Comparative Evaluation of State‑of‑the‑Art Large Language Models for Patient Education Prior to Interventional Radiology procedures

Comparative Evaluation of State‑of‑the‑Art Large Language Models for Patient Education Prior to Interventional Radiology procedures | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Comparative Evaluation of State‑of‑the‑Art Large Language Models for Patient Education Prior to Interventional Radiology procedures Bogdan Levita, Semil Eminovic, Willie Magnus Lüdemann, Dirk Schnapauff, and 5 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7329930/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 13 Oct, 2025 Read the published version in CVIR Endovascular → Version 1 posted 5 You are reading this latest preprint version Abstract Purpose : This study evaluates four large language models’ (LLMs) ability to answer common patient questions preceding transarterial periarticular embolization (TAPE), computed tomography (CT)-guided high-dose-rate (HDR) brachytherapy, and bleomycin electrosclerotherapy (BEST). The goal is to evaluate their potential to enhance clinical workflows and patient comprehension, while also assessing associated risks. Materials and Methods: 35 TAPE, 34 CT‑HDR brachytherapy, and 36 BEST related questions were presented to ChatGPT-4o, DeepSeek-V3, OpenBioLLM-8b, and BioMistral-7b. The LLM-generated responses were independently assessed by two board-certified radiologists. Accuracy was rated on a 5-point Likert scale. Statistics compared LLM performance across question categories for patient-education suitability. Results: DeepSeek-V3 attained the highest mean scores for BEST [4.49 (± 0.77)] and CT-HDR [4.24 (± 0.81)] and demonstrated comparable performance to ChatGPT-4o for TAPE-related questions (DeepSeek-V3 [4.20 (± 0.77)] vs. ChatGPT-4o [4.17 (± 0.64)]; p = 1.000). In contrast, OpenBioLLM-8b (BEST 3.51 (± 1.15), CT-HDR 3.32 (± 1.13), TAPE 3.34 (± 1.16)) and BioMistral-7b (BEST 2.92 (± 1.35), CT-HDR 3.03 (± 1.06), TAPE 3.33 (± 1.28)) performed significantly worse than DeepSeek-V3 and ChatGPT-4o across all procedures. Preparation/Planning was the only category without statistically significant differences across all three procedures. Conclusion: DeepSeek‑V3 and ChatGPT‑4o excelled on TAPE, BEST and CT‑HDR brachytherapy questions, indicating potential to enhance patient education in interventional radiology, where complex but minimally invasive procedures often are explained in brief consultations. However, OpenBioLLM‑8b and BioMistral‑7b exhibited more frequent inaccuracies, suggesting that LLMs cannot replace comprehensive clinical consultations yet. Patient feedback and clinical workflow implementation should validate these findings. Large language models interventional radiology patient education Figures Figure 1 Figure 2 Introduction Large Language Models (LLMs) are progressively being adopted in medicine to aid in clinical decisionsupport, medical documentation, diagnostics, and patient education, demonstrating both significant promise and emerging risks in healthcare workflows [ 1 – 3 ]. An early benchmark study found that ChatGPT achieved near-passing performance on a radiology board-style examination, indicating the depth of domain-specific knowledge LLMs can provide while highlighting the necessity for validation in specialized settings [4]. They can also offer accurate and detailed responses to patient inquiries in Interventional Radiology (IR), providing new opportunities for patient education by reducing fear and improving compliance, though also posing some risks of misinformation [ 5 – 7 ]. IR has significantly expanded its applications over the last years, largely due to advancements in technology, offering advantages such as reduced risk and shorter recovery times compared to traditional open surgery [ 8 ]. However, IR lacks public awareness and understanding by patients: a study found that 65% of patients were unfamiliar with IR and 72% did not recognize IR physicians as such [ 9 ]. As a result, IR physicians must overcome a more significant “gap” in educating patients to help them make well-informed decisions. This gap is particularly evident for less commonly performed procedures, such as Bleomycin electrosclerotherapy (BEST), transarterial periarticular embolization (TAPE) and CT-guided high-dose-rate (HDR) brachytherapy, where patient knowledge is often especially limited. BEST combines electroporation with bleomycin in the treatment of vascular malformations [ 10 , 11 ], improving local drug uptake while reducing systemic effects [ 12 ], and exhibits promising outcomes in multiple studies [ 13 – 15 ]. TAPE is an increasingly popular therapeutic procedure for chronic joint pain, often when drug and physical therapy options have been exhausted, and can be performed comparatively quickly and inexpensively, contributing to significant pain relief [ 16 – 18 ]. CT-HDR brachytherapy is a minimally invasive procedure for liver tumors, demonstrating favorable outcomes with extended survival and excellent local tumor control in numerous studies [ 19 – 21 ]. However, the three procedures are limited to specialized centers and require comprehensive patient education and informed consent [ 22 ], which can be time- and resource-intensive in clinical practice [ 23 , 24 ]. Despite a growing focus on patient empowerment [ 25 ], many patients lack sufficient understanding of medical procedures [ 24 ] and their risks [ 26 ]. The safety and efficacy of LLMs in educating patients about less common procedures like BEST, TAPE, or CT-HDR brachytherapy remain uncertain, as does their potential to reduce clinicians' workload, improve patient satisfaction, or support clinical workflows. Given their rapid evolution, this study evaluates the accuracy of four LLMs in answering common patient questions about CT-HDR brachytherapy, TAPE, and BEST, aiming to assess their clinical relevance and ability to enhance patient communication in IR. Materials and Methods Study Design Due to the human-generated sample dataset, ethics committee approval was not required. In this prospective evaluation study, two radiology residents (fifth and second year, respectively) created 107 questions based on commonly asked patient questions encountered in daily clinical practice before TAPE (35), CT-HDR brachytherapy (34) and BEST (36). To ensure clinical relevance and realism, the questions were subsequently reviewed and validated by two board-certified interventional radiologists (18 and 11 years of experience). Responses were generated by four state-of-the-art LLMs, three open-source models (DeepSeek-V3, OpenBioLLM-8b, BioMistral-7b) and one proprietary model (ChatGPT-4o). The two interventional radiologists independently evaluated responses using a 5-point Likert scale (Supplementary Table S1 ). ChatGPT-4o and DeepSeek-V3 were selected for their public visibility and scientific validation [ 27 , 28 ]. OpenBioLLM-8b and BioMistral-7b enabled comparison with smaller, medically pre-trained language models. Question design and Prompting Question categories were general information, preparation/planning, risks/contraindications, recovery/follow-up and side effects/complications. Two examples of asked questions: “What medication do I have to stop taking before a TAPE and for how long? or “What happens if I cannot withstand the radiation during CT-HDR brachytherapy and have to interrupt the treatment?” All four LLMs were prompted identically with ” I am a patient. I am due to have a CT-HDR brachytherapy and have some questions about this procedure. Can you answer each of the following questions in an understandable way. ”, and respectively the same for TAPE and BEST. All questions for each procedure were subsequently submitted via API to all four LLMs (on December 10th 2024) using identical prompts to ensure comparability and consistency. All questions are provided as supplementary material, all responses and ratings are documented and available upon request. Response Evaluation Response accuracy from DeepSeek-V3, OpenBioLLM-8b, BioMistral-7b, and ChatGPT-4o was rated by two independent board-certified radiologists using a 5-point Likert scale (1 = Very inaccurate / completely false, very likely to mislead; 2 = Inaccurate / mostly false, likely to mislead; 3 = Neutral / moderately accurate, overall acceptable; 4 = Accurate / mostly correct, only very few inaccuracies, unlikely to mislead; 5 = Very accurate / completely correct, very unlikely to mislead; displayed in Supplementary Table S1 ) and subsequently statistically compared. LLMs were anonymized to ensure blinded evaluation by the radiologists. Statistical Analysis Categorical and ordinal ratings are presented as counts and percentages (n, %), and overall scores are reported as mean ± SD, calculated from the radiologists’ individual means. To assess statistically significant differences between models, a two-tailed, four sample Friedman-test was applied (repeated measures, ordinal scale). P-values below 0.05 were considered statistically significant. If so, post-hoc Wilcoxon signed-rank tests with Holm correction were used to assess pairwise statistically significant differences between LLM ratings. Holm correction adjusted the significance level for multiple comparisons. Interrater agreement was assessed using the two-way random effects single-measure intraclass correlation coefficient (ICC) and Cohen’s Kappa to evaluate absolute agreement between both raters per model and across all items (Supplementary Table S2 ). Analyses and visualizations were performed using Python (version 3.9.13) with packages including Pandas (version 2.2.2), NumPy (version 1.23.1), SciPy (version 1.13.1; Scikit-Posthocs Version 0.10.0), Matplotlib (version 3.9.2) and Seaborn (version 0.13.2). Results Performance grading of the four LLMs is displayed in Table 1 (BEST), Table 2 (CT-HDR brachytherapy) and Table 3 (TAPE). Statistics for all questions and all covered categories are illustrated in Table 4 . Cumulative ratings from both radiologists are displayed in Fig. 1 (a)-(c). Table 1 Grading of the performance of four Large Language Models (OpenBioLLM-8b, BioMistral-7b, ChatGPT-4o, DeepSeek-V3) in answering patient questions before a BEST procedure (evaluations of both radiologists combined). With two radiologists grading 36 BEST-related questions there are 72 ratings in total. Rating OpenBioLLM-8b BioMistral-7b ChatGPT-4o DeepSeek-V3 5 21 (29.17%) 14 (19.44%) 39 (54.17%) 48 (66.67%) 4 21 (29.17%) 14 (19.44%) 24 (33.33%) 17 (23.61%) 3 13 (18.06%) 14 (19.44%) 6 (8.33%) 3 (4.17%) 2 8 (11.11%) 12 (16.67%) 2 (2.78%) 2 (2.78%) 1 9 (12.50%) 18 (25.00%) 1 (1.39%) 2 (2.78%) Mean (± SD) 3.51 (± 1.15) 2.92 (± 1.35) 4.36 (± 0.71) 4.49 (± 0.77) Table 2 Grading of the performance of four Large Language Models (OpenBioLLM-8b, BioMistral-7b, ChatGPT-4o, DeepSeek-V3) in answering patient questions before a CT-HDR procedure (evaluations of both radiologists combined). With two radiologists grading 34 CT-HDR-related questions there are 68 ratings in total. Rating OpenBioLLM-8b BioMistral-7b ChatGPT-4o DeepSeek-V3 5 10 (14.71%) 7 (10.29%) 19 (27.94%) 35 (51.47%) 4 28 (41.18%) 17 (25.00%) 26 (38.24%) 21 (30.88%) 3 13 (19.12%) 25 (36.76%) 16 (23.53%) 7 (10.29%) 2 8 (11.76%) 9 (13.24%) 5 (7.35%) 3 (4.41%) 1 9 (13.24%) 10 (14.71%) 2 (2.94%) 2 (2.94%) Mean (± SD) 3.32 (± 1.25) 3.03 (± 1.18) 3.81 (± 1.03) 4.24 (± 1.01) Table 3 Grading of the performance of four Large Language Models (OpenBioLLM-8b, BioMistral-7b, ChatGPT-4o, DeepSeek-V3) in answering patient questions before a TAPE procedure (evaluations of both radiologists combined). With two radiologists grading 35 TAPE-related questions there are 70 ratings in total. Rating OpenBioLLM-8b BioMistral-7b ChatGPT-4o DeepSeek-V3 5 15 (21.43%) 19 (27.14%) 28 (40.00%) 32 (45.71%) 4 19 (27.14%) 15 (21.43%) 30 (42.86%) 23 (32.86%) 3 17 (24.29%) 16 (22.86%) 9 (12.86%) 12 (17.14%) 2 13 (18.57%) 10 (14.29%) 2 (2.86%) 3 (4.29%) 1 6 (8.57%) 10 (14.29%) 1 (1.43%) 0 (0.00%) Mean (± SD) 3.34 (± 1.16) 3.33 (± 1.28) 4.17 (± 0.64) 4.20 (± 0.77) Table 4 Comparative analysis of the performance of four Large Language Models (OpenBioLLM-8b, BioMistral-7b, ChatGPT-4o, DeepSeek-V3) in answering patient questions before an interventional radiology procedure (BEST, CT-HDR, TAPE) OpenBioLLM-8b BioMistral-7b ChatGPT-4o DeepSeek-V3 P-value Friedman-Test BEST All Questions (Mean, SD) 3.51 (± 1.15) 2.92 (± 1.35) 4.36 (± 0.71) 4.49 (± 0.77) < 0.001** General Information (Mean, SD) 2.88 (± 1,30) 2.50 (± 1,46) 4.71 (± 0,50) 4.12 (± 1,11) < 0.001** Preparation / Planning (Mean, SD) 4.10 (± 0,42) 3.30 (± 1,44) 4.50 (± 0,50) 4.60 (± 0,55) 0.186 Risks / Contraindications (Mean, SD) 3.86 (± 1,38) 3.14 (± 1,25) 3.64 (± 1,03) 4.86 (± 0,24) 0.016* Recovery / Follow-Up (Mean, SD) 4.00 (± 0,58) 2.93 (± 1,40) 4.43 (± 0,53) 4.50 (± 0,58) 0.014* Side Effects / Complications (Mean, SD) 3.30 (± 1,04) 3.20 (± 1,35) 4.30 (± 0,45) 4.70 (± 0,45) 0.023* CT-HDR All Questions (Mean, SD) 3.32 (± 1.13) 3.03 (± 1.06) 3.81 (± 0.76) 4.24 (± 0.81) < 0.001** General Information (Mean, SD) 3.05 (± 1,28) 2.95 (± 0,86) 3.50 (± 0,62) 4.05 (± 0,86) 0.025* Preparation / Planning (Mean, SD) 3.56 (± 1,13) 3.39 (± 1,11) 3.67 (± 0,61) 4.33 (± 0,97) 0.178 Risks / Contraindications (Mean, SD) 3.10 (± 1,39) 2.40 (± 1,47) 4.40 (± 0,82) 4.40 (± 0,22) 0.007** Recovery / Follow-Up (Mean, SD) 3.40 (± 1,14) 3.10 (± 0,96) 3.70 (± 1,04) 4.10 (± 0,96) 0.176 Side Effects / Complications (Mean, SD) 3.60 (± 0,74) 3.10 (± 1,08) 4.20 (± 0,67) 4.40 (± 0,82) 0.010* TAPE All Questions (Mean, SD) 3.34 (± 1.16) 3.33 (± 1.28) 4.17 (± 0.64) 4.20 (± 0.77) < 0.001** General Information (Mean, SD) 3.00 (± 1,06) 3.83 (± 0,94) 4.44 (± 0,53) 3.94 (± 0,81) 0.037* Preparation / Planning (Mean, SD) 3.30 (± 1,23) 3.05 (± 1,30) 3.85 (± 0,88) 4.25 (± 0,59) 0.057 Risks / Contraindications (Mean, SD) 3.90 (± 1,08) 2.60 (± 1,71) 4.30 (± 0,57) 3.90 (± 0,96) 0.392 Recovery / Follow-Up (Mean, SD) 3.83 (± 0,75) 3.50 (± 1,05) 4.25 (± 0,42) 4.67 (± 0,61) 0.013* Side Effects / Complications (Mean, SD) 2.90 (± 1,60) 3.50 (± 1,62) 4.10 (± 0,42) 4.30 (± 0,97) 0.182 *: significant (p < 0.050); **: p 0.050 Legend: SD, standard deviation. Table 5 Display of p values of Wilcoxon Signed-Rank test with Holm correction for all questions per intervention P values Wilcoxon Signed-Rank test with Holm correction OpenBioLLM-8b BioMistral-7b ChatGPT-4o DeepSeek-V3 BEST - All Questions OpenBioLLM-8b - 0.073 (n.s.) 0.003 ** < 0.001 ** BioMistral-7b - - < 0.001 ** < 0.001 ** ChatGPT-4o - - - 0.332 (n.s.) DeepSeek-V3 - - - - CT-HDR - All Questions OpenBioLLM-8b - 0.133 (n.s.) 0.017 * 0.001 ** BioMistral-7b - - 0.003 ** < 0.001 ** ChatGPT-4o - - - < 0.010 ** DeepSeek-V3 - - - - TAPE - All Questions OpenBioLLM-8b - 1.000 (n.s.) 0.001 ** 0.002 ** BioMistral-7b - - 0.002 ** 0.002 ** ChatGPT-4o - - - 1.000 (n.s.) DeepSeek-V3 - - - - *: significant (p < 0.050); **: p 0.050 For BEST, DeepSeek-V3 achieved the highest average score [4.49 (± 0.77)], followed by ChatGPT-4o [4.36 (± 0.71)]. OpenBioLLM-8b [3.51 (± 1.15)] and BioMistral-7b [2.92 (± 1.35)] scored significantly lower (p < 0.001). Significant differences were observed in General Information (p < 0.001), Risks / Contraindications (p = 0.016), Recovery / Follow-Up (p = 0.014), and Side Effects / Complications (p = 0.023), but not in Preparation / Planning (p = 0.186). For CT-HDR brachytherapy, DeepSeek-V3 again outperformed other models [4.24 (± 0.81)], followed by ChatGPT-4o [3.81 (± 0.76)], OpenBioLLM-8b [3.32 (± 1.13)], and BioMistral-7b [3.03 (± 1.06)] (p < 0.001). Statistically significant differences were specifically noted in General Information (p = 0.025), Risks / Contraindications (p = 0.007), and Side Effects / Complications (p = 0.010), while Preparation / Planning (p = 0.178) and Recovery / Follow-Up (p = 0.176) showed no significant differences. For TAPE, ChatGPT-4o [4.17 (± 0.64)] and DeepSeek-V3 [4.20 (± 0.77)] performed similarly, significantly surpassing OpenBioLLM-8b [3.34 (± 1.16)] and BioMistral-7b [3.33 (± 1.28)] (p < 0.001). Significant differences were observed in General Information (p = 0.037) and Recovery / Follow-Up (p = 0.013). Categories Preparation / Planning (p = 0.057), Risks / Contraindications (p = 0.392), and Side Effects / Complications (p = 0.182) showed no statistically significant differences. Radar plots in Fig. 2 illustrate average category scores for each procedure. Interrater agreement varied across models and procedures, with highest consistency for BioMistral-7b (ICC up to 0.69), and lowest for ChatGPT-4o, particularly for TAPE-related questions with ICC as low as 0.13 (ICC and Cohen’s Kappa in Supplementary Table S2 ; ratings per radiologist for each procedure are displayed in Supplementary Tables S3-S5). Discussion This study evaluated the performance of four advanced LLMs in addressing potentially relevant and frequently asked patient questions related to TAPE, BEST, and CT-HDR brachytherapy procedures. According to two interventional radiologists, DeepSeek-V3 and ChatGPT-4o demonstrated strong performance by delivering accurate and understandable responses, while the medically pre-trained OpenBioLLM-8b and BioMistral-7b performed significantly worse across all procedures and produced potentially hazardous answers. For instance, OpenBioLLM-8b failed to identify manifest hyperthyroidism as a contraindication, which is concerning given that all three procedures involve iodinated contrast media, which may cause a potentially life threatening thyrotoxic crisis in patients with hyperthyroidism [ 29 ]. BioMistral-7b provided inaccurate information regarding radiation exposure, incorrectly asserting that the radiation dose would be similar or lower than that of a chest radiograph in all three procedures. While OpenBioLLM-8b and BioMistral-7b were developed using biomedical training data, it is likely that the limited size and scope of their training datasets, relative to the larger models ChatGPT-4o and DeepSeek-V3, contributed to the inaccurate and potentially harmful responses observed. Responses generated by ChatGPT-4o and DeepSeek-V3 exhibited superior structural organization, enabling step-by-step comprehension of complex medical information. However, the two models showed only low to moderate interrater agreement, underscoring the subjective way LLM responses are interpreted. In contrast, stronger agreement on the lower-performing models could reflect the ease of identifying clearly incorrect or misleading content. In this study, less commonly performed IR procedures were intentionally selected to evaluate LLM performance under challenging and less-represented clinical settings. Given the evolving nature of IR, assessing how LLMs respond to newer, or niche procedures is of significant relevance. Although this focus limits generalizability, it highlights the potential and robustness of LLMs in underrepresented areas. Several studies have shown that LLMs can provide accurate responses to patient questions in IR [ 30 , 31 ]. ChatGPT-4 showed strong performance in explaining IR complications but requires health literacy to be fully understood [ 32 ]. Combining LLMs with educational videos may enhance patient understanding of IR procedures and satisfaction during informed consent [ 33 ]. However, not all studies reported positive conclusions for LLMs in IR patient education, highlighting incomplete or inaccurate patient education content [ 31 ]. Hofmann and Vairavamurthy found responses incomplete or lacking depth, especially for risks and alternatives, with more experienced reviewers rating LLMs less favorably [ 5 ]. This issue is not exclusive to ChatGPT, as previous studies identified complications in IR cases that had not been documented during the consent process [ 34 ]. Several studies report issues with LLMs when answering radiology-related questions, including errors, jargon, and hallucinations. Jeblick et al. [ 35 ] found that approximately one-third of simplified radiology reports contained errors potentially harmful to patients. Although patients rated ChatGPT-4.0 superior to medical experts for empathy and usefulness, prior studies indicate that patients frequently struggle to recognize harmful or inaccurate advice [ 36 ]. This emphasizes the importance of transparent communication and expert oversight, as specialized medical knowledge remains essential to identify and reduce harmful or misleading information produced by LLMs [ 37 ]. The strong performance of DeepSeek-V3 observed in this study aligns with the literature [ 27 , 38 , 39 ], but concerns about user data privacy and regulatory challenges persist [ 40 ]. Broader ethical concerns include legal accountability and the protection of human rights [ 41 , 42 ]. Determining responsibility in cases of patient harm is particularly complex [ 43 ], and current data collection methods pose additional threats to patient confidentiality and privacy [ 44 ]. Beyond privacy concerns, LLMs could undermine the doctor-patient relationship and reduce trust in medical professionals. However, when appropriately applied, they may improve informed decision-making by providing clear information in complex clinical situations [ 45 ]. Despite these benefits, patient safety must remain the priority in future LLM development, reinforcing the urgent need for the establishment of ethical frameworks to guide responsible LLM use [ 44 ]. This study compared open-source models (DeepSeek-V3, OpenBioLLM-8b, BioMistral-7b) with the proprietary ChatGPT-4o. Although ChatGPT-4o performed well, concerns remain about transparency and data security risks [ 46 ]. Open-source models offer distinct advantages, including enhanced transparency, greater customization opportunities, and increased control over patient data. Given that DeepSeek-V3 is freely accessible and exhibited high performance, it may represent a more practical tool for broader patient access. Obtaining informed consent in IR is often complex and time-consuming. LLMs could streamline this process, enhance patient empowerment, and improve communication efficiency. However, the complexity of certain IR procedures, including risks associated with off-label device use, demands medical expertise [ 47 ]. While DeepSeek-V3 and ChatGPT-4.0 performed well in this study, instances of misinformation produced by the two medically pre-trained LLMs, emphasize the continuing need for physician oversight to ensure patient safety [ 48 ]. Time savings could be achieved by integrating LLMs into structured clinical pathways with outputs limited to low-risk content, while physician oversight is reserved for high-risk areas or responses flagged as low-confidence by the model. This approach could maintain patient safety without requiring full review of all outputs. Future studies should focus on model safety, accuracy and adherence to clinical guidelines. Limitations The questions were developed by two radiology residents based on clinical experience within a single institution, limiting the representativeness. Although validated by two board-certified interventional radiologists, rewording patient questions may also have altered their original nuance. The assessment covered only three less commonly performed IR procedures, reducing generalizability. LLMs were selected based on public awareness and prior validation, introducing possible selection bias. No patients were involved in the design or evaluation of the questions, and standardized prompting does not fully reflect real doctor-patient interactions. Responses were evaluated by only two board-certified radiologists, introducing potential subjective bias. Although the study primarily assessed accuracy and misinformation in LLM-generated responses, aspects like linguistic accessibility and empathetic communication may also matter in patient education. Despite the limitations, the findings provide a valuable foundation for future research, ideally incorporating direct patient participation. Conclusions DeepSeek-V3 and ChatGPT-4o demonstrated a strong performance in answering questions related to TAPE, BEST and CT-HDR brachytherapy, highlighting their potential for patient education and communication improvement. This is especially relevant in IR, where complex but minimally invasive procedures are often explained within tight consultation windows. OpenBioLLM-8b and BioMistral-7b produced more frequent inaccuracies, underscoring the risks of integrating smaller, biomedical-specific models into clinical practice. These findings demonstrate that LLMs cannot substitute comprehensive medical consultations yet. Nevertheless, LLMs will play an increasing role in radiology and patient care. Future research should validate these findings, incorporate patient feedback and evaluate LLM integration into clinical workflows. Abbreviations LLM Large language model CT-HDR Computed tomography-guided high-dose-rate BEST Bleomycin electrosclerotherapy TAPE Transarterial periarticular embolization IR Interventional radiology ICC Intraclass correlation coefficient Declarations Ethics approval and consent to participate: Not applicable. Consent for publication: Not applicable. Availability of data and material: The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request Competing interests: The authors declare that they have no competing interests. Outside the submitted work, T.P. is funded in part by the Berlin Institute of Health (BIH). T.P. also receives funding from Berlin Institute of Health (Advanced Clinician Scientist Grant, Platform Grant), Ministry of Education and Research (BMBF, 01KX2021 (RACOON), 01KX2121 („NUM 2.0“, RACOON), 68GX21001A, 01ZZ2315D), German Research Foundation (DFG, SFB 1340/2), European Union (H2020, CHAIMELEON: 952172, DIGITAL, EUCAIM:101100633). T.P. also declares relationships with the following companies: research agreements (no personal payments) with AGO, Aprea AB, ARCAGY-GINECO, Astellas Pharma Global Inc. (APGD), Astra Zeneca, Clovis Oncology, Inc., Holaira, Incyte Corporation, Karyopharm, Lion Biotechnologies, Inc., MedImmune, Merck Sharp & Dohme Corp, Millennium Pharmaceuticals, Inc., Morphotec Inc., NovoCure Ltd., PharmaMar S.A. and PharmaMar USA, Inc., Roche, Siemens Healthineers, and TESARO Inc., fees for a book translation (Elsevier B.V.), fees for speaking engagements (Bayer Healthcare). J.N. receives funding from Berlin Institute of Health (Digital Health Accelerator), European Union’s Horizon Europe programme (COMFORT, 101079894) and reports personal fees from Eppdata GmbH outside the submitted work. Funding: No funding was received for this study. Author’s contributions: Conceptualization, B.L. and S.E.; methodology, B.L. and S.E.; software, S.E.; validation, B.L., S.E., W.M.L., D.S., J.N., A.D., and T.P.; formal analysis, S.E.; investigation, B.L.; resources, T.P.; data curation, B.L. and S.E.; writing - original draft preparation, B.L. and S.E.; writing -review and editing, B.L., S.E., W.M.L., D.S., J.N., A.D. and T.P.; visualization, S.E.; supervision, T.P.; project administration, T.P.. Acknowledgements: Not applicable. References Wang L, Wan Z, Ni C, Song Q, Li Y, Clayton E et al (2024) Applications and Concerns of ChatGPT and Other Conversational Large Language Models in Health Care: Systematic Review. J Med Internet Res 26(1):e22769. https://doi.org/10.2196/22769 Busch F, Hoffmann L, Rueger C, van Dijk EH, Kader R, Ortiz-Prado E et al (2025) Current applications and challenges in large language models for patient care: a systematic review. Commun Med 5(1):1–13. https://doi.org/10.1038/s43856-024-00717-2 Thirunavukarasu AJ, Ting DSJ, Elangovan K, Gutierrez L, Tan TF, Ting DSW (2023) Large language models in medicine. Nat Med 29(8):1930–1940. https://doi.org/10.1038/s41591-023-02448-8 Bhayana R, Krishna S, Bleakney RR (2023) Performance of ChatGPT on a Radiology Board-style Examination: Insights into Current Strengths and Limitations. Radiology 307(5):e230582. https://doi.org/10.1148/radiol.230582 Hofmann HL, Vairavamurthy J (2024) Large language model doctor: assessing the ability of ChatGPT-4 to deliver interventional radiology procedural information to patients during the consent process. CVIR Endovasc 7:83. https://doi.org/10.1186/s42155-024-00477-z Elhakim T, Brea AR, Fidelis W, Paravastu SS, Malavia M, Omer M et al (2025) Enhanced PROcedural Information READability for Patient-Centered Care in Interventional Radiology With Large Language Models (PRO-READ IR). J Am Coll Radiol JACR 22(1):84–97. https://doi.org/10.1016/j.jacr.2024.08.010 Kooraki S, Hosseiny M, Jalili MH, Rahsepar AA, Imanzadeh A, Kim GH et al (2024) Evaluation of ChatGPT-Generated Educational Patient Pamphlets for Common Interventional Radiology Procedures. Acad Radiol 31(11):4548–4553. https://doi.org/10.1016/j.acra.2024.05.024 Mashoufi R, Mashoufi R (2023) Interventional Radiology for Disease Management: A Narrative Review. Cureus 15(11):e48603. https://doi.org/10.7759/cureus.48603 Heister D, Jackson S, Doherty-Simor M, Newton I (2018) An Evaluation of Trends in Patient and Public Awareness of IR. J Vasc Interv Radiol 29(5):661–668. https://doi.org/10.1016/j.jvir.2017.11.023 Horbach SER, Rigter IM, Smitt JHS, Reekers JA, Spuls PI, van der Horst CMAM (2016) Intralesional Bleomycin Injections for Vascular Malformations: A Systematic Review and Meta-Analysis. Plast Reconstr Surg 137(1):244–256. https://doi.org/10.1097/PRS.0000000000001924 Horbach SER, Wolkerstorfer A, Jolink F, Bloemen PR, van der Horst CMAM (2020) Electrosclerotherapy as a Novel Treatment Option for Hypertrophic Capillary Malformations: A Randomized Controlled Pilot Trial. Dermatol Surg Off Publ Am Soc Dermatol Surg Al 46(4):491–498. https://doi.org/10.1097/DSS.0000000000002191 Giardino R, Fini M, Bonazzi V, Cadossi R, Nicolini A, Carpi A (2006) Electrochemotherapy a novel approach to the treatment of metastatic nodules on the skin and subcutaneous tissues. Biomed Pharmacother Biomedecine Pharmacother 60(8):458–462. https://doi.org/10.1016/j.biopha.2006.07.016 Schmidt VF, Cangir Ö, Meyer L, Goldann C, Hengst S, Brill R et al (2024) Outcome of bleomycin electrosclerotherapy of slow-flow malformations in adults and children. Eur Radiol 34(10):6425–6434. https://doi.org/10.1007/s00330-024-10723-6 Dalmády S, Csoma Z, Besenyi Z, Bottyán K, Oláh J, Kemény L et al (2020) New Treatment Option for Capillary Lymphangioma: Bleomycin-Based Electrochemotherapy of an Infant. Pediatrics 146(6):e20200566. https://doi.org/10.1542/peds.2020-0566 Guntau M, Cucuruz B, Brill R, Bidakov O, Von der Heydt S, Deistung A et al (2023) Individualized treatment of congenital vascular malformations of the tongue. Clin Hemorheol Microcirc 83(4):421–429. https://doi.org/10.3233/CH-221683 Lee SH, Hwang JH, Kim DH, So YH, Park J, Cho SB et al (2019) Clinical Outcomes of Transcatheter Arterial Embolisation for Chronic Knee Pain: Mild-to-Moderate Versus Severe Knee Osteoarthritis. Cardiovasc Intervent Radiol 42(11):1530–1536. https://doi.org/10.1007/s00270-019-02289-4 Little MW, Gibson M, Briggs J, Speirs A, Yoong P, Ariyanayagam T et al (2021) Genicular artEry embolizatioN in patiEnts with oSteoarthrItiS of the Knee (GENESIS) Using Permanent Microspheres: Interim Analysis. Cardiovasc Intervent Radiol 44(6):931–940. https://doi.org/10.1007/s00270-020-02764-3 Taslakian B, Miller LE, Mabud TS, Macaulay W, Samuels J, Attur M et al (2023) Genicular artery embolization for treatment of knee osteoarthritis pain: Systematic review and meta-analysis. Osteoarthr Cartil Open 5(2):100342. https://doi.org/10.1016/j.ocarto.2023.100342 Friedrich S, Busch F, Jonczyk M, Wieners G, Böning G, Lüdemann WM et al (2023) Combined CT-guided high-dose-rate brachytherapy (CT-HDRBT) and transarterial chemoembolization with irinotecan-loaded microspheres improve local tumor control and progression-free survival in patients with unresectable colorectal liver metastases compared with mono-CT-HDRBT. J Contemp Brachytherapy 15(1):15–26. https://doi.org/10.5114/jcb.2023.125480 Böning G, Büttner L, Jonczyk M, Lüdemann WM, Denecke T, Schnapauff D et al (2020) Complications of Computed Tomography-Guided High-Dose-Rate Brachytherapy (CT-HDRBT) and Risk Factors: Results from More than 10 Years of Experience. Cardiovasc Intervent Radiol 43(2):284–294. https://doi.org/10.1007/s00270-019-02386-4 Brinkhaus G, Lock JF, Malinowski M, Denecke T, Neuhaus P, Hamm B et al (2014) CT-guided high-dose-rate brachytherapy of liver tumours does not impair hepatic function and shows high overall safety and favourable survival rates. Ann Surg Oncol 21(13):4284–4292. https://doi.org/10.1245/s10434-014-3835-y O’Dwyer HM, Lyon SM, Fotheringham T, Lee MJ (2003) Informed consent for interventional radiology procedures: a survey detailing current European practice. Cardiovasc Intervent Radiol 26(5):428–433. https://doi.org/10.1007/s00270-003-0058-1 van Leeuwen DJ, van Delden OM (2015) Ethics and informed consent in patients with hepatocellular carcinoma: Changing roles for hepatologist and radiologist. Clin Liver Dis 6(5):122–125. https://doi.org/10.1002/cld.513 Prashar A, Butt S, Castiglione DG, Shaida N (2021) Informed consent in interventional radiology - are we doing enough? Br J Radiol 94(1122):20201368. https://doi.org/10.1259/bjr.20201368 Pinto A, Giurazza F, Califano T, Rea G, Valente T, Niola R et al (2021) Interventional radiology in gynecology and obstetric practice: Safety issues. Semin Ultrasound CT MR 42(1):104–112. https://doi.org/10.1053/j.sult.2020.09.004 Sidhu R, Sakellariou V, Layte P, Soliman A (2006) Patient feedback on helpfulness of postal information packs regarding informed consent for endoscopic procedures. Gastrointest Endosc 64(2):229–234. https://doi.org/10.1016/j.gie.2006.02.055 Sandmann S, Hegselmann S, Fujarski M, Bickmann L, Wild B, Eils R et al (2025) Benchmark evaluation of DeepSeek large language models in clinical decision-making. Nat Med 1–4. https://doi.org/10.1038/s41591-025-03727-2 Newton PM, Summers CJ, Zaheer U, Xiromeriti M, Stokes JR, Bhangu JS et al (2025) Can ChatGPT-4o Really Pass Medical Science Exams? A Pragmatic Analysis Using Novel Questions. Med Sci Educ 35(2):721–729. https://doi.org/10.1007/s40670-025-02293-z Rhee CM, Bhan I, Alexander EK, Brunelli SM (2012) Association Between Iodinated Contrast Media Exposure and Incident Hyperthyroidism and Hypothyroidism. Arch Intern Med 172(2):153–159. https://doi.org/10.1001/archinternmed.2011.677 Scheschenja M, Viniol S, Bastian MB, Wessendorf J, König AM, Mahnken AH (2024) Feasibility of GPT-3 and GPT-4 for in-Depth Patient Education Prior to Interventional Radiological Procedures: A Comparative Analysis. Cardiovasc Intervent Radiol 47(2):245–250. https://doi.org/10.1007/s00270-023-03563-2 McCarthy CJ, Berkowitz S, Ramalingam V, Ahmed M (2023) Evaluation of an Artificial Intelligence Chatbot for Delivery of IR Patient Education Material: A Comparison with Societal Website Content. J Vasc Interv Radiol JVIR 34(10):1760–1768e32. https://doi.org/10.1016/j.jvir.2023.05.037 Kaba E, Beyazal M, Çeliker FB, Yel İ, Vogl TJ (2025) Accuracy and Readability of ChatGPT on Potential Complications of Interventional Radiology Procedures: AI-Powered Patient Interviewing. Acad Radiol 32(3):1547–1553. https://doi.org/10.1016/j.acra.2024.10.028 Makary MS, Jacob CC, Boggs Z, Brankovic R, Paradiso M, Regalado L (2024) Impact of Educational Videos on Patient Understanding of Interventional Radiology Procedures. Acad Radiol 31(11):4554–4559. https://doi.org/10.1016/j.acra.2024.08.020 Forsman T, Silberstein S, Keller EJ (2023) Consent in Interventional Radiology-How Can We Make It Better? Can Assoc Radiol. J J Assoc Can Radiol 74(1):202–210. https://doi.org/10.1177/08465371221101625 Jeblick K, Schachtner B, Dexl J, Mittermeier A, Stüber AT, Topalis J et al (2024) ChatGPT makes medicine easy to swallow: an exploratory case study on simplified radiology reports. Eur Radiol 34(5):2817–2825. https://doi.org/10.1007/s00330-023-10213-1 Armbruster J, Bussmann F, Rothhaas C, Titze N, Grützner PA, Freischmidt H (2024) Doctor ChatGPT, Can You Help Me? The Patient’s Perspective: Cross-Sectional Study. J Med Internet Res 26(1):e58831. https://doi.org/10.2196/58831 Masanneck L, Schmidt L, Seifert A, Kölsche T, Huntemann N, Jansen R et al (2024) Triage Performance Across Large Language Models, ChatGPT, and Untrained Doctors in Emergency Medicine: Comparative Study. J Med Internet Res 26(1):e53297. https://doi.org/10.2196/53297 Liang W, Chen P, Zou X, Lu X, Liu S, Yang J et al (2025) DeepSeek: the Watson to doctors—from assistance to collaboration. J Thorac Dis 17(2):1103–1105. https://doi.org/10.21037/jtd-2025b-03 Jiao C, Rosas E, Asadigandomani H, Delsoz M, Madadi Y, Raja H et al (2025) Diagnostic Performance of Publicly Available Large Language Models in Corneal Diseases: A Comparison with Human Specialists. Diagn Basel Switz 15(10):1221. https://doi.org/10.3390/diagnostics15101221 Temsah A, Alhasan K, Altamimi I, Jamal A, Al-Eyadhy A, Malki KH et al DeepSeek in Healthcare: Revealing Opportunities and Steering Challenges of a New Open-Source Artificial Intelligence Frontier. Cureus 17(2):e79221. https://doi.org/10.7759/cureus.79221 Jeyaraman M, Balaji S, Jeyaraman N, Yadav S (2023) Unraveling the Ethical Enigma: Artificial Intelligence in Healthcare. Cureus 15(8):e43262. https://doi.org/10.7759/cureus.43262 Sallam M (2023) ChatGPT Utility in Healthcare Education, Research, and Practice: Systematic Review on the Promising Perspectives and Valid Concerns. Healthc Basel Switz 11(6):887. https://doi.org/10.3390/healthcare11060887 Wang Y, Peng Y, Wang T, Li H, Zhao Z, Gong L et al (2023) The evolution and current situation in the application of dual-energy computed tomography: a bibliometric study. Quant Imaging Med Surg 13(10):6801–6813. https://doi.org/10.21037/qims-23-467 Ong JCL, Chang SYH, William W, Butte AJ, Shah NH, Chew LST et al (2024) Ethical and regulatory challenges of large language models in medicine. Lancet Digit Health 6(6):e428–e432. https://doi.org/10.1016/S2589-7500(24)00061-X Allen JW, Earp BD, Koplin J, Wilkinson D (2024) Consent-GPT: is it ethical to delegate procedural consent to conversational AI? J Med Ethics 50(2):77–83. https://doi.org/10.1136/jme-2023-109347 Amin K, Khosla P, Doshi R, Chheang S, Forman HP (2023) Artificial Intelligence to Improve Patient Understanding of Radiology Reports. Yale J Biol Med 96(3):407–417. https://doi.org/10.59249/NKOY5498 Zvavanjanja RC, Odetoyinbo TO, Rowlands PC, Healey A, Abdelsalam H, Powell S et al (2012) Off label use of devices and drugs in interventional radiology. Clin Radiol 67(3):239–243. https://doi.org/10.1016/j.crad.2011.06.017 Boyd K (2015) The impossibility of informed consent? J Med Ethics 41(1):44–47. https://doi.org/10.1136/medethics-2014-102308 Supplementary Files BLLLMIRPatientEducationSupplementaryMaterial.docx Cite Share Download PDF Status: Published Journal Publication published 13 Oct, 2025 Read the published version in CVIR Endovascular → Version 1 posted Editorial decision: Minor revision 08 Sep, 2025 Reviewers agreed at journal 20 Aug, 2025 Reviewers invited by journal 15 Aug, 2025 Editor assigned by journal 12 Aug, 2025 First submitted to journal 11 Aug, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7329930","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":500881692,"identity":"5c14f5db-ee05-4153-82d3-22e031bb8aa2","order_by":0,"name":"Bogdan Levita","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABFUlEQVRIiWNgGAWjYBACAyCWAGIZBgbGBgaGCoYECIOBmaAWHqDKxgaGM2AtQAZxWoCmM7aBtICtwa3FnL394o2fOxh4+CWS2x/8nGeXxz/7cPuDDwzWcri0WPacKbbsPcPAIzkjsbGxd1tyscQ5IGMGQ7oxTofdyEmT4G1j4DE4c7CxgXfbgcSGM4yNzTwMhxMb8GiR/AvUYg/U0vh3zoHE+SAtfxgO1+PWkn5MGmwLe2NjM2/DgcQNIC0MDIcT8PiF2Vq2TYJH4nhj42yZY8mJG4FaZvYYpBvisgUYYg9vvm2zkeNvZn/w8U2NXeK8M+wPPvyosJbHZQswRmBRg+pg3BoYGNgf4JMdBaNgFIyCUcDAAAB7yV2P/Pn1KwAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0009-0009-0603-1186","institution":"Charité Universitätsmedizin Berlin CVK: Charite Universitatsmedizin Berlin - Campus Virchow-Klinikum","correspondingAuthor":true,"prefix":"","firstName":"Bogdan","middleName":"","lastName":"Levita","suffix":""},{"id":500881693,"identity":"a4eca33e-613c-4f2a-9f1a-c98a425e9140","order_by":1,"name":"Semil Eminovic","email":"","orcid":"","institution":"Charité Universitätsmedizin Berlin CVK: Charite Universitatsmedizin Berlin - Campus Virchow-Klinikum","correspondingAuthor":false,"prefix":"","firstName":"Semil","middleName":"","lastName":"Eminovic","suffix":""},{"id":500881694,"identity":"51c44eca-ba2a-422c-90ca-14942a65e3dc","order_by":2,"name":"Willie Magnus Lüdemann","email":"","orcid":"","institution":"Charité Universitätsmedizin Berlin CVK: Charite Universitatsmedizin Berlin - Campus Virchow-Klinikum","correspondingAuthor":false,"prefix":"","firstName":"Willie","middleName":"Magnus","lastName":"Lüdemann","suffix":""},{"id":500881695,"identity":"69db3857-da2b-492e-93d6-df487eb53748","order_by":3,"name":"Dirk Schnapauff","email":"","orcid":"","institution":"Charité Universitätsmedizin Berlin CVK: Charite Universitatsmedizin Berlin - Campus Virchow-Klinikum","correspondingAuthor":false,"prefix":"","firstName":"Dirk","middleName":"","lastName":"Schnapauff","suffix":""},{"id":500881696,"identity":"fba3bf87-c2bc-4fc1-acac-a997ff30113e","order_by":4,"name":"Robin Schmidt","email":"","orcid":"","institution":"Charité Universitätsmedizin Berlin CVK: Charite Universitatsmedizin Berlin - Campus Virchow-Klinikum","correspondingAuthor":false,"prefix":"","firstName":"Robin","middleName":"","lastName":"Schmidt","suffix":""},{"id":500881697,"identity":"64c7e740-a64b-47aa-b73d-8e19f4e7733d","order_by":5,"name":"Anna-Maria Haack","email":"","orcid":"","institution":"Charité Universitätsmedizin Berlin CVK: Charite Universitatsmedizin Berlin - Campus Virchow-Klinikum","correspondingAuthor":false,"prefix":"","firstName":"Anna-Maria","middleName":"","lastName":"Haack","suffix":""},{"id":500881698,"identity":"0766361a-16b9-4346-b918-d2d1c0f06578","order_by":6,"name":"Andrea Dell’Orco","email":"","orcid":"","institution":"Charité Universitätsmedizin Berlin CVK: Charite Universitatsmedizin Berlin - Campus Virchow-Klinikum","correspondingAuthor":false,"prefix":"","firstName":"Andrea","middleName":"","lastName":"Dell’Orco","suffix":""},{"id":500881699,"identity":"b13d3937-5d65-4aa1-90bb-be55224e3489","order_by":7,"name":"Jawed Nawabi","email":"","orcid":"","institution":"Charité Universitätsmedizin Berlin CVK: Charite Universitatsmedizin Berlin - Campus Virchow-Klinikum","correspondingAuthor":false,"prefix":"","firstName":"Jawed","middleName":"","lastName":"Nawabi","suffix":""},{"id":500881700,"identity":"c7d7d27c-acf1-4a33-bcbc-6ff1f70b5128","order_by":8,"name":"Tobias Penzkofer","email":"","orcid":"","institution":"Charité Universitätsmedizin Berlin CVK: Charite Universitatsmedizin Berlin - Campus Virchow-Klinikum","correspondingAuthor":false,"prefix":"","firstName":"Tobias","middleName":"","lastName":"Penzkofer","suffix":""}],"badges":[],"createdAt":"2025-08-08 19:48:27","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7329930/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7329930/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s42155-025-00609-z","type":"published","date":"2025-10-13T15:58:08+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":89673179,"identity":"5757c841-96f9-488b-82ee-a092d1199e76","added_by":"auto","created_at":"2025-08-22 13:12:43","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":30331,"visible":true,"origin":"","legend":"\u003cp\u003eBar charts that display the cumulative grading of the performance of four LLMs (OpenBioLLM-8b, BioMistral-7b, ChatGPT-4o, DeepSeek-V3) assessed by two radiologists in answering patient questions before the interventional radiology procedures \u003cstrong\u003e(a)\u003c/strong\u003e BEST, \u003cstrong\u003e(b)\u003c/strong\u003eTAPE and \u003cstrong\u003e(c)\u003c/strong\u003e CT HDR.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7329930/v1/461bc951daee5f6bcf03e1ae.png"},{"id":89673180,"identity":"28357e18-e594-471b-bcfb-11b1697af4cf","added_by":"auto","created_at":"2025-08-22 13:12:43","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":118796,"visible":true,"origin":"","legend":"\u003cp\u003eRadar plots that display the illustrate the average performance of four large language models (OpenBioLLM-8b, BioMistral-7b, ChatGPT-4o, DeepSeek-V3) in answering patient questions before interventional radiology procedures \u003cstrong\u003e(a)\u003c/strong\u003e BEST, \u003cstrong\u003e(b)\u003c/strong\u003e CT HDR and \u003cstrong\u003e(c)\u003c/strong\u003eTAPE across five informational categories.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7329930/v1/c701a724e0c807db8e085391.png"},{"id":93956034,"identity":"53e6517f-13dd-4f0e-b36e-c908d6bad3f7","added_by":"auto","created_at":"2025-10-20 16:09:33","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1090211,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7329930/v1/784f16c8-fd41-4acc-8205-06e9ab3740ad.pdf"},{"id":89673711,"identity":"f796b081-d649-4fef-80a4-b6431edb03bc","added_by":"auto","created_at":"2025-08-22 13:20:43","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":33133,"visible":true,"origin":"","legend":"","description":"","filename":"BLLLMIRPatientEducationSupplementaryMaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-7329930/v1/a1cd0e0da8f913b4e1a2bf12.docx"}],"financialInterests":"","formattedTitle":"Comparative Evaluation of State‑of‑the‑Art Large Language Models for Patient Education Prior to Interventional Radiology procedures","fulltext":[{"header":"Introduction","content":"\u003cp\u003eLarge Language Models (LLMs) are progressively being adopted in medicine to aid in clinical decisionsupport, medical documentation, diagnostics, and patient education, demonstrating both significant promise and emerging risks in healthcare workflows [\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. An early benchmark study found that ChatGPT achieved near-passing performance on a radiology board-style examination, indicating the depth of domain-specific knowledge LLMs can provide while highlighting the necessity for validation in specialized settings [4]. They can also offer accurate and detailed responses to patient inquiries in Interventional Radiology (IR), providing new opportunities for patient education by reducing fear and improving compliance, though also posing some risks of misinformation [\u003cspan additionalcitationids=\"CR6\" citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIR has significantly expanded its applications over the last years, largely due to advancements in technology, offering advantages such as reduced risk and shorter recovery times compared to traditional open surgery [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. However, IR lacks public awareness and understanding by patients: a study found that 65% of patients were unfamiliar with IR and 72% did not recognize IR physicians as such [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. As a result, IR physicians must overcome a more significant \u0026ldquo;gap\u0026rdquo; in educating patients to help them make well-informed decisions.\u003c/p\u003e\u003cp\u003eThis gap is particularly evident for less commonly performed procedures, such as Bleomycin electrosclerotherapy (BEST), transarterial periarticular embolization (TAPE) and CT-guided high-dose-rate (HDR) brachytherapy, where patient knowledge is often especially limited.\u003c/p\u003e\u003cp\u003eBEST combines electroporation with bleomycin in the treatment of vascular malformations [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e], improving local drug uptake while reducing systemic effects [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e], and exhibits promising outcomes in multiple studies [\u003cspan additionalcitationids=\"CR14\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. TAPE is an increasingly popular therapeutic procedure for chronic joint pain, often when drug and physical therapy options have been exhausted, and can be performed comparatively quickly and inexpensively, contributing to significant pain relief [\u003cspan additionalcitationids=\"CR17\" citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. CT-HDR brachytherapy is a minimally invasive procedure for liver tumors, demonstrating favorable outcomes with extended survival and excellent local tumor control in numerous studies [\u003cspan additionalcitationids=\"CR20\" citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. However, the three procedures are limited to specialized centers and require comprehensive patient education and informed consent [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e], which can be time- and resource-intensive in clinical practice [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Despite a growing focus on patient empowerment [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e], many patients lack sufficient understanding of medical procedures [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] and their risks [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThe safety and efficacy of LLMs in educating patients about less common procedures like BEST, TAPE, or CT-HDR brachytherapy remain uncertain, as does their potential to reduce clinicians' workload, improve patient satisfaction, or support clinical workflows. Given their rapid evolution, this study evaluates the accuracy of four LLMs in answering common patient questions about CT-HDR brachytherapy, TAPE, and BEST, aiming to assess their clinical relevance and ability to enhance patient communication in IR.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eStudy Design\u003c/h2\u003e\u003cp\u003eDue to the human-generated sample dataset, ethics committee approval was not required.\u003c/p\u003e\u003cp\u003eIn this prospective evaluation study, two radiology residents (fifth and second year, respectively) created 107 questions based on commonly asked patient questions encountered in daily clinical practice before TAPE (35), CT-HDR brachytherapy (34) and BEST (36). To ensure clinical relevance and realism, the questions were subsequently reviewed and validated by two board-certified interventional radiologists (18 and 11 years of experience).\u003c/p\u003e\u003cp\u003eResponses were generated by four state-of-the-art LLMs, three open-source models (DeepSeek-V3, OpenBioLLM-8b, BioMistral-7b) and one proprietary model (ChatGPT-4o). The two interventional radiologists independently evaluated responses using a 5-point Likert scale (Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e). ChatGPT-4o and DeepSeek-V3 were selected for their public visibility and scientific validation [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. OpenBioLLM-8b and BioMistral-7b enabled comparison with smaller, medically pre-trained language models.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eQuestion design and Prompting\u003c/h3\u003e\n\u003cp\u003eQuestion categories were general information, preparation/planning, risks/contraindications, recovery/follow-up and side effects/complications. Two examples of asked questions:\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003e\u003cem\u003e\u0026ldquo;What medication do I have to stop taking before a TAPE and for how long?\u003c/em\u003e\u003c/p\u003e\u003cp\u003eor \u003cem\u003e\u0026ldquo;What happens if I cannot withstand the radiation during CT-HDR brachytherapy and have to interrupt the treatment?\u0026rdquo;\u003c/em\u003e\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eAll four LLMs were prompted identically with\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003e\u0026rdquo;\u003cem\u003eI am a patient. I am due to have a\u003c/em\u003e CT-HDR brachytherapy \u003cem\u003eand have some questions about this procedure. Can you answer each of the following questions in an understandable way.\u003c/em\u003e\u0026rdquo;, and respectively the same for TAPE and BEST.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eAll questions for each procedure were subsequently submitted via API to all four LLMs (on December 10th 2024) using identical prompts to ensure comparability and consistency. All questions are provided as supplementary material, all responses and ratings are documented and available upon request.\u003c/p\u003e\n\u003ch3\u003eResponse Evaluation\u003c/h3\u003e\n\u003cp\u003eResponse accuracy from DeepSeek-V3, OpenBioLLM-8b, BioMistral-7b, and ChatGPT-4o was rated by two independent board-certified radiologists using a 5-point Likert scale (1\u0026thinsp;=\u0026thinsp;Very inaccurate / completely false, very likely to mislead; 2\u0026thinsp;=\u0026thinsp;Inaccurate / mostly false, likely to mislead; 3\u0026thinsp;=\u0026thinsp;Neutral / moderately accurate, overall acceptable; 4\u0026thinsp;=\u0026thinsp;Accurate / mostly correct, only very few inaccuracies, unlikely to mislead; 5\u0026thinsp;=\u0026thinsp;Very accurate / completely correct, very unlikely to mislead; displayed in Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e) and subsequently statistically compared. LLMs were anonymized to ensure blinded evaluation by the radiologists.\u003c/p\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003eStatistical Analysis\u003c/h2\u003e\u003cp\u003eCategorical and ordinal ratings are presented as counts and percentages (n, %), and overall scores are reported as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD, calculated from the radiologists\u0026rsquo; individual means. To assess statistically significant differences between models, a two-tailed, four sample Friedman-test was applied (repeated measures, ordinal scale). P-values below 0.05 were considered statistically significant. If so, post-hoc Wilcoxon signed-rank tests with Holm correction were used to assess pairwise statistically significant differences between LLM ratings. Holm correction adjusted the significance level for multiple comparisons. Interrater agreement was assessed using the two-way random effects single-measure intraclass correlation coefficient (ICC) and Cohen\u0026rsquo;s Kappa to evaluate absolute agreement between both raters per model and across all items (Supplementary Table \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e). Analyses and visualizations were performed using Python (version 3.9.13) with packages including Pandas (version 2.2.2), NumPy (version 1.23.1), SciPy (version 1.13.1; Scikit-Posthocs Version 0.10.0), Matplotlib (version 3.9.2) and Seaborn (version 0.13.2).\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003ePerformance grading of the four LLMs is displayed in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e (BEST), Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e (CT-HDR brachytherapy) and Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e (TAPE). Statistics for all questions and all covered categories are illustrated in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e. Cumulative ratings from both radiologists are displayed in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e (a)-(c).\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eGrading of the performance of four Large Language Models (OpenBioLLM-8b, BioMistral-7b, ChatGPT-4o, DeepSeek-V3) in answering patient questions before a \u003cb\u003eBEST\u003c/b\u003e procedure (evaluations of both radiologists combined). With two radiologists grading 36 BEST-related questions there are 72 ratings in total.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRating\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eOpenBioLLM-8b\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eBioMistral-7b\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eChatGPT-4o\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eDeepSeek-V3\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e5\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e21 (29.17%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e14 (19.44%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e39 (54.17%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e48 (66.67%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e4\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e21 (29.17%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e14 (19.44%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e24 (33.33%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e17 (23.61%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e3\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e13 (18.06%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e14 (19.44%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e6 (8.33%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e3 (4.17%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e2\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e8 (11.11%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e12 (16.67%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e2 (2.78%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e2 (2.78%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e1\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e9 (12.50%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e18 (25.00%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e1 (1.39%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e2 (2.78%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eMean (\u0026plusmn;\u0026thinsp;SD)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e\u003cem\u003e3.51 (\u0026plusmn;\u0026thinsp;1.15)\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e\u003cem\u003e2.92 (\u0026plusmn;\u0026thinsp;1.35)\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e\u003cem\u003e4.36 (\u0026plusmn;\u0026thinsp;0.71)\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e\u003cem\u003e4.49 (\u0026plusmn;\u0026thinsp;0.77)\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eGrading of the performance of four Large Language Models (OpenBioLLM-8b, BioMistral-7b, ChatGPT-4o, DeepSeek-V3) in answering patient questions before a \u003cb\u003eCT-HDR\u003c/b\u003e procedure (evaluations of both radiologists combined). With two radiologists grading 34 CT-HDR-related questions there are 68 ratings in total.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRating\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eOpenBioLLM-8b\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eBioMistral-7b\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eChatGPT-4o\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eDeepSeek-V3\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e5\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e10 (14.71%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e7 (10.29%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e19 (27.94%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e35 (51.47%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e4\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e28 (41.18%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e17 (25.00%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e26 (38.24%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e21 (30.88%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e3\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e13 (19.12%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e25 (36.76%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e16 (23.53%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e7 (10.29%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e2\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e8 (11.76%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e9 (13.24%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e5 (7.35%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e3 (4.41%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e1\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e9 (13.24%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e10 (14.71%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e2 (2.94%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e2 (2.94%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eMean (\u0026plusmn;\u0026thinsp;SD)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e\u003cem\u003e3.32 (\u0026plusmn;\u0026thinsp;1.25)\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e\u003cem\u003e3.03 (\u0026plusmn;\u0026thinsp;1.18)\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e\u003cem\u003e3.81 (\u0026plusmn;\u0026thinsp;1.03)\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e\u003cem\u003e4.24 (\u0026plusmn;\u0026thinsp;1.01)\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eGrading of the performance of four Large Language Models (OpenBioLLM-8b, BioMistral-7b, ChatGPT-4o, DeepSeek-V3) in answering patient questions before a \u003cb\u003eTAPE\u003c/b\u003e procedure (evaluations of both radiologists combined). With two radiologists grading 35 TAPE-related questions there are 70 ratings in total.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRating\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eOpenBioLLM-8b\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eBioMistral-7b\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eChatGPT-4o\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eDeepSeek-V3\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e5\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e15 (21.43%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e19 (27.14%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e28 (40.00%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e32 (45.71%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e4\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e19 (27.14%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e15 (21.43%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e30 (42.86%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e23 (32.86%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e3\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e17 (24.29%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e16 (22.86%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e9 (12.86%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e12 (17.14%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e2\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e13 (18.57%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e10 (14.29%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e2 (2.86%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e3 (4.29%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003e1\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e6 (8.57%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e10 (14.29%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e1 (1.43%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0 (0.00%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eMean (\u0026plusmn;\u0026thinsp;SD)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e\u003cem\u003e3.34 (\u0026plusmn;\u0026thinsp;1.16)\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e\u003cem\u003e3.33 (\u0026plusmn;\u0026thinsp;1.28)\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e\u003cem\u003e4.17 (\u0026plusmn;\u0026thinsp;0.64)\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e\u003cem\u003e4.20 (\u0026plusmn;\u0026thinsp;0.77)\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eComparative analysis of the performance of four Large Language Models (OpenBioLLM-8b, BioMistral-7b, ChatGPT-4o, DeepSeek-V3) in answering patient questions before an interventional radiology procedure (BEST, CT-HDR, TAPE)\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eOpenBioLLM-8b\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eBioMistral-7b\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eChatGPT-4o\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eDeepSeek-V3\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eP-value\u003c/p\u003e\u003cp\u003eFriedman-Test\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colspan=\"6\" nameend=\"c6\" namest=\"c1\"\u003e\u003cp\u003eBEST\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eAll Questions\u003c/b\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003e(Mean, SD)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.51 (\u0026plusmn;\u0026thinsp;1.15)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2.92 (\u0026plusmn;\u0026thinsp;1.35)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e4.36 (\u0026plusmn;\u0026thinsp;0.71)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e4.49 (\u0026plusmn;\u0026thinsp;0.77)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001**\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGeneral Information (Mean, SD)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2.88 (\u0026plusmn;\u0026thinsp;1,30)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2.50 (\u0026plusmn;\u0026thinsp;1,46)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e4.71 (\u0026plusmn;\u0026thinsp;0,50)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e4.12 (\u0026plusmn;\u0026thinsp;1,11)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001**\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePreparation / Planning (Mean, SD)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e4.10 (\u0026plusmn;\u0026thinsp;0,42)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3.30 (\u0026plusmn;\u0026thinsp;1,44)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e4.50 (\u0026plusmn;\u0026thinsp;0,50)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e4.60 (\u0026plusmn;\u0026thinsp;0,55)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.186\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRisks / Contraindications (Mean, SD)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.86 (\u0026plusmn;\u0026thinsp;1,38)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3.14 (\u0026plusmn;\u0026thinsp;1,25)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e3.64 (\u0026plusmn;\u0026thinsp;1,03)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e4.86 (\u0026plusmn;\u0026thinsp;0,24)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e0.016*\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRecovery / Follow-Up (Mean, SD)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e4.00 (\u0026plusmn;\u0026thinsp;0,58)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2.93 (\u0026plusmn;\u0026thinsp;1,40)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e4.43 (\u0026plusmn;\u0026thinsp;0,53)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e4.50 (\u0026plusmn;\u0026thinsp;0,58)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e0.014*\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSide Effects / Complications (Mean, SD)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.30 (\u0026plusmn;\u0026thinsp;1,04)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3.20 (\u0026plusmn;\u0026thinsp;1,35)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e4.30 (\u0026plusmn;\u0026thinsp;0,45)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e4.70 (\u0026plusmn;\u0026thinsp;0,45)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e0.023*\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"6\" nameend=\"c6\" namest=\"c1\"\u003e\u003cp\u003e\u003cb\u003eCT-HDR\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eAll Questions\u003c/b\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003e(Mean, SD)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.32 (\u0026plusmn;\u0026thinsp;1.13)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3.03 (\u0026plusmn;\u0026thinsp;1.06)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e3.81 (\u0026plusmn;\u0026thinsp;0.76)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e4.24 (\u0026plusmn;\u0026thinsp;0.81)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001**\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGeneral Information (Mean, SD)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.05 (\u0026plusmn;\u0026thinsp;1,28)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2.95 (\u0026plusmn;\u0026thinsp;0,86)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e3.50 (\u0026plusmn;\u0026thinsp;0,62)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e4.05 (\u0026plusmn;\u0026thinsp;0,86)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e0.025*\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePreparation / Planning (Mean, SD)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.56 (\u0026plusmn;\u0026thinsp;1,13)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3.39 (\u0026plusmn;\u0026thinsp;1,11)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e3.67 (\u0026plusmn;\u0026thinsp;0,61)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e4.33 (\u0026plusmn;\u0026thinsp;0,97)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.178\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRisks / Contraindications (Mean, SD)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.10 (\u0026plusmn;\u0026thinsp;1,39)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2.40 (\u0026plusmn;\u0026thinsp;1,47)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e4.40 (\u0026plusmn;\u0026thinsp;0,82)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e4.40 (\u0026plusmn;\u0026thinsp;0,22)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e0.007**\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRecovery / Follow-Up (Mean, SD)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.40 (\u0026plusmn;\u0026thinsp;1,14)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3.10 (\u0026plusmn;\u0026thinsp;0,96)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e3.70 (\u0026plusmn;\u0026thinsp;1,04)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e4.10 (\u0026plusmn;\u0026thinsp;0,96)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.176\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSide Effects / Complications (Mean, SD)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.60 (\u0026plusmn;\u0026thinsp;0,74)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3.10 (\u0026plusmn;\u0026thinsp;1,08)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e4.20 (\u0026plusmn;\u0026thinsp;0,67)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e4.40 (\u0026plusmn;\u0026thinsp;0,82)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.010*\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"6\" nameend=\"c6\" namest=\"c1\"\u003e\u003cp\u003e\u003cb\u003eTAPE\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eAll Questions\u003c/b\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003e(Mean, SD)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.34 (\u0026plusmn;\u0026thinsp;1.16)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3.33 (\u0026plusmn;\u0026thinsp;1.28)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e4.17 (\u0026plusmn;\u0026thinsp;0.64)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e4.20 (\u0026plusmn;\u0026thinsp;0.77)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001**\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGeneral Information (Mean, SD)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.00 (\u0026plusmn;\u0026thinsp;1,06)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3.83 (\u0026plusmn;\u0026thinsp;0,94)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e4.44 (\u0026plusmn;\u0026thinsp;0,53)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e3.94 (\u0026plusmn;\u0026thinsp;0,81)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e0.037*\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePreparation / Planning (Mean, SD)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.30 (\u0026plusmn;\u0026thinsp;1,23)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3.05 (\u0026plusmn;\u0026thinsp;1,30)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e3.85 (\u0026plusmn;\u0026thinsp;0,88)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e4.25 (\u0026plusmn;\u0026thinsp;0,59)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.057\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRisks / Contraindications (Mean, SD)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.90 (\u0026plusmn;\u0026thinsp;1,08)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2.60 (\u0026plusmn;\u0026thinsp;1,71)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e4.30 (\u0026plusmn;\u0026thinsp;0,57)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e3.90 (\u0026plusmn;\u0026thinsp;0,96)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.392\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRecovery / Follow-Up (Mean, SD)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.83 (\u0026plusmn;\u0026thinsp;0,75)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3.50 (\u0026plusmn;\u0026thinsp;1,05)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e4.25 (\u0026plusmn;\u0026thinsp;0,42)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e4.67 (\u0026plusmn;\u0026thinsp;0,61)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e\u003cb\u003e0.013*\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSide Effects / Complications (Mean, SD)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2.90 (\u0026plusmn;\u0026thinsp;1,60)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3.50 (\u0026plusmn;\u0026thinsp;1,62)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e4.10 (\u0026plusmn;\u0026thinsp;0,42)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e4.30 (\u0026plusmn;\u0026thinsp;0,97)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.182\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"6\"\u003e*: significant (p\u0026thinsp;\u0026lt;\u0026thinsp;0.050); **: p\u0026thinsp;\u0026lt;\u0026thinsp;0.010; n.s.: p\u0026thinsp;\u0026gt;\u0026thinsp;0.050\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cem\u003eLegend:\u003c/em\u003e SD, standard deviation.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eDisplay of p values of Wilcoxon Signed-Rank test with Holm correction for all questions per intervention\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e\u003cp\u003eP values Wilcoxon Signed-Rank test with Holm correction\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u003cem\u003eOpenBioLLM-8b\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cem\u003eBioMistral-7b\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cem\u003eChatGPT-4o\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cem\u003eDeepSeek-V3\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e\u003cp\u003e\u003cb\u003eBEST -\u003c/b\u003e All Questions\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eOpenBioLLM-8b\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u003cb\u003e-\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.073 (n.s.)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e0.003 **\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001 **\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eBioMistral-7b\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001 **\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001 **\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eChatGPT-4o\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.332 (n.s.)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eDeepSeek-V3\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e\u003cp\u003e\u003cb\u003eCT-HDR -\u003c/b\u003e All Questions\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eOpenBioLLM-8b\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u003cb\u003e-\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.133 (n.s.)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e0.017 *\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e0.001 **\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eBioMistral-7b\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e0.003 **\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001 **\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eChatGPT-4o\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.010 **\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eDeepSeek-V3\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"5\" nameend=\"c5\" namest=\"c1\"\u003e\u003cp\u003e\u003cb\u003eTAPE -\u003c/b\u003e All Questions\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eOpenBioLLM-8b\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u003cb\u003e-\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.000 (n.s.)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e0.001 **\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e0.002 **\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eBioMistral-7b\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u003cb\u003e0.002 **\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e\u003cb\u003e0.002 **\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eChatGPT-4o\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.000 (n.s.)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cem\u003eDeepSeek-V3\u003c/em\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"5\"\u003e*: significant (p\u0026thinsp;\u0026lt;\u0026thinsp;0.050); **: p\u0026thinsp;\u0026lt;\u0026thinsp;0.010; n.s.: p\u0026thinsp;\u0026gt;\u0026thinsp;0.050\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eFor BEST, DeepSeek-V3 achieved the highest average score [4.49 (\u0026plusmn;\u0026thinsp;0.77)], followed by ChatGPT-4o [4.36 (\u0026plusmn;\u0026thinsp;0.71)]. OpenBioLLM-8b [3.51 (\u0026plusmn;\u0026thinsp;1.15)] and BioMistral-7b [2.92 (\u0026plusmn;\u0026thinsp;1.35)] scored significantly lower (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Significant differences were observed in General Information (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), Risks / Contraindications (p\u0026thinsp;=\u0026thinsp;0.016), Recovery / Follow-Up (p\u0026thinsp;=\u0026thinsp;0.014), and Side Effects / Complications (p\u0026thinsp;=\u0026thinsp;0.023), but not in Preparation / Planning (p\u0026thinsp;=\u0026thinsp;0.186).\u003c/p\u003e\u003cp\u003eFor CT-HDR brachytherapy, DeepSeek-V3 again outperformed other models [4.24 (\u0026plusmn;\u0026thinsp;0.81)], followed by ChatGPT-4o [3.81 (\u0026plusmn;\u0026thinsp;0.76)], OpenBioLLM-8b [3.32 (\u0026plusmn;\u0026thinsp;1.13)], and BioMistral-7b [3.03 (\u0026plusmn;\u0026thinsp;1.06)] (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Statistically significant differences were specifically noted in General Information (p\u0026thinsp;=\u0026thinsp;0.025), Risks / Contraindications (p\u0026thinsp;=\u0026thinsp;0.007), and Side Effects / Complications (p\u0026thinsp;=\u0026thinsp;0.010), while Preparation / Planning (p\u0026thinsp;=\u0026thinsp;0.178) and Recovery / Follow-Up (p\u0026thinsp;=\u0026thinsp;0.176) showed no significant differences.\u003c/p\u003e\u003cp\u003eFor TAPE, ChatGPT-4o [4.17 (\u0026plusmn;\u0026thinsp;0.64)] and DeepSeek-V3 [4.20 (\u0026plusmn;\u0026thinsp;0.77)] performed similarly, significantly surpassing OpenBioLLM-8b [3.34 (\u0026plusmn;\u0026thinsp;1.16)] and BioMistral-7b [3.33 (\u0026plusmn;\u0026thinsp;1.28)] (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Significant differences were observed in General Information (p\u0026thinsp;=\u0026thinsp;0.037) and Recovery / Follow-Up (p\u0026thinsp;=\u0026thinsp;0.013). Categories Preparation / Planning (p\u0026thinsp;=\u0026thinsp;0.057), Risks / Contraindications (p\u0026thinsp;=\u0026thinsp;0.392), and Side Effects / Complications (p\u0026thinsp;=\u0026thinsp;0.182) showed no statistically significant differences.\u003c/p\u003e\u003cp\u003eRadar plots in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e illustrate average category scores for each procedure. Interrater agreement varied across models and procedures, with highest consistency for BioMistral-7b (ICC up to 0.69), and lowest for ChatGPT-4o, particularly for TAPE-related questions with ICC as low as 0.13 (ICC and Cohen\u0026rsquo;s Kappa in Supplementary Table \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e; ratings per radiologist for each procedure are displayed in Supplementary Tables S3-S5).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study evaluated the performance of four advanced LLMs in addressing potentially relevant and frequently asked patient questions related to TAPE, BEST, and CT-HDR brachytherapy procedures. According to two interventional radiologists, DeepSeek-V3 and ChatGPT-4o demonstrated strong performance by delivering accurate and understandable responses, while the medically pre-trained OpenBioLLM-8b and BioMistral-7b performed significantly worse across all procedures and produced potentially hazardous answers. For instance, OpenBioLLM-8b failed to identify manifest hyperthyroidism as a contraindication, which is concerning given that all three procedures involve iodinated contrast media, which may cause a potentially life threatening thyrotoxic crisis in patients with hyperthyroidism [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. BioMistral-7b provided inaccurate information regarding radiation exposure, incorrectly asserting that the radiation dose would be similar or lower than that of a chest radiograph in all three procedures.\u003c/p\u003e\u003cp\u003eWhile OpenBioLLM-8b and BioMistral-7b were developed using biomedical training data, it is likely that the limited size and scope of their training datasets, relative to the larger models ChatGPT-4o and DeepSeek-V3, contributed to the inaccurate and potentially harmful responses observed. Responses generated by ChatGPT-4o and DeepSeek-V3 exhibited superior structural organization, enabling step-by-step comprehension of complex medical information.\u003c/p\u003e\u003cp\u003eHowever, the two models showed only low to moderate interrater agreement, underscoring the subjective way LLM responses are interpreted. In contrast, stronger agreement on the lower-performing models could reflect the ease of identifying clearly incorrect or misleading content.\u003c/p\u003e\u003cp\u003eIn this study, less commonly performed IR procedures were intentionally selected to evaluate LLM performance under challenging and less-represented clinical settings. Given the evolving nature of IR, assessing how LLMs respond to newer, or niche procedures is of significant relevance. Although this focus limits generalizability, it highlights the potential and robustness of LLMs in underrepresented areas.\u003c/p\u003e\u003cp\u003eSeveral studies have shown that LLMs can provide accurate responses to patient questions in IR [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. ChatGPT-4 showed strong performance in explaining IR complications but requires health literacy to be fully understood [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. Combining LLMs with educational videos may enhance patient understanding of IR procedures and satisfaction during informed consent [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. However, not all studies reported positive conclusions for LLMs in IR patient education, highlighting incomplete or inaccurate patient education content [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Hofmann and Vairavamurthy found responses incomplete or lacking depth, especially for risks and alternatives, with more experienced reviewers rating LLMs less favorably [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. This issue is not exclusive to ChatGPT, as previous studies identified complications in IR cases that had not been documented during the consent process [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Several studies report issues with LLMs when answering radiology-related questions, including errors, jargon, and hallucinations. Jeblick et al. [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e] found that approximately one-third of simplified radiology reports contained errors potentially harmful to patients. Although patients rated ChatGPT-4.0 superior to medical experts for empathy and usefulness, prior studies indicate that patients frequently struggle to recognize harmful or inaccurate advice [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. This emphasizes the importance of transparent communication and expert oversight, as specialized medical knowledge remains essential to identify and reduce harmful or misleading information produced by LLMs [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThe strong performance of DeepSeek-V3 observed in this study aligns with the literature [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e], but concerns about user data privacy and regulatory challenges persist [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. Broader ethical concerns include legal accountability and the protection of human rights [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. Determining responsibility in cases of patient harm is particularly complex [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e], and current data collection methods pose additional threats to patient confidentiality and privacy [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eBeyond privacy concerns, LLMs could undermine the doctor-patient relationship and reduce trust in medical professionals. However, when appropriately applied, they may improve informed decision-making by providing clear information in complex clinical situations [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e]. Despite these benefits, patient safety must remain the priority in future LLM development, reinforcing the urgent need for the establishment of ethical frameworks to guide responsible LLM use [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThis study compared open-source models (DeepSeek-V3, OpenBioLLM-8b, BioMistral-7b) with the proprietary ChatGPT-4o. Although ChatGPT-4o performed well, concerns remain about transparency and data security risks [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e]. Open-source models offer distinct advantages, including enhanced transparency, greater customization opportunities, and increased control over patient data. Given that DeepSeek-V3 is freely accessible and exhibited high performance, it may represent a more practical tool for broader patient access.\u003c/p\u003e\u003cp\u003eObtaining informed consent in IR is often complex and time-consuming. LLMs could streamline this process, enhance patient empowerment, and improve communication efficiency. However, the complexity of certain IR procedures, including risks associated with off-label device use, demands medical expertise [\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e]. While DeepSeek-V3 and ChatGPT-4.0 performed well in this study, instances of misinformation produced by the two medically pre-trained LLMs, emphasize the continuing need for physician oversight to ensure patient safety [\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e]. Time savings could be achieved by integrating LLMs into structured clinical pathways with outputs limited to low-risk content, while physician oversight is reserved for high-risk areas or responses flagged as low-confidence by the model. This approach could maintain patient safety without requiring full review of all outputs.\u003c/p\u003e\u003cp\u003eFuture studies should focus on model safety, accuracy and adherence to clinical guidelines.\u003c/p\u003e\n\u003ch3\u003eLimitations\u003c/h3\u003e\n\u003cp\u003eThe questions were developed by two radiology residents based on clinical experience within a single institution, limiting the representativeness. Although validated by two board-certified interventional radiologists, rewording patient questions may also have altered their original nuance. The assessment covered only three less commonly performed IR procedures, reducing generalizability. LLMs were selected based on public awareness and prior validation, introducing possible selection bias. No patients were involved in the design or evaluation of the questions, and standardized prompting does not fully reflect real doctor-patient interactions. Responses were evaluated by only two board-certified radiologists, introducing potential subjective bias. Although the study primarily assessed accuracy and misinformation in LLM-generated responses, aspects like linguistic accessibility and empathetic communication may also matter in patient education. Despite the limitations, the findings provide a valuable foundation for future research, ideally incorporating direct patient participation.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eDeepSeek-V3 and ChatGPT-4o demonstrated a strong performance in answering questions related to TAPE, BEST and CT-HDR brachytherapy, highlighting their potential for patient education and communication improvement. This is especially relevant in IR, where complex but minimally invasive procedures are often explained within tight consultation windows. OpenBioLLM-8b and BioMistral-7b produced more frequent inaccuracies, underscoring the risks of integrating smaller, biomedical-specific models into clinical practice. These findings demonstrate that LLMs cannot substitute comprehensive medical consultations yet. Nevertheless, LLMs will play an increasing role in radiology and patient care. Future research should validate these findings, incorporate patient feedback and evaluate LLM integration into clinical workflows.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eLLM Large language model\u003c/p\u003e\u003cp\u003eCT-HDR Computed tomography-guided high-dose-rate\u003c/p\u003e\u003cp\u003eBEST Bleomycin electrosclerotherapy\u003c/p\u003e\u003cp\u003eTAPE Transarterial periarticular embolization\u003c/p\u003e\u003cp\u003eIR Interventional radiology\u003c/p\u003e\u003cp\u003eICC Intraclass correlation coefficient\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eEthics approval and consent to participate:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eNot applicable.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eConsent for publication:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eNot applicable.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAvailability of data and material:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe datasets used and/or analysed during the current study are available from the corresponding author on reasonable request\u003c/p\u003e\n\u003cp\u003eCompeting interests:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003eOutside the submitted work, T.P. is funded in part by the Berlin Institute of Health (BIH). T.P. also receives funding from Berlin Institute of Health (Advanced Clinician Scientist Grant, Platform Grant), Ministry of Education and Research (BMBF, 01KX2021 (RACOON), 01KX2121 (\u0026bdquo;NUM 2.0\u0026ldquo;, RACOON), 68GX21001A, 01ZZ2315D), German Research Foundation (DFG, SFB 1340/2), European Union (H2020, CHAIMELEON: 952172, DIGITAL, EUCAIM:101100633). T.P. also declares relationships with the following companies: research agreements (no personal payments) with AGO, Aprea AB, ARCAGY-GINECO, Astellas Pharma Global Inc. (APGD), Astra Zeneca, Clovis Oncology, Inc., Holaira, Incyte Corporation, Karyopharm, Lion Biotechnologies, Inc., MedImmune, Merck Sharp \u0026amp; Dohme Corp, Millennium Pharmaceuticals, Inc., Morphotec Inc., NovoCure Ltd., PharmaMar S.A. and PharmaMar USA, Inc., Roche, Siemens Healthineers, and TESARO Inc., fees for a book translation (Elsevier B.V.), fees for speaking engagements (Bayer Healthcare). J.N. receives funding from Berlin Institute of Health (Digital Health Accelerator), European Union\u0026rsquo;s Horizon Europe programme (COMFORT, 101079894) and reports personal fees from Eppdata GmbH outside the submitted work.\u003c/p\u003e\n\u003cp\u003eFunding:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eNo funding was received for this study.\u003c/p\u003e\n\u003cp\u003eAuthor\u0026rsquo;s contributions:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eConceptualization, B.L. and S.E.; methodology, B.L. and S.E.; software, S.E.; validation, B.L., S.E., W.M.L., D.S., J.N., A.D., and T.P.; formal analysis, S.E.; investigation, B.L.; resources, T.P.; data curation, B.L. and S.E.; writing - original draft preparation, B.L. and S.E.; writing -review and editing, B.L., S.E., W.M.L., D.S., J.N., A.D. and T.P.; visualization, S.E.; supervision, T.P.; project administration, T.P..\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAcknowledgements:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eWang L, Wan Z, Ni C, Song Q, Li Y, Clayton E et al (2024) Applications and Concerns of ChatGPT and Other Conversational Large Language Models in Health Care: Systematic Review. J Med Internet Res 26(1):e22769. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2196/22769\u003c/span\u003e\u003cspan address=\"10.2196/22769\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBusch F, Hoffmann L, Rueger C, van Dijk EH, Kader R, Ortiz-Prado E et al (2025) Current applications and challenges in large language models for patient care: a systematic review. Commun Med 5(1):1\u0026ndash;13. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s43856-024-00717-2\u003c/span\u003e\u003cspan address=\"10.1038/s43856-024-00717-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eThirunavukarasu AJ, Ting DSJ, Elangovan K, Gutierrez L, Tan TF, Ting DSW (2023) Large language models in medicine. Nat Med 29(8):1930\u0026ndash;1940. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41591-023-02448-8\u003c/span\u003e\u003cspan address=\"10.1038/s41591-023-02448-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBhayana R, Krishna S, Bleakney RR (2023) Performance of ChatGPT on a Radiology Board-style Examination: Insights into Current Strengths and Limitations. Radiology 307(5):e230582. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1148/radiol.230582\u003c/span\u003e\u003cspan address=\"10.1148/radiol.230582\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHofmann HL, Vairavamurthy J (2024) Large language model doctor: assessing the ability of ChatGPT-4 to deliver interventional radiology procedural information to patients during the consent process. CVIR Endovasc 7:83. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s42155-024-00477-z\u003c/span\u003e\u003cspan address=\"10.1186/s42155-024-00477-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eElhakim T, Brea AR, Fidelis W, Paravastu SS, Malavia M, Omer M et al (2025) Enhanced PROcedural Information READability for Patient-Centered Care in Interventional Radiology With Large Language Models (PRO-READ IR). J Am Coll Radiol JACR 22(1):84\u0026ndash;97. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jacr.2024.08.010\u003c/span\u003e\u003cspan address=\"10.1016/j.jacr.2024.08.010\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKooraki S, Hosseiny M, Jalili MH, Rahsepar AA, Imanzadeh A, Kim GH et al (2024) Evaluation of ChatGPT-Generated Educational Patient Pamphlets for Common Interventional Radiology Procedures. Acad Radiol 31(11):4548\u0026ndash;4553. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.acra.2024.05.024\u003c/span\u003e\u003cspan address=\"10.1016/j.acra.2024.05.024\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMashoufi R, Mashoufi R (2023) Interventional Radiology for Disease Management: A Narrative Review. Cureus 15(11):e48603. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.7759/cureus.48603\u003c/span\u003e\u003cspan address=\"10.7759/cureus.48603\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHeister D, Jackson S, Doherty-Simor M, Newton I (2018) An Evaluation of Trends in Patient and Public Awareness of IR. J Vasc Interv Radiol 29(5):661\u0026ndash;668. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jvir.2017.11.023\u003c/span\u003e\u003cspan address=\"10.1016/j.jvir.2017.11.023\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHorbach SER, Rigter IM, Smitt JHS, Reekers JA, Spuls PI, van der Horst CMAM (2016) Intralesional Bleomycin Injections for Vascular Malformations: A Systematic Review and Meta-Analysis. Plast Reconstr Surg 137(1):244\u0026ndash;256. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/PRS.0000000000001924\u003c/span\u003e\u003cspan address=\"10.1097/PRS.0000000000001924\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHorbach SER, Wolkerstorfer A, Jolink F, Bloemen PR, van der Horst CMAM (2020) Electrosclerotherapy as a Novel Treatment Option for Hypertrophic Capillary Malformations: A Randomized Controlled Pilot Trial. Dermatol Surg Off Publ Am Soc Dermatol Surg Al 46(4):491\u0026ndash;498. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/DSS.0000000000002191\u003c/span\u003e\u003cspan address=\"10.1097/DSS.0000000000002191\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGiardino R, Fini M, Bonazzi V, Cadossi R, Nicolini A, Carpi A (2006) Electrochemotherapy a novel approach to the treatment of metastatic nodules on the skin and subcutaneous tissues. Biomed Pharmacother Biomedecine Pharmacother 60(8):458\u0026ndash;462. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.biopha.2006.07.016\u003c/span\u003e\u003cspan address=\"10.1016/j.biopha.2006.07.016\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSchmidt VF, Cangir \u0026Ouml;, Meyer L, Goldann C, Hengst S, Brill R et al (2024) Outcome of bleomycin electrosclerotherapy of slow-flow malformations in adults and children. Eur Radiol 34(10):6425\u0026ndash;6434. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00330-024-10723-6\u003c/span\u003e\u003cspan address=\"10.1007/s00330-024-10723-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDalm\u0026aacute;dy S, Csoma Z, Besenyi Z, Botty\u0026aacute;n K, Ol\u0026aacute;h J, Kem\u0026eacute;ny L et al (2020) New Treatment Option for Capillary Lymphangioma: Bleomycin-Based Electrochemotherapy of an Infant. Pediatrics 146(6):e20200566. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1542/peds.2020-0566\u003c/span\u003e\u003cspan address=\"10.1542/peds.2020-0566\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGuntau M, Cucuruz B, Brill R, Bidakov O, Von der Heydt S, Deistung A et al (2023) Individualized treatment of congenital vascular malformations of the tongue. Clin Hemorheol Microcirc 83(4):421\u0026ndash;429. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3233/CH-221683\u003c/span\u003e\u003cspan address=\"10.3233/CH-221683\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLee SH, Hwang JH, Kim DH, So YH, Park J, Cho SB et al (2019) Clinical Outcomes of Transcatheter Arterial Embolisation for Chronic Knee Pain: Mild-to-Moderate Versus Severe Knee Osteoarthritis. Cardiovasc Intervent Radiol 42(11):1530\u0026ndash;1536. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00270-019-02289-4\u003c/span\u003e\u003cspan address=\"10.1007/s00270-019-02289-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLittle MW, Gibson M, Briggs J, Speirs A, Yoong P, Ariyanayagam T et al (2021) Genicular artEry embolizatioN in patiEnts with oSteoarthrItiS of the Knee (GENESIS) Using Permanent Microspheres: Interim Analysis. Cardiovasc Intervent Radiol 44(6):931\u0026ndash;940. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00270-020-02764-3\u003c/span\u003e\u003cspan address=\"10.1007/s00270-020-02764-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTaslakian B, Miller LE, Mabud TS, Macaulay W, Samuels J, Attur M et al (2023) Genicular artery embolization for treatment of knee osteoarthritis pain: Systematic review and meta-analysis. Osteoarthr Cartil Open 5(2):100342. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ocarto.2023.100342\u003c/span\u003e\u003cspan address=\"10.1016/j.ocarto.2023.100342\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFriedrich S, Busch F, Jonczyk M, Wieners G, B\u0026ouml;ning G, L\u0026uuml;demann WM et al (2023) Combined CT-guided high-dose-rate brachytherapy (CT-HDRBT) and transarterial chemoembolization with irinotecan-loaded microspheres improve local tumor control and progression-free survival in patients with unresectable colorectal liver metastases compared with mono-CT-HDRBT. J Contemp Brachytherapy 15(1):15\u0026ndash;26. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5114/jcb.2023.125480\u003c/span\u003e\u003cspan address=\"10.5114/jcb.2023.125480\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eB\u0026ouml;ning G, B\u0026uuml;ttner L, Jonczyk M, L\u0026uuml;demann WM, Denecke T, Schnapauff D et al (2020) Complications of Computed Tomography-Guided High-Dose-Rate Brachytherapy (CT-HDRBT) and Risk Factors: Results from More than 10 Years of Experience. Cardiovasc Intervent Radiol 43(2):284\u0026ndash;294. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00270-019-02386-4\u003c/span\u003e\u003cspan address=\"10.1007/s00270-019-02386-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBrinkhaus G, Lock JF, Malinowski M, Denecke T, Neuhaus P, Hamm B et al (2014) CT-guided high-dose-rate brachytherapy of liver tumours does not impair hepatic function and shows high overall safety and favourable survival rates. Ann Surg Oncol 21(13):4284\u0026ndash;4292. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1245/s10434-014-3835-y\u003c/span\u003e\u003cspan address=\"10.1245/s10434-014-3835-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eO\u0026rsquo;Dwyer HM, Lyon SM, Fotheringham T, Lee MJ (2003) Informed consent for interventional radiology procedures: a survey detailing current European practice. Cardiovasc Intervent Radiol 26(5):428\u0026ndash;433. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00270-003-0058-1\u003c/span\u003e\u003cspan address=\"10.1007/s00270-003-0058-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003evan Leeuwen DJ, van Delden OM (2015) Ethics and informed consent in patients with hepatocellular carcinoma: Changing roles for hepatologist and radiologist. Clin Liver Dis 6(5):122\u0026ndash;125. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/cld.513\u003c/span\u003e\u003cspan address=\"10.1002/cld.513\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePrashar A, Butt S, Castiglione DG, Shaida N (2021) Informed consent in interventional radiology - are we doing enough? Br J Radiol 94(1122):20201368. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1259/bjr.20201368\u003c/span\u003e\u003cspan address=\"10.1259/bjr.20201368\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePinto A, Giurazza F, Califano T, Rea G, Valente T, Niola R et al (2021) Interventional radiology in gynecology and obstetric practice: Safety issues. Semin Ultrasound CT MR 42(1):104\u0026ndash;112. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1053/j.sult.2020.09.004\u003c/span\u003e\u003cspan address=\"10.1053/j.sult.2020.09.004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSidhu R, Sakellariou V, Layte P, Soliman A (2006) Patient feedback on helpfulness of postal information packs regarding informed consent for endoscopic procedures. Gastrointest Endosc 64(2):229\u0026ndash;234. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.gie.2006.02.055\u003c/span\u003e\u003cspan address=\"10.1016/j.gie.2006.02.055\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSandmann S, Hegselmann S, Fujarski M, Bickmann L, Wild B, Eils R et al (2025) Benchmark evaluation of DeepSeek large language models in clinical decision-making. Nat Med 1\u0026ndash;4. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41591-025-03727-2\u003c/span\u003e\u003cspan address=\"10.1038/s41591-025-03727-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNewton PM, Summers CJ, Zaheer U, Xiromeriti M, Stokes JR, Bhangu JS et al (2025) Can ChatGPT-4o Really Pass Medical Science Exams? A Pragmatic Analysis Using Novel Questions. Med Sci Educ 35(2):721\u0026ndash;729. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s40670-025-02293-z\u003c/span\u003e\u003cspan address=\"10.1007/s40670-025-02293-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRhee CM, Bhan I, Alexander EK, Brunelli SM (2012) Association Between Iodinated Contrast Media Exposure and Incident Hyperthyroidism and Hypothyroidism. Arch Intern Med 172(2):153\u0026ndash;159. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1001/archinternmed.2011.677\u003c/span\u003e\u003cspan address=\"10.1001/archinternmed.2011.677\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eScheschenja M, Viniol S, Bastian MB, Wessendorf J, K\u0026ouml;nig AM, Mahnken AH (2024) Feasibility of GPT-3 and GPT-4 for in-Depth Patient Education Prior to Interventional Radiological Procedures: A Comparative Analysis. Cardiovasc Intervent Radiol 47(2):245\u0026ndash;250. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00270-023-03563-2\u003c/span\u003e\u003cspan address=\"10.1007/s00270-023-03563-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMcCarthy CJ, Berkowitz S, Ramalingam V, Ahmed M (2023) Evaluation of an Artificial Intelligence Chatbot for Delivery of IR Patient Education Material: A Comparison with Societal Website Content. J Vasc Interv Radiol JVIR 34(10):1760\u0026ndash;1768e32. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jvir.2023.05.037\u003c/span\u003e\u003cspan address=\"10.1016/j.jvir.2023.05.037\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKaba E, Beyazal M, \u0026Ccedil;eliker FB, Yel İ, Vogl TJ (2025) Accuracy and Readability of ChatGPT on Potential Complications of Interventional Radiology Procedures: AI-Powered Patient Interviewing. Acad Radiol 32(3):1547\u0026ndash;1553. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.acra.2024.10.028\u003c/span\u003e\u003cspan address=\"10.1016/j.acra.2024.10.028\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMakary MS, Jacob CC, Boggs Z, Brankovic R, Paradiso M, Regalado L (2024) Impact of Educational Videos on Patient Understanding of Interventional Radiology Procedures. Acad Radiol 31(11):4554\u0026ndash;4559. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.acra.2024.08.020\u003c/span\u003e\u003cspan address=\"10.1016/j.acra.2024.08.020\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eForsman T, Silberstein S, Keller EJ (2023) Consent in Interventional Radiology-How Can We Make It Better? Can Assoc Radiol. J J Assoc Can Radiol 74(1):202\u0026ndash;210. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/08465371221101625\u003c/span\u003e\u003cspan address=\"10.1177/08465371221101625\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJeblick K, Schachtner B, Dexl J, Mittermeier A, St\u0026uuml;ber AT, Topalis J et al (2024) ChatGPT makes medicine easy to swallow: an exploratory case study on simplified radiology reports. Eur Radiol 34(5):2817\u0026ndash;2825. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00330-023-10213-1\u003c/span\u003e\u003cspan address=\"10.1007/s00330-023-10213-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eArmbruster J, Bussmann F, Rothhaas C, Titze N, Gr\u0026uuml;tzner PA, Freischmidt H (2024) Doctor ChatGPT, Can You Help Me? The Patient\u0026rsquo;s Perspective: Cross-Sectional Study. J Med Internet Res 26(1):e58831. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2196/58831\u003c/span\u003e\u003cspan address=\"10.2196/58831\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMasanneck L, Schmidt L, Seifert A, K\u0026ouml;lsche T, Huntemann N, Jansen R et al (2024) Triage Performance Across Large Language Models, ChatGPT, and Untrained Doctors in Emergency Medicine: Comparative Study. J Med Internet Res 26(1):e53297. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2196/53297\u003c/span\u003e\u003cspan address=\"10.2196/53297\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiang W, Chen P, Zou X, Lu X, Liu S, Yang J et al (2025) DeepSeek: the Watson to doctors\u0026mdash;from assistance to collaboration. J Thorac Dis 17(2):1103\u0026ndash;1105. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.21037/jtd-2025b-03\u003c/span\u003e\u003cspan address=\"10.21037/jtd-2025b-03\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJiao C, Rosas E, Asadigandomani H, Delsoz M, Madadi Y, Raja H et al (2025) Diagnostic Performance of Publicly Available Large Language Models in Corneal Diseases: A Comparison with Human Specialists. Diagn Basel Switz 15(10):1221. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/diagnostics15101221\u003c/span\u003e\u003cspan address=\"10.3390/diagnostics15101221\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTemsah A, Alhasan K, Altamimi I, Jamal A, Al-Eyadhy A, Malki KH et al DeepSeek in Healthcare: Revealing Opportunities and Steering Challenges of a New Open-Source Artificial Intelligence Frontier. Cureus 17(2):e79221. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.7759/cureus.79221\u003c/span\u003e\u003cspan address=\"10.7759/cureus.79221\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJeyaraman M, Balaji S, Jeyaraman N, Yadav S (2023) Unraveling the Ethical Enigma: Artificial Intelligence in Healthcare. Cureus 15(8):e43262. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.7759/cureus.43262\u003c/span\u003e\u003cspan address=\"10.7759/cureus.43262\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSallam M (2023) ChatGPT Utility in Healthcare Education, Research, and Practice: Systematic Review on the Promising Perspectives and Valid Concerns. Healthc Basel Switz 11(6):887. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/healthcare11060887\u003c/span\u003e\u003cspan address=\"10.3390/healthcare11060887\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang Y, Peng Y, Wang T, Li H, Zhao Z, Gong L et al (2023) The evolution and current situation in the application of dual-energy computed tomography: a bibliometric study. Quant Imaging Med Surg 13(10):6801\u0026ndash;6813. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.21037/qims-23-467\u003c/span\u003e\u003cspan address=\"10.21037/qims-23-467\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOng JCL, Chang SYH, William W, Butte AJ, Shah NH, Chew LST et al (2024) Ethical and regulatory challenges of large language models in medicine. Lancet Digit Health 6(6):e428\u0026ndash;e432. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S2589-7500(24)00061-X\u003c/span\u003e\u003cspan address=\"10.1016/S2589-7500(24)00061-X\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAllen JW, Earp BD, Koplin J, Wilkinson D (2024) Consent-GPT: is it ethical to delegate procedural consent to conversational AI? J Med Ethics 50(2):77\u0026ndash;83. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/jme-2023-109347\u003c/span\u003e\u003cspan address=\"10.1136/jme-2023-109347\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAmin K, Khosla P, Doshi R, Chheang S, Forman HP (2023) Artificial Intelligence to Improve Patient Understanding of Radiology Reports. Yale J Biol Med 96(3):407\u0026ndash;417. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.59249/NKOY5498\u003c/span\u003e\u003cspan address=\"10.59249/NKOY5498\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZvavanjanja RC, Odetoyinbo TO, Rowlands PC, Healey A, Abdelsalam H, Powell S et al (2012) Off label use of devices and drugs in interventional radiology. Clin Radiol 67(3):239\u0026ndash;243. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.crad.2011.06.017\u003c/span\u003e\u003cspan address=\"10.1016/j.crad.2011.06.017\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBoyd K (2015) The impossibility of informed consent? J Med Ethics 41(1):44\u0026ndash;47. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/medethics-2014-102308\u003c/span\u003e\u003cspan address=\"10.1136/medethics-2014-102308\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"cvir-endovascular","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"cire","sideBox":"Learn more about [CVIR Endovascular](https://www.springer.com/journal/42155)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/cire/default.aspx","title":"CVIR Endovascular","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Large language models, interventional radiology, patient education","lastPublishedDoi":"10.21203/rs.3.rs-7329930/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7329930/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cem\u003e\u003cstrong\u003ePurpose\u003c/strong\u003e\u003c/em\u003e: This study evaluates four large language models’ (LLMs) ability to answer common patient questions preceding transarterial periarticular embolization (TAPE), computed tomography (CT)-guided high-dose-rate (HDR) brachytherapy, and bleomycin electrosclerotherapy (BEST). The goal is to evaluate their potential to enhance clinical workflows and patient comprehension, while also assessing associated risks.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cstrong\u003eMaterials and Methods: \u003c/strong\u003e\u003c/em\u003e35 TAPE, 34 CT‑HDR brachytherapy, and 36 BEST related questions were presented to ChatGPT-4o, DeepSeek-V3, OpenBioLLM-8b, and BioMistral-7b. The LLM-generated responses were independently assessed by two board-certified radiologists. Accuracy was rated on a 5-point Likert scale. Statistics compared LLM performance across question categories for patient-education suitability.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cstrong\u003eResults: \u003c/strong\u003e\u003c/em\u003eDeepSeek-V3 attained the highest mean scores for BEST [4.49 (± 0.77)] and CT-HDR [4.24 (± 0.81)] and demonstrated comparable performance to ChatGPT-4o for TAPE-related questions (DeepSeek-V3 [4.20 (± 0.77)] vs. ChatGPT-4o [4.17 (± 0.64)]; p = 1.000). In contrast, OpenBioLLM-8b (BEST 3.51 (± 1.15), CT-HDR 3.32 (± 1.13), TAPE 3.34 (± 1.16)) and BioMistral-7b (BEST 2.92 (± 1.35), CT-HDR 3.03 (± 1.06), TAPE 3.33 (± 1.28)) performed significantly worse than DeepSeek-V3 and ChatGPT-4o across all procedures. Preparation/Planning was the only category without statistically significant differences across all three procedures.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cstrong\u003eConclusion: \u003c/strong\u003e\u003c/em\u003eDeepSeek‑V3 and ChatGPT‑4o excelled on TAPE, BEST and CT‑HDR brachytherapy questions, indicating potential to enhance patient education in interventional radiology, where complex but minimally invasive procedures often are explained in brief consultations. However, OpenBioLLM‑8b and BioMistral‑7b exhibited more frequent inaccuracies, suggesting that LLMs cannot replace comprehensive clinical consultations yet. Patient feedback and clinical workflow implementation should validate these findings.\u003c/p\u003e","manuscriptTitle":"Comparative Evaluation of State‑of‑the‑Art Large Language Models for Patient Education Prior to Interventional Radiology procedures","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-08-22 13:12:38","doi":"10.21203/rs.3.rs-7329930/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Minor revision","date":"2025-09-08T09:30:47+00:00","index":"","fulltext":""},{"type":"reviewerAgreed","content":"","date":"2025-08-20T06:11:58+00:00","index":0,"fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-08-15T07:44:29+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-08-12T22:02:06+00:00","index":"","fulltext":""},{"type":"submitted","content":"CVIR Endovascular","date":"2025-08-11T09:05:27+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"cvir-endovascular","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"cire","sideBox":"Learn more about [CVIR Endovascular](https://www.springer.com/journal/42155)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/cire/default.aspx","title":"CVIR Endovascular","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"dc5fb424-9326-44cc-a073-81ed806a45ce","owner":[],"postedDate":"August 22nd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-10-20T16:03:12+00:00","versionOfRecord":{"articleIdentity":"rs-7329930","link":"https://doi.org/10.1186/s42155-025-00609-z","journal":{"identity":"cvir-endovascular","isVorOnly":false,"title":"CVIR Endovascular"},"publishedOn":"2025-10-13 15:58:08","publishedOnDateReadable":"October 13th, 2025"},"versionCreatedAt":"2025-08-22 13:12:38","video":"","vorDoi":"10.1186/s42155-025-00609-z","vorDoiUrl":"https://doi.org/10.1186/s42155-025-00609-z","workflowStages":[]},"version":"v1","identity":"rs-7329930","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7329930","identity":"rs-7329930","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}