Influence of deep learning image reconstruction and adaptive statistical iterative reconstruction-V on automated Alberta Stroke Program Early CT Score- evaluation

Influence of deep learning image reconstruction and adaptive statistical iterative reconstruction-V on automated Alberta Stroke Program Early CT Score- evaluation | 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 Influence of deep learning image reconstruction and adaptive statistical iterative reconstruction-V on automated Alberta Stroke Program Early CT Score- evaluation Estelle Akl, Daniel Cantré, Matthias Lütgens, Wiebke Hermann, and 5 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7904400/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 18 Feb, 2026 Read the published version in Neuroradiology → Version 1 posted You are reading this latest preprint version Abstract Purpose The Alberta Stroke Program Early CT Score (ASPECTS) and advances in CT reconstruction play an important role in the neurodiagnostic workflow. This study examines the effect of these reconstruction techniques on automated ASPECTS. Methods In a retrospective study, 173 patients (median age 79 years, 39% female) with suspected middle cerebral artery infarction underwent non-contrast CT scans reconstructed with Filtered Back Projection (FBP), ASIR-V (30% and 60%), and DLIR (low, medium, and high). Automated ASPECTS were analyzed, with FBP as the reference standard. Results Bland–Altman analysis revealed a mean bias of ASIR and DLIR underestimating ASPECTS compared to FBP, which was less pronounced for ASIR-V 30% (-0.057 ) and DLIR-L (-0.069) compared to ASIR-V 60% (-0.126), DLIR-M (-0.121), and DLIR-H (-0.086). The region with the highest overestimation, compared to FBP, was M3 (n = 23), and with the highest underestimation was the insular ribbon (n = 51). Regarding the ASPECTS < 6 threshold, most patients were re-classified from ASPECTS ≤ 5 to ASPECTS ≥ 6 with DLIR-M (n = 5 ), which also showed the strongest agreement with expert consensus (κ = 0.352). Conclusion Both ASIR-V and DLIR led to only a minor underestimation of ASPECTS compared to FBP. However, more patients were overestimated to ASPECTS ≥ 6, making them available for endovascular therapy, which was most pronounced for DLIR-M. DLIR-M also exhibited the highest agreement with expert consensus for automated ASPECTS. Careful selection of reconstruction parameters, as well as further optimization and standardization of these techniques, is therefore essential for broader application in stroke imaging. Stroke ASPECTS computed tomography deep learning iterative reconstruction Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Key Points CT Reconstruction techniques influence automated ASPECT scoring reliability. Insula, as well as the cortical M3 ASPECTS region, seemed more susceptible to algorithmic variability. More patients were overestimated from ASPECTS ≤ 5 to ASPECTS ≥ 6, making them available for endovascular therapy. DLIR-M demonstrated the highest agreement between automated ASPECTS assessment and expert-derived ground truth. Background The Alberta Stroke Program Early Computed Tomography Score (ASPECTS) is widely used to evaluate early ischemic changes in patients with acute ischemic stroke, offering a standardized approach for estimating the extent of the ischemic region [ 1 ]. In recent guidelines, an ASPECTS score of 6 or higher is a requirement or criterion for selecting patients who should receive endovascular therapy [ 2 , 3 ]. ASPECTS is also used for the evaluation of stroke severity and prognosis in acute ischemic stroke [ 4 , 5 ]. Traditional manual ASPECT scoring, however, can be subject to considerable variability among observers. To address these limitations, automated ASPECTS tools based on artificial intelligence (AI), including deep learning (DL) and convolutional neural networks, have emerged, capable of complex tasks such as feature extraction, classification, and segmentation in medical imaging [ 6 , 7 ]. The integration of AI into stroke imaging has also transformed other aspects of the diagnostic workflow [ 8 ]. Historically, Filtered Back Projection (FBP) was the primary method for CT image reconstruction due to its computational efficiency. However, FBP has well-documented limitations, including its inability to effectively manage noise, particularly at lower radiation doses or with obese patients. Iterative Reconstruction (IR) methods were subsequently introduced, offering improved noise management and dose reduction. More recently, Deep Learning Image Reconstruction (DLIR) has emerged as a powerful alternative to FBP and IR, enhancing noise reduction and improving image quality, while maintaining an acceptable radiation dose. DLIR utilizes trained artificial neural networks to refine images and improve diagnostic confidence, particularly in challenging imaging conditions. Recent studies have demonstrated that DLIR-based methods outperform traditional IR by producing images with reduced noise and superior contrast-to-noise ratios, thereby enhancing the diagnostic potential for various pathologies [ 9 – 11 ]. In this study, we aim to evaluate the impact of different reconstruction algorithms for CT on the performance of automated ASPECT scoring in patients with acute ischemic stroke. Understanding the differences among these reconstruction algorithms could contribute to optimizing imaging protocols and ultimately improving clinical outcomes for stroke patients. Methods Patient selection and study design This study was designed as a retrospective, single-center study. Patients were identified by a retrospective search of our radiology information system (Centricity 5.0, GE Healthcare, Barrington, Illinois). Inclusion criteria were patients who received a non-contrast head CT (NCCT) and additional CT angiography (CTA) at our institution by suspicion of an acute stroke in the middle cerebral artery (MCA) territory between May 13, 2024, and November 19, 2024 (n=490). We excluded patients with other pathologies (e.g. tumor, trauma, headache…), (n=159), symptoms suggesting an involvement of the anterior cerebral artery territory and posterior circulation (n=64), hemorrhagic strokes (n=39), large, chronic, ischemic areas in the MCA territory (n= 15) and more than 24 hours after the onset of MCA stroke symptoms (n=8). Twenty patients were excluded due to technical issues with the automated ASPECTS software, and 12 patients were excluded because their datasets were incomplete. One patient was excluded due to artifacts caused by a cochlear implant (Fig. 1). Human Ethics and Consent to Participate declarations The study protocol was approved by the Ethics Committee of Rostock University Medical Faculty and informed consent to participate was waived due to the retrospective nature of this study. The study was conducted in compliance with the Declaration of Helsinki in its current form. CT acquisition protocol All included patients underwent CT imaging of the head as per the standard acute stroke protocol, which was conducted using a 128-slice CT scanner (Revolution CT, GE Healthcare). Scanning parameters included a slice thickness of 0,625 mm, reconstruction of 3,0 mm, voltage of 120 KV, and a current using a tube current modulation, determined based on the Scout in mA. The median CT dose index was 47 mGy (IQR 46-52 mGy). Image Reconstruction and Analysis For all patients, NCCT images of the brain were reconstructed on the Revolution CT scanner (software version 2.1B) using standard Filtered Back Projection (FBP), a new generation of adaptive statistical iterative reconstruction (ASIR-V at 30% and 60% strength), as well as three strength levels of deep-learning-based reconstruction (low (DLIR_L), medium (DLIR_M), and high (DLIR_H); “TrueFidelityTM”, GE Healthcare, Chicago) (Figure 2). The image slice thickness used for image reconstruction, automated ASPECT score determination, was reconstructed using the standard convolution kernel and a slice thickness of 2.5 mm, as recommended by the manufacturer for optimal ASPECT score evaluation. The Alberta Stroke Program Early CT Score (ASPECTS) was obtained automatically using commercially available software (AW server 3.2 software, GE HealthCare). The scoring system examines ten specific regions within the middle cerebral artery (MCA) territory: the caudate nucleus (C), lentiform nucleus (L), internal capsule (IC), insular ribbon (I), as well as the cortical regions M1, M2, and M3 (anterior, middle, and posterior third of the lower MCA territory, respectively), and M4, M5, and M6 (anterior, middle, and posterior third of the higher MCA territory, respectively). Each region demonstrating signs of early ischemic changes is deducted from the total of ten, resulting in a final ASPECTS ranging from 0 (all areas affected by early ischemic changes) to 10 (no changes) [7]. Ground truth was established by expert consensus between one radiology and one neuroradiology consultant (with 11 and 12 years of experience in stroke imaging) by using all available imaging at the acute stage, including NCCT with FBP, ASIR, and DLIR technique, CT angiography, CT perfusion, digital subtraction angiography, and follow-up CT or MRI. They were both blinded to the results of the automated ASPECTS rating. Statistical Analysis Statistical analysis was performed using GraphPad Prism version 9.0.0 (GraphPad Software LLC). FBP was selected as the reference for comparison with ASIR and DLIR. Bland-Altman plots were used to determine the mean difference and limits of agreement between FBP and the other reconstruction techniques [12–14]. The agreement between inconsistent automated ASPECTS ratings of FBP, ASIR, and DLIR reconstructions and the ground truth was assessed using weighted kappa analysis. Results Patient Characteristics The final study population included 173 participants with suspected MCA infarction and symptom onset within 24 hours. The median time from stroke onset to CT imaging was 3 hours (IQR 1.5-6 hours). They had a median age of 79 years (IQR 71–86 years) and 39% (n = 68) were female. In 117 (68%) patients, the results of the automated ASPECTS did not differ despite the use of different CT reconstruction techniques. In 56 patients (32%), one or more score results of the automated ASPECTS differed between the different CT reconstruction techniques. The patients ' characteristics are summarized in Table 1 . Table 1 Patient Characteristics Characteristics Patients included 173 Median age in years (IQR) 79 (71–86) Female n (%) 68 (39%) Median time from stroke onset to CT imaging in hours (IQR) 2.5 (1.5-6) Inconsistency of automated ASPECTS 56 (32%) ASPECTS, Alberta Stroke Program Early CT Score. Agreement of total ASPECTS between FBP and ASIR-V or DLIR Examples of ASPECTS analysis of different CT reconstruction techniques in a representative patient are shown in Fig. 3 . Bland-Altman plots of all reconstruction techniques (ASIR-V at 30% and 60%, DLIR-L, DLIR-M, and DLIR-H) compared to FBP as the reference standard were obtained. ASIR-V 30% showed the best agreement regarding ASPECTS compared to FBP, with a mean bias of − 0.057, whereas ASIR-V 60% had the highest deviation in scoring results, with a mean bias of − 0.126. The mean bias for DLIR-L, DLIR-M, and DLIR-H was – 0.069, − 0.121, and – 0.086 points, respectively. Detailed results of the Bland–Altman analysis are shown in Table 2 . Table 2 Bland–Altman Analysis for ASPECTS of FBP vs. ASIR-V and DLIR 95% Limits of Agreement Comparison Mean Bias SD of Bias Upper Lower FBP Reference ASIR-V 30% -0.057 0.489 -1.017 0.902 ASIR-V 60% -0.126 0.605 -1.311 1.058 DLIR-L -0.069 0.632 -1.308 1.170 DLIR-M -0.121 0.682 -1.457 1.216 DLIR-H -0.086 0.767 -1.589 1.417 FBP serves as the reference standard. ASIR-V, adaptive statistical iterative reconstruction-V; DLIR, deep learning-based image reconstruction in low (L), medium (M), or high (H) strength; FBP, filtered back projection. Impact on the Classification of Patients into Risk Categories Based on ASPECTS Reclassifications of ASPECTS categories with FBP as a reference standard were found for each reconstruction technique. Most of the re-classifications were from moderate (score 5–7) to high ASPECTS (score 8–10). With ASIR-V at 30% and 60% strength, three and five patients were re-classified from moderate to high ASPECTS. DLIR at low, medium, and high strength resulted in seven, five, and ten patients, respectively, being re-classified from moderate to high ASPECTS. However, patients were also re-classified from a high to moderate ASPECTS with one and two patients regarding ASIR-V at 30% and 60% strength, as well as four, two, and three patients regarding DLIR at low, medium, and high strength, respectively. With ASIR-V at 60% and DLIR at medium and high strength, one case was reclassified from a low (score 0–4) to a medium ASPECTS (score 5–7). With DLIR at low strength, two patients were reclassified from a low to a medium ASPECTS, and with ASIR-V at 30% strength, one patient was reclassified from a medium to a low ASPECTS. With ASIR-V at 30% and 60% strength, one patient was re-classified from ASPECTS ≤ 5 to ASPECTS ≥ 6, as well as three, five, and four patients with DLIR at low, medium, and high strength. With ASIR-V at 30% and DLIR at high strength one patient was re-classified from ASPECTS ≥ 6 to ASPECTS ≤ 5. Detailed classification of the patient cohort within different reconstruction techniques can be found in Fig. 4 and the reclassification is displayed in Table 3 . Table 3 Change in ASPECTS Categories comparing FBP to ASIR-V and DLIR. ASPECTS category Net Change in Number of Patients in Risk Groups ASIR-V 30% ASIR-V 60% DLIR-L DLIR-M DLIR-H high (8–10) -1 + 3 -2 + 5 -4 + 7 -2 + 5 -3 + 10 moderate (5–7) -4 + 1 -5 + 3 -7 + 6 -5 + 3 -10 + 4 low (0–4) 0 + 1 -1 0 -2 0 -1 0 -1 0 ≥ 6 -1 + 1 0 + 1 0 + 3 0 + 5 -1 + 4 ≤ 5 -1 + 1 -1 0 -3 0 -5 0 -4 + 1 ASIR-V, adaptive statistical iterative reconstruction-V; DLIR, deep learning-based image reconstruction in low (L), medium (M), or high (H) strength; FBP, filtered back projection. FBP served as the reference standard. Impact on Regional Distribution of Ischemic Changes Based on ASPECTS Over- and underestimation of ischemic change per ASPECTS region of all reconstruction techniques (ASIR-V at 30% and 60%, DLIR-L, DLIR-M, and DLIR-H) compared to FBP as the reference standard is shown in Fig. 5 . The region with the highest overestimation, summarized from all reconstruction techniques in comparison to FBP was M3 (n = 23) (Fig. 6 ), followed by M2 (n = 18), insula (n = 11), M5 (n = 9), M1 (n = 6), M6 (n = 5), lentiform nucleus as well as caudate nucleus (n = 3) and M4 (n = 3). No overestimation was found for the internal capsule (n = 0). The region with the highest underestimation compared to FBP was the insular ribbon (n = 51) (Fig. 7 ), followed by M4 (n = 29), M2 (n = 23), M5 (n = 16), M1 (n = 14), M1 (n = 11), M3 (n = 10), lentiform nucleus (n = 8) followed by caudate nucleus as well as the internal capsule (n = 1). With regards to the reconstruction techniques, the highest number of overestimation of ASPECTS regions, compared to FBP, was found with DLIR-H (n = 28), followed by DLIR-L (n = 19), DLIR-M (n = 17), ASIR-60% (= n = 10), and ASIR-30% (n = 9). The highest number of underestimation of ASPECTS regions, compared to FBP, was found with DLIR-H (n = 43), followed by DLIR-M (n = 38), DLIR-L (n = 33), ASIR-60% (n = 31) and ASIR-30% (n = 19). Agreement between consensus and automated ASPECTS of FBP, ASIR, and DLIR In 56 patients (32%), inconsistent results of the automated ASPECTS were found between different CT reconstruction techniques and were compared to expert consensus. DLIR-M showed the highest agreement with expert consensus (κ = 0.352), followed by ASIR-60% (κ = 0.346), ASIR-30% (κ = 0.291), DLIR-L (κ = 0.263), FBP (κ = 0.229), and DLIR-H (κ = 0.220). Discussion In this study, we investigated the effect of CT reconstruction techniques on the automated ASPECTS in patients with suspected MCA infarction. In total, ASIR-V and DLIR reconstructions showed only a minor underestimation of ASPECTS results compared to FBP. Regarding the ASPECTS < 6 threshold, however, more patients were overestimated when transitioning from ASPECTS ≤ 5 to ASPECTS ≥ 6, making them eligible for endovascular therapy, which was more pronounced for DLIR, especially DLIR-M, compared to ASIR-V. Interestingly, the ASPECTS region with the highest overestimation compared to FBP was M3, and the region with the highest underestimation was the insular ribbon. The highest number of over- and underestimation of ASPECTS regions was found with DLIR-H. Comparing the inconsistent results of the automated ASPECTS between different CT reconstruction techniques and expert consensus as the ground truth, DLIR-M showed the highest agreement. Over the last decade, iterative reconstruction (IR) algorithms have significantly advanced CT technology as a potential alternative to FBP [ 10 , 11 , 15 , 16 ]. These methods improve image quality and reduce radiation dose, making them a criterion standard in CT imaging [ 9 , 15 – 17 ]. Unlike IR and FBP, DLIR utilizes artificial neural networks to iteratively learn noise elimination while preserving anatomical details, therefore representing another step forward in both image quality and dose optimization [ 11 , 16 ]. All advancements in CT reconstruction techniques have reshaped the landscape of neuroradiological CT imaging [ 18 – 21 ] and may therefore also hold significant potential to refine automated ASPECTS, a crucial tool in the evaluation of acute ischemic stroke. Similar studies have primarily focused on the agreement between ASPECTS scores obtained by automated software and those determined by human readers, which have shown overall good [ 22 , 23 ] or even an improvement in the interobserver agreement and accuracy of neuroradiologists and neurologists in their interpretation of ASPECTS [ 24 ]. Löffler et al. compared ASPECTS assessments, which were based on hybrid IR or iterative model reconstruction (IMR) and carried out by humans and software in comparison to a consensus reference. They reported that the automated software improved with IMR, which can further reduce noise levels and improve image contrasts at a given dose level and slice thickness compared to hybrid IR [ 25 ]. Seker et al. NECT compared automated ASPECTS of FBP with Sinogram-Affirmed Iterative Reconstruction (SAFIRE) in strength levels 2 and 4 out of 5. This study included 43 patients with acute occlusion of the M1 segment of the middle cerebral artery and found that automated ASPECTS of SAFIRE, at a strength level of 2, showed the highest consistency when compared to a ground truth provided by an expert with unrestricted data access [ 26 ]. However, in both studies mentioned above, different reconstruction techniques were investigated in comparison to our study, and a detailed analysis of the ASPECTS categories and regions was not performed. Konno et al. compared similar CT reconstruction techniques to those in our study (FBP, ASIR-V at 50%, and DLIR), which did not lead to significant differences in the ASPECTS evaluation of 30 patients. However, only human readers were used for ASPECTS assessment and no automated software [ 27 ]. In our study, the highest number of over- and underestimation of ASPECTS regions was observed with DLIR-H. The insula was the most frequently underestimated ASPECTS region, while overestimation was most notable for the cortical M3 ASPECTS region, suggesting that both regions are more susceptible to algorithmic variability. On the other hand, subcortical regions, such as the internal capsule, lentiform nucleus, and caudate nucleus, displayed greater consistency across reconstruction techniques, suggesting these regions are less influenced by algorithmic variability. Austein et al. also found disagreement in the characteristics of ASPECTS regions scored, which was most notable for the insula and internal capsule, when comparing between humans and two different automated ASPECTS packages. They also reported variability in cortical scoring (M1–M6), which they concluded might be due to the challenges of consistently defining the anatomical borders of these regions [ 22 ]. Interestingly, van Horn et al, who evaluated interrater reliability in a large group of human readers, found that the ASPECTS region with the highest level of agreement was the insular cortex, and with the lowest level of agreement, the M3 region [ 28 ]. The most important clinical use of ASPECTS is in guiding decisions regarding the eligibility for thrombolysis and endovascular therapy [ 1 , 12 ]. Variability in ASPECTS due to reconstruction technique, such as over- or underestimation of the total score, may significantly influence patient outcomes by altering the choice of therapy. According to established guidelines, an ASPECTS score of 6 or higher is a requirement or criterion for endovascular therapy [ 2 , 3 ]. In our study, most patients were reclassified from ASPECTS ≤ 5 to ASPECTS ≥ 6 with DLIR-M, making them eligible for endovascular therapy. Additionally, DLIR-M showed the highest agreement between the consensus expert ground truth and the automated ASPECTS. Our study has several limitations. The evaluation was performed using reconstruction algorithms from a single CT manufacturer, which may limit the generalizability of the findings to other systems. Secondly, while FBP was employed as the reference standard, it is not a definitive method for ASPECTS due to the lack of a universally accepted gold standard. Furthermore, dose reduction was not incorporated into the CT acquisition protocol, although deep learning-based reconstruction and iterative reconstruction techniques have previously shown potential for dose reductions in head CT [ 29 , 30 ]. This was due to the retrospective character of this study and ethical considerations, since a correct ASPECTS has therapeutic consequences and may outweigh radiation protection in the acute emergency setting. Additionally, due to the retrospective nature of this study, simultaneous diffusion-weighted MRI, which could likely provide the most accurate ground truth in cases of acute stroke, was not feasible for every patient in the present study. We attempted to compensate for this by establishing an expert-based ground truth for comparing automated and reader-based ASPECT, with access to clinical information, baseline CT, CT angiography, and CT perfusion imaging, as well as follow-up imaging. In conclusion, different CT reconstruction techniques resulted in only a minor underestimation of automated ASPECTS compared to FBP. Especially the insula, as well as the cortical M3 ASPECTS, seemed more susceptible to algorithmic variability. However, more patients were overestimated from ASPECTS ≤ 5 to ASPECTS ≥ 6, making them available for endovascular therapy, which was most pronounced for DLIR-M. DLIR-M also exhibited the highest agreement with consensus expert ground truth for automated ASPECTS. Careful selection of reconstruction parameters, as well as further optimization and standardization of these techniques, is therefore essential for broader application in stroke imaging to minimize discrepancies and improve consistent and reliable patient management. Declarations Author Contribution E.A., E.B. and M.L. collected data. E.A. and E.B. wrote the main manuscript text and prepared all figures. D.C. and E.B. analyzed the data as part of the "expert consensus". F.M., A-C. K. and E.B. performed the statistics. W.H. and S.L. added neurological clinical aspects. All authors reviewed the manuscript. Statements and Declarations: The authors declare that they have no competing interests. Human Ethics and Consent to Participate declarations: Not applicable. References Uchida K, Shindo S, Yoshimura S, et al (2022) Association Between Alberta Stroke Program Early Computed Tomography Score and Efficacy and Safety Outcomes With Endovascular Therapy in Patients With Stroke From Large-Vessel Occlusion: A Secondary Analysis of the Recovery by Endovascular Salvage for Cerebral Ultra-acute Embolism—Japan Large Ischemic Core Trial (RESCUE-Japan LIMIT). JAMA Neurol 79:1260. https://doi.org/10.1001/jamaneurol.2022.3285 Boulanger J, Lindsay M, Gubitz G, et al (2018) Canadian Stroke Best Practice Recommendations for Acute Stroke Management: Prehospital, Emergency Department, and Acute Inpatient Stroke Care, 6th Edition, Update 2018 . Int J Stroke 13:949–984. https://doi.org/10.1177/1747493018786616 Powers WJ, Rabinstein AA, Ackerson T, et al (2018) 2018 Guidelines for the Early Management of Patients With Acute Ischemic Stroke: A Guideline for Healthcare Professionals From the American Heart Association/American Stroke Association. Stroke 49:. https://doi.org/10.1161/STR.0000000000000158 Esmael A, Elsherief M, Eltoukhy K (2021) Predictive Value of the Alberta Stroke Program Early CT Score (ASPECTS) in the Outcome of the Acute Ischemic Stroke and Its Correlation with Stroke Subtypes, NIHSS, and Cognitive Impairment. Stroke Res Treat 2021:1–10. https://doi.org/10.1155/2021/5935170 Elawady SS, Saway BF, Matsukawa H, et al (2024) Thrombectomy in Stroke Patients With Low Alberta Stroke Program Early Computed Tomography Score: Is Modified Thrombolysis in Cerebral Infarction (mTICI) 2c/3 Superior to mTICI 2b? J Stroke 26:95–103. https://doi.org/10.5853/jos.2023.02292 Soun JE, Chow DS, Nagamine M, et al (2021) Artificial Intelligence and Acute Stroke Imaging. Am J Neuroradiol 42:2–11. https://doi.org/10.3174/ajnr.A6883 Maegerlein C, Fischer J, Mönch S, et al (2019) Automated Calculation of the Alberta Stroke Program Early CT Score: Feasibility and Reliability. Radiology 291:141–148. https://doi.org/10.1148/radiol.2019181228 Liu Y, Wen Z, Wang Y, et al (2024) Artificial intelligence in ischemic stroke images: current applications and future directions. Front Neurol 15:1418060. https://doi.org/10.3389/fneur.2024.1418060 Heinrich A, Streckenbach F, Beller E, et al (2021) Deep Learning-Based Image Reconstruction for CT Angiography of the Aorta. Diagnostics 11:2037. https://doi.org/10.3390/diagnostics11112037 Koetzier LR, Mastrodicasa D, Szczykutowicz TP, et al (2023) Deep Learning Image Reconstruction for CT: Technical Principles and Clinical Prospects. Radiology 306:e221257. https://doi.org/10.1148/radiol.221257 Kim I, Kang H, Yoon HJ, et al (2021) Deep learning–based image reconstruction for brain CT: improved image quality compared with adaptive statistical iterative reconstruction-Veo (ASIR-V). Neuroradiology 63:905–912. https://doi.org/10.1007/s00234-020-02574-x al. R Luis SanBerkhemer, Olvert A et (2018) Imaging features and safety and efficacy of endovascular stroke treatment: a meta-analysis of individual patient-level data. Lancet Neurol 17:895–904 Yoo AJ, Zaidat OO, Chaudhry ZA, et al (2014) Impact of Pretreatment Noncontrast CT Alberta Stroke Program Early CT Score on Clinical Outcome After Intra-Arterial Stroke Therapy. Stroke 45:746–751. https://doi.org/10.1161/STROKEAHA.113.004260 Kaveeta C, Alhabli I, Bala F, et al (2025) The treatment effect across ASPECTS in acute ischemic stroke: Analysis from the AcT trial. Int J Stroke 20:64–74. https://doi.org/10.1177/17474930241273561 Klemenz A-C, Beckert L, Manzke M, et al (2024) Influence of Deep Learning Based Image Reconstruction on Quantitative Results of Coronary Artery Calcium Scoring. Acad Radiol 31:2259–2267. https://doi.org/10.1016/j.acra.2024.03.020 Pula M, Kucharczyk E, Piersiak M, et al (2025) Improved CTA imaging for stroke evaluation – deep learning and iterative reconstruction comparative study. Neuroradiology. https://doi.org/10.1007/s00234-025-03733-8 Nishikawa M, Machida H, Shimizu Y, et al (2022) Image quality and radiologists’ subjective acceptance using model-based iterative and deep learning reconstructions as adjuncts to ultrahigh-resolution CT in low-dose contrast-enhanced abdominopelvic CT: phantom and clinical pilot studies. Abdom Radiol 47:891–902. https://doi.org/10.1007/s00261-021-03373-5 Altmann S, Abello Mercado MA, Ucar FA, et al (2023) Ultra-High-Resolution CT of the Head and Neck with Deep Learning Reconstruction—Assessment of Image Quality and Radiation Exposure and Intraindividual Comparison with Normal-Resolution CT. Diagnostics 13:1534. https://doi.org/10.3390/diagnostics13091534 Sun J, Li H, Wang B, et al (2021) Application of a deep learning image reconstruction (DLIR) algorithm in head CT imaging for children to improve image quality and lesion detection. BMC Med Imaging 21:108. https://doi.org/10.1186/s12880-021-00637-w Alagic Z, Diaz Cardenas J, Halldorsson K, et al (2022) Deep learning versus iterative image reconstruction algorithm for head CT in trauma. Emerg Radiol 29:339–352. https://doi.org/10.1007/s10140-021-02012-2 Huang X, Zhao W, Wang G, et al (2023) Improving image quality with deep learning image reconstruction in double-low-dose head CT angiography compared with standard dose and adaptive statistical iterative reconstruction. Br J Radiol 96:20220625. https://doi.org/10.1259/bjr.20220625 Austein F, Wodarg F, Jürgensen N, et al (2019) Automated versus manual imaging assessment of early ischemic changes in acute stroke: comparison of two software packages and expert consensus. Eur Radiol 29:6285–6292. https://doi.org/10.1007/s00330-019-06252-2 Neuhaus A, Seyedsaadat SM, Mihal D, et al (2020) Region-specific agreement in ASPECTS estimation between neuroradiologists and e-ASPECTS software. J NeuroInterventional Surg 12:720–724. https://doi.org/10.1136/neurintsurg-2019-015442 Waleed Brinjikji, Mehdi Abbasi, Catherine Arnold, et al (2021) E-ASPECTS software improves interobserver agreement and accuracy of interpretation of aspects score. Interv Neuroradiol 27:781 Löffler MT, Sollmann N, Mönch S, et al (2021) Improved Reliability of Automated ASPECTS Evaluation Using Iterative Model Reconstruction from Head CT Scans. J Neuroimaging 31:341–347. https://doi.org/10.1111/jon.12810 Fatih Seker, Johannes Pfaff, Simon Nagel, et al (2018) CT Reconstruction Levels Affect Automated and Reader-Based ASPECTS Ratings in Acute Ischemic Stroke. J Neuroimaging 29:62–64 Konno M, Otani T, Matsuda M, et al (2025) Deep learning-based computed tomography reconstruction improves image quality but does not significantly affect Alberta stroke program early CT score evaluation in acute middle cerebral artery territory infarction. Neuroradiology. https://doi.org/10.1007/s00234-025-03804-w Van Horn N, Kniep H, Broocks G, et al (2021) ASPECTS Interobserver Agreement of 100 Investigators from the TENSION Study. Clin Neuroradiol 31:1093–1100. https://doi.org/10.1007/s00062-020-00988-x Kaul D, Kahn J, Huizing L, et al (2015) Reducing Radiation Dose in Adult Head CT using Iterative Reconstruction – A Clinical Study in 177 Patients. RöFo - Fortschritte Auf Dem Geb Röntgenstrahlen Bildgeb Verfahr 188:155–162. https://doi.org/10.1055/s-0041-107200 Nagayama Y, Iwashita K, Maruyama N, et al (2023) Deep learning-based reconstruction can improve the image quality of low radiation dose head CT. Eur Radiol 33:3253–3265. https://doi.org/10.1007/s00330-023-09559-3 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 18 Feb, 2026 Read the published version in Neuroradiology → Version 1 posted 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-7904400","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":542764777,"identity":"c18fa72b-4b28-4b08-8b8d-d7981f3201f0","order_by":0,"name":"Estelle Akl","email":"","orcid":"","institution":"University Medical Center Rostock","correspondingAuthor":false,"prefix":"","firstName":"Estelle","middleName":"","lastName":"Akl","suffix":""},{"id":542764778,"identity":"d40ee0c0-88e4-4719-95d8-e5b9c0fc5e5e","order_by":1,"name":"Daniel Cantré","email":"","orcid":"","institution":"University Medical Center Rostock","correspondingAuthor":false,"prefix":"","firstName":"Daniel","middleName":"","lastName":"Cantré","suffix":""},{"id":542764779,"identity":"bdda6fcd-1469-4a47-b898-54e02c940bb2","order_by":2,"name":"Matthias Lütgens","email":"","orcid":"","institution":"University Medical Center Rostock","correspondingAuthor":false,"prefix":"","firstName":"Matthias","middleName":"","lastName":"Lütgens","suffix":""},{"id":542764780,"identity":"f2b2792c-b02c-43d9-a8ee-bd2e0907f169","order_by":3,"name":"Wiebke Hermann","email":"","orcid":"","institution":"University of Rostock","correspondingAuthor":false,"prefix":"","firstName":"Wiebke","middleName":"","lastName":"Hermann","suffix":""},{"id":542764781,"identity":"92ef1d94-12cc-45d4-bca7-ffe50ddb50ab","order_by":4,"name":"Sönke Langner","email":"","orcid":"","institution":"University Medical Center Rostock","correspondingAuthor":false,"prefix":"","firstName":"Sönke","middleName":"","lastName":"Langner","suffix":""},{"id":542764782,"identity":"b345e26b-4e2f-436e-a6ee-6f3a646094ef","order_by":5,"name":"Marc-André Weber","email":"","orcid":"","institution":"University Medical Center Rostock","correspondingAuthor":false,"prefix":"","firstName":"Marc-André","middleName":"","lastName":"Weber","suffix":""},{"id":542764783,"identity":"c58ad510-ef61-423a-ab24-096af771511a","order_by":6,"name":"Ann-Christin Klemenz","email":"","orcid":"","institution":"University Medical Center Rostock","correspondingAuthor":false,"prefix":"","firstName":"Ann-Christin","middleName":"","lastName":"Klemenz","suffix":""},{"id":542764784,"identity":"d1dcdecc-aca1-4fe1-9cd0-aa7b4bcd37e6","order_by":7,"name":"Felix G. Meinel","email":"","orcid":"","institution":"University Medical Center Rostock","correspondingAuthor":false,"prefix":"","firstName":"Felix","middleName":"G.","lastName":"Meinel","suffix":""},{"id":542764785,"identity":"99d60001-44ee-4f8f-a0be-ee4d89bdcc9b","order_by":8,"name":"Ebba Beller","email":"data:image/png;base64,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","orcid":"","institution":"University Medical Center Rostock","correspondingAuthor":true,"prefix":"","firstName":"Ebba","middleName":"","lastName":"Beller","suffix":""}],"badges":[],"createdAt":"2025-10-20 09:53:03","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7904400/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7904400/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s00234-026-03938-5","type":"published","date":"2026-02-18T15:59:13+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":95806495,"identity":"53e884c1-d059-4ccf-8eeb-b23d3a819d7d","added_by":"auto","created_at":"2025-11-13 08:47:32","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1013748,"visible":true,"origin":"","legend":"","description":"","filename":"anonymizedASPECTManuscriptNeuroradiology.docx","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/d5f31795b110bb9eaf80373d.docx"},{"id":95806094,"identity":"62cba1f6-97a6-482b-94c6-402ce8ac8260","added_by":"auto","created_at":"2025-11-13 08:47:15","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9823,"visible":true,"origin":"","legend":"","description":"","filename":"3c12180ccf8546608a5fea237a12ef37.json","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/9374d380de1d1be1c84dc50e.json"},{"id":95818947,"identity":"6342432f-998d-47e4-aec4-394e1954d2de","added_by":"auto","created_at":"2025-11-13 10:35:58","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":89212,"visible":true,"origin":"","legend":"","description":"","filename":"3c12180ccf8546608a5fea237a12ef371enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/d4a025f13161d2fee9ca8c8f.xml"},{"id":95806487,"identity":"baa6e85e-591b-4c54-b61f-1db502cfbedb","added_by":"auto","created_at":"2025-11-13 08:47:32","extension":"jpeg","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":39768,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/155d42d1e923313286719ccb.jpeg"},{"id":95806218,"identity":"e9b1e5af-bf3e-47e9-a387-79db799159de","added_by":"auto","created_at":"2025-11-13 08:47:20","extension":"jpeg","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":269408,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/7a30391d29327eee066aa60c.jpeg"},{"id":95806375,"identity":"39c03081-e0cf-4744-8a93-466860424512","added_by":"auto","created_at":"2025-11-13 08:47:26","extension":"jpeg","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":298606,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/45a3e8cf754ad3886690dec0.jpeg"},{"id":95806072,"identity":"d9bad95d-1e6d-4277-a25e-5c656f95662f","added_by":"auto","created_at":"2025-11-13 08:47:14","extension":"emf","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":104276,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.emf","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/c7324cc4940d9c53229aa779.emf"},{"id":95806282,"identity":"6d9d562e-1415-44dc-88aa-a5621a99ebe0","added_by":"auto","created_at":"2025-11-13 08:47:22","extension":"emf","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":95876,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.emf","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/44206dc589b57a34d3b40db4.emf"},{"id":95806326,"identity":"b19afb74-ff4f-48eb-84f6-e0f0bf9c27c1","added_by":"auto","created_at":"2025-11-13 08:47:23","extension":"jpeg","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":116536,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/3b85480bd3d662237835449b.jpeg"},{"id":95806492,"identity":"ecf5993d-6ec6-4761-b04c-ceed323ca756","added_by":"auto","created_at":"2025-11-13 08:47:32","extension":"jpeg","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":132155,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/5918c4459df75901f6f05fa6.jpeg"},{"id":95806210,"identity":"5a922774-eacc-4dda-a6f8-3b439fb60ee1","added_by":"auto","created_at":"2025-11-13 08:47:20","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":96414,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/3f1d5b9d2e7493eefac4f5ad.png"},{"id":95806119,"identity":"f5eda97d-ebe2-4163-b52f-7c878bcae060","added_by":"auto","created_at":"2025-11-13 08:47:16","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":217247,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/4ae18a87122838f835b85a0c.png"},{"id":95806165,"identity":"c42dccf3-f851-4506-b6f6-f4e6fdb485ad","added_by":"auto","created_at":"2025-11-13 08:47:18","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":136994,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/15e2465058027f3336d0edaa.png"},{"id":95806293,"identity":"da961d80-e49a-4d7f-bbe3-eb06d8652ee6","added_by":"auto","created_at":"2025-11-13 08:47:22","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4606,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/b3ac9019e027ad7ec2344748.png"},{"id":95806290,"identity":"c3474085-fc5c-4c61-9644-c9a2c006acf6","added_by":"auto","created_at":"2025-11-13 08:47:22","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":12070,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/19965bedb87612628f4ddcaa.png"},{"id":95806428,"identity":"0107b76d-2fae-4e6b-b73b-c830ff53055a","added_by":"auto","created_at":"2025-11-13 08:47:29","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":75255,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/309634d77be5b9796575c7b0.png"},{"id":95806249,"identity":"4678c590-bbfb-4cd3-83d9-3bec8d29865f","added_by":"auto","created_at":"2025-11-13 08:47:21","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":91915,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/f9e14f24bad797ff96388649.png"},{"id":95806557,"identity":"2ada79f6-480f-42e7-a251-808f053ee24c","added_by":"auto","created_at":"2025-11-13 08:47:36","extension":"xml","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":89048,"visible":true,"origin":"","legend":"","description":"","filename":"3c12180ccf8546608a5fea237a12ef371structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/69ee41cbf19b12a7c7f3dcee.xml"},{"id":95806256,"identity":"1a2e12b8-a57c-408f-9eb3-43706cb1ebdb","added_by":"auto","created_at":"2025-11-13 08:47:21","extension":"html","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":96752,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/0c4c6bb1e5c8811d42a01b3a.html"},{"id":95806175,"identity":"d9ba439e-52a2-4aca-9ff2-93b761afb994","added_by":"auto","created_at":"2025-11-13 08:47:18","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":39768,"visible":true,"origin":"","legend":"\u003cp\u003eAxial non-contrast CT scan with hyperdense MCA sign on the right side of a 91-year-old patient presenting with left-sided hemiplegia, ipsilateral horizontal gaze palsy (to the left) with déviation conjugée and anarthria. The patient was excluded from the study due to severe artifacts caused by a cochlear implant in the right temporal region\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/4cc2818ffd629a81bc94e567.jpeg"},{"id":95806502,"identity":"8a23a924-adb4-43b1-98c1-fee7f7695b31","added_by":"auto","created_at":"2025-11-13 08:47:33","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":269408,"visible":true,"origin":"","legend":"\u003cp\u003eAxial non-contrast CT images demonstrating early ischemic changes in the left temporal region with a hyperdense middle cerebral artery sign. Images were reconstructed using six different algorithms: Filtered Back Projection (FBP) (a), Adaptive Statistical Iterative Reconstruction-V (ASIR-V) at 30% (b) and 60% (c), and Deep Learning Image Reconstruction (DLIR) at low (L) (d), medium (M) (e), and high (H) (f) strengths\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/afbc046f3b87307472c47efe.jpeg"},{"id":95806379,"identity":"533d5eec-e3f4-4baa-b99d-5dd35574595e","added_by":"auto","created_at":"2025-11-13 08:47:26","extension":"jpeg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":298606,"visible":true,"origin":"","legend":"\u003cp\u003eAutomated ASPECTS analysis of a NCCT of a 51-year-old male patient with right-sided paresis and global aphasia, 1.5 hours prior to imaging. ASPECTS was calculated for each different reconstruction technique: Filtered Back Projection (FBP) (a); Adaptive Statistical Iterative Reconstruction-V (ASIR-V) at 30% (b) and 60% (c), and Deep Learning Image Reconstruction (DLIR) at low (L) (d), medium (M) (e), and high (H) (f) strength. Areas with presumed early ischemic changes were automatically contoured in red and automated ASPECTS displayed on the right with a score of 4 for FBP, as well as ASIR-V at 30% and a score 5 for ASIR-V at 60% and DLIR-L, -M, as well as -H\u003c/p\u003e","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/8cd54e99f32a9b1c92a4dbcb.jpeg"},{"id":95806280,"identity":"848f4a72-c610-4f24-85b3-f82420fdd8dc","added_by":"auto","created_at":"2025-11-13 08:47:22","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":81907,"visible":true,"origin":"","legend":"\u003cp\u003eDistribution of ASPECTS categories by reconstruction technique: ASIR-V, adaptive statistical iterative reconstruction-V; FBP, filtered back projection; DLIR, deep learning- based image reconstruction in low (L), medium (M) or high (H) strength.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/8104c30494e0b66d0e3183c4.png"},{"id":95806505,"identity":"c335e9f3-7966-4d10-8a7e-02eb54266477","added_by":"auto","created_at":"2025-11-13 08:47:33","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":40962,"visible":true,"origin":"","legend":"\u003cp\u003eOver- and underestimation of ischemic change per ASPECTS region as the sum of all reconstruction techniques in comparison to FBP (upper row) and of each reconstruction technique compared to FBP (lower row). ASIR-V, adaptive statistical iterative reconstruction-V; FBP, filtered back projection; DLIR, deep learning-based image reconstruction in low (L), medium (M) or high (H) strength; C, caudate nucleus; L, lentiform nucleus; IC, internal capsule; I, insular ribbon; M1-6, six cortical areas of the middle cerebral artery territory.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/cfb3b7670d104e5cc9df13bd.png"},{"id":95806224,"identity":"81a7f5b5-7e02-41bc-b06c-7ed263a248da","added_by":"auto","created_at":"2025-11-13 08:47:20","extension":"jpeg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":116536,"visible":true,"origin":"","legend":"\u003cp\u003eRepresentative example of overestimation of the M3 region with automated ASPECTS comparing Filtered Back Projection (FBP) (a) with Deep Learning Image Reconstruction at high strength (DLIR-H) (b). With FBP (a), no early ischemic changes were identified, resulting in an ASPECTS of 10, whereas with DLIR-H (b), the M3 cortex area of the left hemisphere was marked as early ischemic region (in red), resulting in an ASPECTS of 9\u003c/p\u003e","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/640e89f16ed5341a6e4cf98a.jpeg"},{"id":95806441,"identity":"ee7fc02a-94f4-42b7-92d6-b550653ea35c","added_by":"auto","created_at":"2025-11-13 08:47:30","extension":"jpeg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":132155,"visible":true,"origin":"","legend":"\u003cp\u003eRepresentative example of underestimation of the insular ribbon with automated ASPECTS comparing Filtered Back Projection (FBP) (a) with Deep Learning Image Reconstruction at high strength (DLIR-H) (b). With FBP (a), early ischemic changes were identified in the insula (I) as well as in the M2-region of the left middle cerebral artery territory (in red), resulting in an automated ASPECTS of 8, whereas with DLIR-H (b), only the M2-region was marked as affected, resulting in an ASPECTS of 9\u003c/p\u003e","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/75417253c4b28b7b69e1d49a.jpeg"},{"id":103251191,"identity":"08309318-d2d6-48bd-b73d-958b4451566f","added_by":"auto","created_at":"2026-02-23 16:05:50","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1665315,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7904400/v1/d66bd0d0-e1e0-4151-8638-55618ac719eb.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Influence of deep learning image reconstruction and adaptive statistical iterative reconstruction-V on automated Alberta Stroke Program Early CT Score- evaluation","fulltext":[{"header":"Key Points","content":"\u003col start=\"1\" type=\"1\"\u003e\n \u003cli\u003eCT Reconstruction techniques influence automated ASPECT scoring reliability.\u003c/li\u003e\n \u003cli\u003eInsula, as well as the cortical M3 ASPECTS region, seemed more susceptible to algorithmic variability.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eMore patients were overestimated from ASPECTS \u0026le; 5 to ASPECTS \u0026ge; 6, making them available for endovascular therapy.\u003c/li\u003e\n \u003cli\u003eDLIR-M demonstrated the highest agreement between automated ASPECTS assessment and expert-derived ground truth.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Background","content":"\u003cp\u003eThe Alberta Stroke Program Early Computed Tomography Score (ASPECTS) is widely used to evaluate early ischemic changes in patients with acute ischemic stroke, offering a standardized approach for estimating the extent of the ischemic region [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. In recent guidelines, an ASPECTS score of 6 or higher is a requirement or criterion for selecting patients who should receive endovascular therapy [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. ASPECTS is also used for the evaluation of stroke severity and prognosis in acute ischemic stroke [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Traditional manual ASPECT scoring, however, can be subject to considerable variability among observers. To address these limitations, automated ASPECTS tools based on artificial intelligence (AI), including deep learning (DL) and convolutional neural networks, have emerged, capable of complex tasks such as feature extraction, classification, and segmentation in medical imaging [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThe integration of AI into stroke imaging has also transformed other aspects of the diagnostic workflow [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Historically, Filtered Back Projection (FBP) was the primary method for CT image reconstruction due to its computational efficiency. However, FBP has well-documented limitations, including its inability to effectively manage noise, particularly at lower radiation doses or with obese patients. Iterative Reconstruction (IR) methods were subsequently introduced, offering improved noise management and dose reduction. More recently, Deep Learning Image Reconstruction (DLIR) has emerged as a powerful alternative to FBP and IR, enhancing noise reduction and improving image quality, while maintaining an acceptable radiation dose. DLIR utilizes trained artificial neural networks to refine images and improve diagnostic confidence, particularly in challenging imaging conditions. Recent studies have demonstrated that DLIR-based methods outperform traditional IR by producing images with reduced noise and superior contrast-to-noise ratios, thereby enhancing the diagnostic potential for various pathologies [\u003cspan additionalcitationids=\"CR10\" citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIn this study, we aim to evaluate the impact of different reconstruction algorithms for CT on the performance of automated ASPECT scoring in patients with acute ischemic stroke. Understanding the differences among these reconstruction algorithms could contribute to optimizing imaging protocols and ultimately improving clinical outcomes for stroke patients.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cem\u003ePatient selection and study design\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThis study was designed as a retrospective, single-center study. Patients were identified by a retrospective search of our radiology information system (Centricity 5.0, GE Healthcare, Barrington, Illinois). Inclusion criteria were patients who received a non-contrast head CT (NCCT) and additional CT angiography (CTA) at our institution by suspicion of an acute stroke in the middle cerebral artery (MCA) territory between May 13, 2024, and November 19, 2024 (n=490). We excluded patients with other pathologies (e.g. tumor, trauma, headache\u0026hellip;), (n=159), symptoms suggesting an involvement of the anterior cerebral artery territory and posterior circulation (n=64), hemorrhagic strokes (n=39), large, chronic, ischemic areas in the MCA territory (n= 15) and more than 24 hours after the onset of MCA stroke symptoms (n=8). Twenty patients were excluded due to technical issues with the automated ASPECTS software, and 12 patients were excluded because their datasets were incomplete. One patient was excluded due to artifacts caused by a cochlear implant (Fig. 1).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eHuman Ethics and Consent to Participate declarations\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe study protocol was approved by the Ethics Committee of Rostock University Medical Faculty and informed consent to participate was waived due to the retrospective nature of this study. The study was conducted in compliance with the Declaration of Helsinki in its current form. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eCT acquisition protocol\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eAll included patients underwent CT imaging of the head as per the standard acute stroke protocol, which was conducted using a 128-slice CT scanner (Revolution CT, GE Healthcare). Scanning parameters included a slice thickness of 0,625 mm, reconstruction of 3,0 mm, voltage of 120 KV, and a current using a tube current modulation, determined based on the Scout in mA. The median CT dose index was 47 mGy (IQR 46-52 mGy).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eImage Reconstruction and Analysis\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eFor all patients, NCCT images of the brain were reconstructed on the Revolution CT scanner (software version 2.1B) using standard Filtered Back Projection (FBP), a new generation of adaptive statistical iterative reconstruction (ASIR-V at 30% and 60% strength), as well as three strength levels of deep-learning-based reconstruction (low (DLIR_L), medium (DLIR_M), and high (DLIR_H); \u0026ldquo;TrueFidelityTM\u0026rdquo;, GE Healthcare, Chicago) (Figure 2). The image slice thickness used for image reconstruction, automated ASPECT score determination, was reconstructed using the standard convolution kernel and a slice thickness of 2.5 mm, as recommended by the manufacturer for optimal ASPECT score evaluation. The Alberta Stroke Program Early CT Score (ASPECTS) was obtained automatically using commercially available software (AW server 3.2 software, GE HealthCare). The scoring system examines ten specific regions within the middle cerebral artery (MCA) territory: the caudate nucleus (C), lentiform nucleus (L), internal capsule (IC), insular ribbon (I), as well as the cortical regions M1, M2, and M3 (anterior, middle, and posterior third of the lower MCA territory, respectively), and M4, M5, and M6 (anterior, middle, and posterior third of the higher MCA territory, respectively). Each region demonstrating signs of early ischemic changes is deducted from the total of ten, resulting in a final ASPECTS ranging from 0 (all areas affected by early ischemic changes) to 10 (no changes) [7]. Ground truth was established by expert consensus between one radiology and one neuroradiology consultant (with 11 and 12 years of experience in stroke imaging) by using all available imaging at the acute stage, including NCCT with FBP, ASIR, and DLIR technique, CT angiography, CT perfusion, digital subtraction angiography, and follow-up CT or MRI. They were both blinded to the results of the automated ASPECTS rating.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003c/em\u003e\u003cem\u003eStatistical Analysis\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eStatistical analysis was performed using GraphPad Prism version 9.0.0 (GraphPad Software LLC). FBP was selected as the reference for comparison with ASIR and DLIR. Bland-Altman plots were used to determine the mean difference and limits of agreement between FBP and the other reconstruction techniques [12\u0026ndash;14]. The agreement between inconsistent automated ASPECTS ratings of FBP, ASIR, and DLIR reconstructions and the ground truth was assessed using weighted kappa analysis.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\u003ch2\u003ePatient Characteristics\u003c/h2\u003e\u003cp\u003eThe final study population included 173 participants with suspected MCA infarction and symptom onset within 24 hours. The median time from stroke onset to CT imaging was 3 hours (IQR 1.5-6 hours). They had a median age of 79 years (IQR 71\u0026ndash;86 years) and 39% (n\u0026thinsp;=\u0026thinsp;68) were female. In 117 (68%) patients, the results of the automated ASPECTS did not differ despite the use of different CT reconstruction techniques. In 56 patients (32%), one or more score results of the automated ASPECTS differed between the different CT reconstruction techniques. The patients ' characteristics are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\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\u003ePatient Characteristics\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"2\"\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\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCharacteristics\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePatients included\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e173\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMedian age in years (IQR)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e79 (71\u0026ndash;86)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFemale n (%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e68 (39%)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMedian time from stroke onset\u003c/p\u003e\u003cp\u003eto CT imaging in hours (IQR)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2.5 (1.5-6)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eInconsistency of\u003c/p\u003e\u003cp\u003eautomated ASPECTS\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e56 (32%)\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\u003eASPECTS, Alberta Stroke Program Early CT Score.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eAgreement of total ASPECTS between FBP and ASIR-V or DLIR\u003c/h3\u003e\n\u003cp\u003eExamples of ASPECTS analysis of different CT reconstruction techniques in a representative patient are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. Bland-Altman plots of all reconstruction techniques (ASIR-V at 30% and 60%, DLIR-L, DLIR-M, and DLIR-H) compared to FBP as the reference standard were obtained. ASIR-V 30% showed the best agreement regarding ASPECTS compared to FBP, with a mean bias of \u0026minus;\u0026thinsp;0.057, whereas ASIR-V 60% had the highest deviation in scoring results, with a mean bias of \u0026minus;\u0026thinsp;0.126. The mean bias for DLIR-L, DLIR-M, and DLIR-H was \u0026ndash; 0.069, \u0026minus; 0.121, and \u0026ndash; 0.086 points, respectively. Detailed results of the Bland\u0026ndash;Altman analysis are shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\u003cp\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\u003eBland\u0026ndash;Altman Analysis for ASPECTS of FBP vs. ASIR-V and DLIR\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\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e\u003cp\u003e95% Limits of Agreement\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eComparison\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMean Bias\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSD of Bias\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eUpper\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eLower\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFBP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eReference\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eASIR-V 30%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-0.057\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.489\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-1.017\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.902\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eASIR-V 60%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-0.126\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.605\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-1.311\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.058\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDLIR-L\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-0.069\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.632\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-1.308\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.170\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDLIR-M\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-0.121\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.682\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-1.457\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.216\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDLIR-H\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-0.086\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.767\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-1.589\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1.417\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\u003eFBP serves as the reference standard. ASIR-V, adaptive statistical iterative reconstruction-V; DLIR, deep learning-based image reconstruction in low (L), medium (M), or high (H) strength; FBP, filtered back projection.\u003c/p\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eImpact on the Classification of Patients into Risk Categories Based on ASPECTS\u003c/h2\u003e\u003cp\u003eReclassifications of ASPECTS categories with FBP as a reference standard were found for each reconstruction technique. Most of the re-classifications were from moderate (score 5\u0026ndash;7) to high ASPECTS (score 8\u0026ndash;10). With ASIR-V at 30% and 60% strength, three and five patients were re-classified from moderate to high ASPECTS. DLIR at low, medium, and high strength resulted in seven, five, and ten patients, respectively, being re-classified from moderate to high ASPECTS. However, patients were also re-classified from a high to moderate ASPECTS with one and two patients regarding ASIR-V at 30% and 60% strength, as well as four, two, and three patients regarding DLIR at low, medium, and high strength, respectively. With ASIR-V at 60% and DLIR at medium and high strength, one case was reclassified from a low (score 0\u0026ndash;4) to a medium ASPECTS (score 5\u0026ndash;7). With DLIR at low strength, two patients were reclassified from a low to a medium ASPECTS, and with ASIR-V at 30% strength, one patient was reclassified from a medium to a low ASPECTS. With ASIR-V at 30% and 60% strength, one patient was re-classified from ASPECTS\u0026thinsp;\u0026le;\u0026thinsp;5 to ASPECTS\u0026thinsp;\u0026ge;\u0026thinsp;6, as well as three, five, and four patients with DLIR at low, medium, and high strength. With ASIR-V at 30% and DLIR at high strength one patient was re-classified from ASPECTS\u0026thinsp;\u0026ge;\u0026thinsp;6 to ASPECTS\u0026thinsp;\u0026le;\u0026thinsp;5. Detailed classification of the patient cohort within different reconstruction techniques can be found in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e and the reclassification is displayed in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e\u003cp\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\u003eChange in ASPECTS Categories comparing FBP to ASIR-V and DLIR.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"11\"\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\u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eASPECTS category\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"10\" nameend=\"c11\" namest=\"c2\"\u003e\u003cp\u003eNet Change in Number of Patients in Risk Groups\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e\u003cp\u003eASIR-V 30%\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e\u003cp\u003eASIR-V 60%\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c7\" namest=\"c6\"\u003e\u003cp\u003eDLIR-L\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e\u003cp\u003eDLIR-M\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c11\" namest=\"c10\"\u003e\u003cp\u003eDLIR-H\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ehigh (8\u0026ndash;10)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e+\u0026thinsp;3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e+\u0026thinsp;5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e+\u0026thinsp;7\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e-2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e+\u0026thinsp;5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c10\"\u003e\u003cp\u003e-3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c11\"\u003e\u003cp\u003e+\u0026thinsp;10\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003emoderate (5\u0026ndash;7)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e+\u0026thinsp;1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e+\u0026thinsp;3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-7\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e+\u0026thinsp;6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e-5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e+\u0026thinsp;3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c10\"\u003e\u003cp\u003e-10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c11\"\u003e\u003cp\u003e+\u0026thinsp;4\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003elow (0\u0026ndash;4)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e+\u0026thinsp;1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e-1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c10\"\u003e\u003cp\u003e-1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c11\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u0026ge;\u0026thinsp;6\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e+\u0026thinsp;1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e+\u0026thinsp;1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e+\u0026thinsp;3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e+\u0026thinsp;5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c10\"\u003e\u003cp\u003e-1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c11\"\u003e\u003cp\u003e+\u0026thinsp;4\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u0026le;\u0026thinsp;5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e+\u0026thinsp;1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e\u003cp\u003e-5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c9\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c10\"\u003e\u003cp\u003e-4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c11\"\u003e\u003cp\u003e+\u0026thinsp;1\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\u003eASIR-V, adaptive statistical iterative reconstruction-V; DLIR, deep learning-based image reconstruction in low (L), medium (M), or high (H) strength; FBP, filtered back projection. FBP served as the reference standard.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003eImpact on Regional Distribution of Ischemic Changes Based on ASPECTS\u003c/h2\u003e\u003cp\u003eOver- and underestimation of ischemic change per ASPECTS region of all reconstruction techniques (ASIR-V at 30% and 60%, DLIR-L, DLIR-M, and DLIR-H) compared to FBP as the reference standard is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e. The region with the highest overestimation, summarized from all reconstruction techniques in comparison to FBP was M3 (n\u0026thinsp;=\u0026thinsp;23) (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e), followed by M2 (n\u0026thinsp;=\u0026thinsp;18), insula (n\u0026thinsp;=\u0026thinsp;11), M5 (n\u0026thinsp;=\u0026thinsp;9), M1 (n\u0026thinsp;=\u0026thinsp;6), M6 (n\u0026thinsp;=\u0026thinsp;5), lentiform nucleus as well as caudate nucleus (n\u0026thinsp;=\u0026thinsp;3) and M4 (n\u0026thinsp;=\u0026thinsp;3). No overestimation was found for the internal capsule (n\u0026thinsp;=\u0026thinsp;0). The region with the highest underestimation compared to FBP was the insular ribbon (n\u0026thinsp;=\u0026thinsp;51) (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e), followed by M4 (n\u0026thinsp;=\u0026thinsp;29), M2 (n\u0026thinsp;=\u0026thinsp;23), M5 (n\u0026thinsp;=\u0026thinsp;16), M1 (n\u0026thinsp;=\u0026thinsp;14), M1 (n\u0026thinsp;=\u0026thinsp;11), M3 (n\u0026thinsp;=\u0026thinsp;10), lentiform nucleus (n\u0026thinsp;=\u0026thinsp;8) followed by caudate nucleus as well as the internal capsule (n\u0026thinsp;=\u0026thinsp;1). With regards to the reconstruction techniques, the highest number of overestimation of ASPECTS regions, compared to FBP, was found with DLIR-H (n\u0026thinsp;=\u0026thinsp;28), followed by DLIR-L (n\u0026thinsp;=\u0026thinsp;19), DLIR-M (n\u0026thinsp;=\u0026thinsp;17), ASIR-60% (=\u0026thinsp;n\u0026thinsp;=\u0026thinsp;10), and ASIR-30% (n\u0026thinsp;=\u0026thinsp;9). The highest number of underestimation of ASPECTS regions, compared to FBP, was found with DLIR-H (n\u0026thinsp;=\u0026thinsp;43), followed by DLIR-M (n\u0026thinsp;=\u0026thinsp;38), DLIR-L (n\u0026thinsp;=\u0026thinsp;33), ASIR-60% (n\u0026thinsp;=\u0026thinsp;31) and ASIR-30% (n\u0026thinsp;=\u0026thinsp;19).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003eAgreement between consensus and automated ASPECTS of FBP, ASIR, and DLIR\u003c/h2\u003e\u003cp\u003eIn 56 patients (32%), inconsistent results of the automated ASPECTS were found between different CT reconstruction techniques and were compared to expert consensus. DLIR-M showed the highest agreement with expert consensus (κ\u0026thinsp;=\u0026thinsp;0.352), followed by ASIR-60% (κ\u0026thinsp;=\u0026thinsp;0.346), ASIR-30% (κ\u0026thinsp;=\u0026thinsp;0.291), DLIR-L (κ\u0026thinsp;=\u0026thinsp;0.263), FBP (κ\u0026thinsp;=\u0026thinsp;0.229), and DLIR-H (κ\u0026thinsp;=\u0026thinsp;0.220).\u003c/p\u003e\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this study, we investigated the effect of CT reconstruction techniques on the automated ASPECTS in patients with suspected MCA infarction. In total, ASIR-V and DLIR reconstructions showed only a minor underestimation of ASPECTS results compared to FBP. Regarding the ASPECTS\u0026thinsp;\u0026lt;\u0026thinsp;6 threshold, however, more patients were overestimated when transitioning from ASPECTS\u0026thinsp;\u0026le;\u0026thinsp;5 to ASPECTS\u0026thinsp;\u0026ge;\u0026thinsp;6, making them eligible for endovascular therapy, which was more pronounced for DLIR, especially DLIR-M, compared to ASIR-V. Interestingly, the ASPECTS region with the highest overestimation compared to FBP was M3, and the region with the highest underestimation was the insular ribbon. The highest number of over- and underestimation of ASPECTS regions was found with DLIR-H. Comparing the inconsistent results of the automated ASPECTS between different CT reconstruction techniques and expert consensus as the ground truth, DLIR-M showed the highest agreement.\u003c/p\u003e\u003cp\u003eOver the last decade, iterative reconstruction (IR) algorithms have significantly advanced CT technology as a potential alternative to FBP [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. These methods improve image quality and reduce radiation dose, making them a criterion standard in CT imaging [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan additionalcitationids=\"CR16\" citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Unlike IR and FBP, DLIR utilizes artificial neural networks to iteratively learn noise elimination while preserving anatomical details, therefore representing another step forward in both image quality and dose optimization [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. All advancements in CT reconstruction techniques have reshaped the landscape of neuroradiological CT imaging [\u003cspan additionalcitationids=\"CR19 CR20\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e] and may therefore also hold significant potential to refine automated ASPECTS, a crucial tool in the evaluation of acute ischemic stroke.\u003c/p\u003e\u003cp\u003eSimilar studies have primarily focused on the agreement between ASPECTS scores obtained by automated software and those determined by human readers, which have shown overall good [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] or even an improvement in the interobserver agreement and accuracy of neuroradiologists and neurologists in their interpretation of ASPECTS [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. L\u0026ouml;ffler et al. compared ASPECTS assessments, which were based on hybrid IR or iterative model reconstruction (IMR) and carried out by humans and software in comparison to a consensus reference. They reported that the automated software improved with IMR, which can further reduce noise levels and improve image contrasts at a given dose level and slice thickness compared to hybrid IR [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Seker et al. NECT compared automated ASPECTS of FBP with Sinogram-Affirmed Iterative Reconstruction (SAFIRE) in strength levels 2 and 4 out of 5. This study included 43 patients with acute occlusion of the M1 segment of the middle cerebral artery and found that automated ASPECTS of SAFIRE, at a strength level of 2, showed the highest consistency when compared to a ground truth provided by an expert with unrestricted data access [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. However, in both studies mentioned above, different reconstruction techniques were investigated in comparison to our study, and a detailed analysis of the ASPECTS categories and regions was not performed. Konno et al. compared similar CT reconstruction techniques to those in our study (FBP, ASIR-V at 50%, and DLIR), which did not lead to significant differences in the ASPECTS evaluation of 30 patients. However, only human readers were used for ASPECTS assessment and no automated software [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIn our study, the highest number of over- and underestimation of ASPECTS regions was observed with DLIR-H. The insula was the most frequently underestimated ASPECTS region, while overestimation was most notable for the cortical M3 ASPECTS region, suggesting that both regions are more susceptible to algorithmic variability. On the other hand, subcortical regions, such as the internal capsule, lentiform nucleus, and caudate nucleus, displayed greater consistency across reconstruction techniques, suggesting these regions are less influenced by algorithmic variability. Austein et al. also found disagreement in the characteristics of ASPECTS regions scored, which was most notable for the insula and internal capsule, when comparing between humans and two different automated ASPECTS packages. They also reported variability in cortical scoring (M1\u0026ndash;M6), which they concluded might be due to the challenges of consistently defining the anatomical borders of these regions [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Interestingly, van Horn et al, who evaluated interrater reliability in a large group of human readers, found that the ASPECTS region with the highest level of agreement was the insular cortex, and with the lowest level of agreement, the M3 region [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThe most important clinical use of ASPECTS is in guiding decisions regarding the eligibility for thrombolysis and endovascular therapy [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Variability in ASPECTS due to reconstruction technique, such as over- or underestimation of the total score, may significantly influence patient outcomes by altering the choice of therapy. According to established guidelines, an ASPECTS score of 6 or higher is a requirement or criterion for endovascular therapy [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. In our study, most patients were reclassified from ASPECTS\u0026thinsp;\u0026le;\u0026thinsp;5 to ASPECTS\u0026thinsp;\u0026ge;\u0026thinsp;6 with DLIR-M, making them eligible for endovascular therapy. Additionally, DLIR-M showed the highest agreement between the consensus expert ground truth and the automated ASPECTS.\u003c/p\u003e\u003cp\u003eOur study has several limitations. The evaluation was performed using reconstruction algorithms from a single CT manufacturer, which may limit the generalizability of the findings to other systems. Secondly, while FBP was employed as the reference standard, it is not a definitive method for ASPECTS due to the lack of a universally accepted gold standard. Furthermore, dose reduction was not incorporated into the CT acquisition protocol, although deep learning-based reconstruction and iterative reconstruction techniques have previously shown potential for dose reductions in head CT [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. This was due to the retrospective character of this study and ethical considerations, since a correct ASPECTS has therapeutic consequences and may outweigh radiation protection in the acute emergency setting. Additionally, due to the retrospective nature of this study, simultaneous diffusion-weighted MRI, which could likely provide the most accurate ground truth in cases of acute stroke, was not feasible for every patient in the present study. We attempted to compensate for this by establishing an expert-based ground truth for comparing automated and reader-based ASPECT, with access to clinical information, baseline CT, CT angiography, and CT perfusion imaging, as well as follow-up imaging.\u003c/p\u003e\u003cp\u003eIn conclusion, different CT reconstruction techniques resulted in only a minor underestimation of automated ASPECTS compared to FBP. Especially the insula, as well as the cortical M3 ASPECTS, seemed more susceptible to algorithmic variability. However, more patients were overestimated from ASPECTS\u0026thinsp;\u0026le;\u0026thinsp;5 to ASPECTS\u0026thinsp;\u0026ge;\u0026thinsp;6, making them available for endovascular therapy, which was most pronounced for DLIR-M. DLIR-M also exhibited the highest agreement with consensus expert ground truth for automated ASPECTS. Careful selection of reconstruction parameters, as well as further optimization and standardization of these techniques, is therefore essential for broader application in stroke imaging to minimize discrepancies and improve consistent and reliable patient management.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eE.A., E.B. and M.L. collected data. E.A. and E.B. wrote the main manuscript text and prepared all figures. D.C. and E.B. analyzed the data as part of the \"expert consensus\". F.M., A-C. K. and E.B. performed the statistics. W.H. and S.L. added neurological clinical aspects. All authors reviewed the manuscript.\u003c/p\u003e\u003ch2\u003eStatements and Declarations:\u003c/h2\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003ch2\u003eHuman Ethics and Consent to Participate declarations:\u003c/h2\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eUchida K, Shindo S, Yoshimura S, et al (2022) Association Between Alberta Stroke Program Early Computed Tomography Score and Efficacy and Safety Outcomes With Endovascular Therapy in Patients With Stroke From Large-Vessel Occlusion: A Secondary Analysis of the Recovery by Endovascular Salvage for Cerebral Ultra-acute Embolism\u0026mdash;Japan Large Ischemic Core Trial (RESCUE-Japan LIMIT). JAMA Neurol 79:1260. https://doi.org/10.1001/jamaneurol.2022.3285\u003c/li\u003e\n\u003cli\u003eBoulanger J, Lindsay M, Gubitz G, et al (2018) Canadian Stroke Best Practice Recommendations for Acute Stroke Management: \u003cem\u003ePrehospital, Emergency Department, and Acute Inpatient Stroke Care, 6th Edition, Update 2018\u003c/em\u003e. Int J Stroke 13:949\u0026ndash;984. https://doi.org/10.1177/1747493018786616\u003c/li\u003e\n\u003cli\u003ePowers WJ, Rabinstein AA, Ackerson T, et al (2018) 2018 Guidelines for the Early Management of Patients With Acute Ischemic Stroke: A Guideline for Healthcare Professionals From the American Heart Association/American Stroke Association. Stroke 49:. https://doi.org/10.1161/STR.0000000000000158\u003c/li\u003e\n\u003cli\u003eEsmael A, Elsherief M, Eltoukhy K (2021) Predictive Value of the Alberta Stroke Program Early CT Score (ASPECTS) in the Outcome of the Acute Ischemic Stroke and Its Correlation with Stroke Subtypes, NIHSS, and Cognitive Impairment. Stroke Res Treat 2021:1\u0026ndash;10. https://doi.org/10.1155/2021/5935170\u003c/li\u003e\n\u003cli\u003eElawady SS, Saway BF, Matsukawa H, et al (2024) Thrombectomy in Stroke Patients With Low Alberta Stroke Program Early Computed Tomography Score: Is Modified Thrombolysis in Cerebral Infarction (mTICI) 2c/3 Superior to mTICI 2b? J Stroke 26:95\u0026ndash;103. https://doi.org/10.5853/jos.2023.02292\u003c/li\u003e\n\u003cli\u003eSoun JE, Chow DS, Nagamine M, et al (2021) Artificial Intelligence and Acute Stroke Imaging. Am J Neuroradiol 42:2\u0026ndash;11. https://doi.org/10.3174/ajnr.A6883\u003c/li\u003e\n\u003cli\u003eMaegerlein C, Fischer J, M\u0026ouml;nch S, et al (2019) Automated Calculation of the Alberta Stroke Program Early CT Score: Feasibility and Reliability. Radiology 291:141\u0026ndash;148. https://doi.org/10.1148/radiol.2019181228\u003c/li\u003e\n\u003cli\u003eLiu Y, Wen Z, Wang Y, et al (2024) Artificial intelligence in ischemic stroke images: current applications and future directions. Front Neurol 15:1418060. https://doi.org/10.3389/fneur.2024.1418060\u003c/li\u003e\n\u003cli\u003eHeinrich A, Streckenbach F, Beller E, et al (2021) Deep Learning-Based Image Reconstruction for CT Angiography of the Aorta. Diagnostics 11:2037. https://doi.org/10.3390/diagnostics11112037\u003c/li\u003e\n\u003cli\u003eKoetzier LR, Mastrodicasa D, Szczykutowicz TP, et al (2023) Deep Learning Image Reconstruction for CT: Technical Principles and Clinical Prospects. Radiology 306:e221257. https://doi.org/10.1148/radiol.221257\u003c/li\u003e\n\u003cli\u003eKim I, Kang H, Yoon HJ, et al (2021) Deep learning\u0026ndash;based image reconstruction for brain CT: improved image quality compared with adaptive statistical iterative reconstruction-Veo (ASIR-V). Neuroradiology 63:905\u0026ndash;912. https://doi.org/10.1007/s00234-020-02574-x\u003c/li\u003e\n\u003cli\u003eal. R Luis SanBerkhemer, Olvert A et (2018) Imaging features and safety and efficacy of endovascular stroke treatment: a meta-analysis of individual patient-level data. Lancet Neurol 17:895\u0026ndash;904\u003c/li\u003e\n\u003cli\u003eYoo AJ, Zaidat OO, Chaudhry ZA, et al (2014) Impact of Pretreatment Noncontrast CT Alberta Stroke Program Early CT Score on Clinical Outcome After Intra-Arterial Stroke Therapy. Stroke 45:746\u0026ndash;751. https://doi.org/10.1161/STROKEAHA.113.004260\u003c/li\u003e\n\u003cli\u003eKaveeta C, Alhabli I, Bala F, et al (2025) The treatment effect across ASPECTS in acute ischemic stroke: Analysis from the AcT trial. Int J Stroke 20:64\u0026ndash;74. https://doi.org/10.1177/17474930241273561\u003c/li\u003e\n\u003cli\u003eKlemenz A-C, Beckert L, Manzke M, et al (2024) Influence of Deep Learning Based Image Reconstruction on Quantitative Results of Coronary Artery Calcium Scoring. Acad Radiol 31:2259\u0026ndash;2267. https://doi.org/10.1016/j.acra.2024.03.020\u003c/li\u003e\n\u003cli\u003ePula M, Kucharczyk E, Piersiak M, et al (2025) Improved CTA imaging for stroke evaluation \u0026ndash; deep learning and iterative reconstruction comparative study. Neuroradiology. https://doi.org/10.1007/s00234-025-03733-8\u003c/li\u003e\n\u003cli\u003eNishikawa M, Machida H, Shimizu Y, et al (2022) Image quality and radiologists\u0026rsquo; subjective acceptance using model-based iterative and deep learning reconstructions as adjuncts to ultrahigh-resolution CT in low-dose contrast-enhanced abdominopelvic CT: phantom and clinical pilot studies. Abdom Radiol 47:891\u0026ndash;902. https://doi.org/10.1007/s00261-021-03373-5\u003c/li\u003e\n\u003cli\u003eAltmann S, Abello Mercado MA, Ucar FA, et al (2023) Ultra-High-Resolution CT of the Head and Neck with Deep Learning Reconstruction\u0026mdash;Assessment of Image Quality and Radiation Exposure and Intraindividual Comparison with Normal-Resolution CT. Diagnostics 13:1534. https://doi.org/10.3390/diagnostics13091534\u003c/li\u003e\n\u003cli\u003eSun J, Li H, Wang B, et al (2021) Application of a deep learning image reconstruction (DLIR) algorithm in head CT imaging for children to improve image quality and lesion detection. BMC Med Imaging 21:108. https://doi.org/10.1186/s12880-021-00637-w\u003c/li\u003e\n\u003cli\u003eAlagic Z, Diaz Cardenas J, Halldorsson K, et al (2022) Deep learning versus iterative image reconstruction algorithm for head CT in trauma. Emerg Radiol 29:339\u0026ndash;352. https://doi.org/10.1007/s10140-021-02012-2\u003c/li\u003e\n\u003cli\u003eHuang X, Zhao W, Wang G, et al (2023) Improving image quality with deep learning image reconstruction in double-low-dose head CT angiography compared with standard dose and adaptive statistical iterative reconstruction. Br J Radiol 96:20220625. https://doi.org/10.1259/bjr.20220625\u003c/li\u003e\n\u003cli\u003eAustein F, Wodarg F, J\u0026uuml;rgensen N, et al (2019) Automated versus manual imaging assessment of early ischemic changes in acute stroke: comparison of two software packages and expert consensus. Eur Radiol 29:6285\u0026ndash;6292. https://doi.org/10.1007/s00330-019-06252-2\u003c/li\u003e\n\u003cli\u003eNeuhaus A, Seyedsaadat SM, Mihal D, et al (2020) Region-specific agreement in ASPECTS estimation between neuroradiologists and e-ASPECTS software. J NeuroInterventional Surg 12:720\u0026ndash;724. https://doi.org/10.1136/neurintsurg-2019-015442\u003c/li\u003e\n\u003cli\u003eWaleed Brinjikji, Mehdi Abbasi, Catherine Arnold, et al (2021) E-ASPECTS software improves interobserver agreement and accuracy of interpretation of aspects score. Interv Neuroradiol 27:781\u003c/li\u003e\n\u003cli\u003eL\u0026ouml;ffler MT, Sollmann N, M\u0026ouml;nch S, et al (2021) Improved Reliability of Automated ASPECTS Evaluation Using Iterative Model Reconstruction from Head CT Scans. J Neuroimaging 31:341\u0026ndash;347. https://doi.org/10.1111/jon.12810\u003c/li\u003e\n\u003cli\u003eFatih Seker, Johannes Pfaff, Simon Nagel, et al (2018) CT Reconstruction Levels Affect Automated and Reader-Based ASPECTS Ratings in Acute Ischemic Stroke. J Neuroimaging 29:62\u0026ndash;64\u003c/li\u003e\n\u003cli\u003eKonno M, Otani T, Matsuda M, et al (2025) Deep learning-based computed tomography reconstruction improves image quality but does not significantly affect Alberta stroke program early CT score evaluation in acute middle cerebral artery territory infarction. Neuroradiology. https://doi.org/10.1007/s00234-025-03804-w\u003c/li\u003e\n\u003cli\u003eVan Horn N, Kniep H, Broocks G, et al (2021) ASPECTS Interobserver Agreement of 100 Investigators from the TENSION Study. Clin Neuroradiol 31:1093\u0026ndash;1100. https://doi.org/10.1007/s00062-020-00988-x\u003c/li\u003e\n\u003cli\u003eKaul D, Kahn J, Huizing L, et al (2015) Reducing Radiation Dose in Adult Head CT using Iterative Reconstruction \u0026ndash; A Clinical Study in 177 Patients. R\u0026ouml;Fo - Fortschritte Auf Dem Geb R\u0026ouml;ntgenstrahlen Bildgeb Verfahr 188:155\u0026ndash;162. https://doi.org/10.1055/s-0041-107200\u003c/li\u003e\n\u003cli\u003eNagayama Y, Iwashita K, Maruyama N, et al (2023) Deep learning-based reconstruction can improve the image quality of low radiation dose head CT. Eur Radiol 33:3253\u0026ndash;3265. https://doi.org/10.1007/s00330-023-09559-3\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"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":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Stroke, ASPECTS, computed tomography, deep learning, iterative reconstruction","lastPublishedDoi":"10.21203/rs.3.rs-7904400/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7904400/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cb\u003ePurpose\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe Alberta Stroke Program Early CT Score (ASPECTS) and advances in CT reconstruction play an important role in the neurodiagnostic workflow. This study examines the effect of these reconstruction techniques on automated ASPECTS.\u003c/p\u003e\u003cp\u003e\u003cb\u003eMethods\u003c/b\u003e\u003c/p\u003e\u003cp\u003eIn a retrospective study, 173 patients (median age 79 years, 39% female) with suspected middle cerebral artery infarction underwent non-contrast CT scans reconstructed with Filtered Back Projection (FBP), ASIR-V (30% and 60%), and DLIR (low, medium, and high). Automated ASPECTS were analyzed, with FBP as the reference standard.\u003c/p\u003e\u003cp\u003e\u003cb\u003eResults\u003c/b\u003e\u003c/p\u003e\u003cp\u003eBland\u0026ndash;Altman analysis revealed a mean bias of ASIR and DLIR underestimating ASPECTS compared to FBP, which was less pronounced for ASIR-V 30% (-0.057 ) and DLIR-L (-0.069) compared to ASIR-V 60% (-0.126), DLIR-M (-0.121), and DLIR-H (-0.086). The region with the highest overestimation, compared to FBP, was M3 (n\u0026thinsp;=\u0026thinsp;23), and with the highest underestimation was the insular ribbon (n\u0026thinsp;=\u0026thinsp;51). Regarding the ASPECTS\u0026thinsp;\u0026lt;\u0026thinsp;6 threshold, most patients were re-classified from ASPECTS\u0026thinsp;\u0026le;\u0026thinsp;5 to ASPECTS\u0026thinsp;\u0026ge;\u0026thinsp;6 with DLIR-M (n\u0026thinsp;=\u0026thinsp;5 ), which also showed the strongest agreement with expert consensus (κ\u0026thinsp;=\u0026thinsp;0.352).\u003c/p\u003e\u003cp\u003e\u003cb\u003eConclusion\u003c/b\u003e\u003c/p\u003e\u003cp\u003eBoth ASIR-V and DLIR led to only a minor underestimation of ASPECTS compared to FBP. However, more patients were overestimated to ASPECTS\u0026thinsp;\u0026ge;\u0026thinsp;6, making them available for endovascular therapy, which was most pronounced for DLIR-M. DLIR-M also exhibited the highest agreement with expert consensus for automated ASPECTS. Careful selection of reconstruction parameters, as well as further optimization and standardization of these techniques, is therefore essential for broader application in stroke imaging.\u003c/p\u003e","manuscriptTitle":"Influence of deep learning image reconstruction and adaptive statistical iterative reconstruction-V on automated Alberta Stroke Program Early CT Score- evaluation","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-13 07:53:08","doi":"10.21203/rs.3.rs-7904400/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"5b32c165-0c62-4f4e-aa8a-42e1e8cb748a","owner":[],"postedDate":"November 13th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-02-23T16:02:49+00:00","versionOfRecord":{"articleIdentity":"rs-7904400","link":"https://doi.org/10.1007/s00234-026-03938-5","journal":{"identity":"neuroradiology","isVorOnly":false,"title":"Neuroradiology"},"publishedOn":"2026-02-18 15:59:13","publishedOnDateReadable":"February 18th, 2026"},"versionCreatedAt":"2025-11-13 07:53:08","video":"","vorDoi":"10.1007/s00234-026-03938-5","vorDoiUrl":"https://doi.org/10.1007/s00234-026-03938-5","workflowStages":[]},"version":"v1","identity":"rs-7904400","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7904400","identity":"rs-7904400","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}