Use of Artificial Intelligence in Analysis of Endoscopic Images Following Complete Clinical Response in Rectal Cancer

Use of Artificial Intelligence in Analysis of Endoscopic Images Following Complete Clinical Response in Rectal Cancer | 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 Use of Artificial Intelligence in Analysis of Endoscopic Images Following Complete Clinical Response in Rectal Cancer Muhammad Ahsan Javed, Miguel Mascarenhas, Francisco Mendes, Eduardo Carvalho, and 5 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7775722/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 9 You are reading this latest preprint version Abstract Background Patients with locally advanced rectal cancer (LARC) who have a complete clinical response (cCR) to neoadjuvant chemoradiotherapy (nCRT) may opt for organ preservation, wait and watch – (W&W). This consists of an intense surveillance program including serial endoscopies, pelvic MRI, CEA, and CT scans to detect recurrent disease at an early stage. However, identifying residual or recurrent lesions endoscopically in these cases can be challenging due to mucosal changes such as friability and neovascularization. We developed a novel deep learning model to assist in the detection of residual or recurrent rectal cancer lesions during proctosigmoidoscopy. Methods We trained a convolutional neural network (Wide ResNet-101-2) on a dataset of 1,795 annotated frames from proctosigmoidoscopy exams of 97 patients treated at a tertiary referral centre. Residual or recurrent disease was defined by histopathological confirmation. The dataset was split into training and testing cohorts using a 90/10% patient-level split. Results Out of 97 patients, 24 (363 frames) had confirmed residual or recurrent disease, while 73 (1,432 frames) presented normal rectal mucosa. The model achieved an overall accuracy of 92.8%, with a sensitivity of 80.0%, specificity of 97.3%, PPV of 90.9%, NPV of 94.0%, and an AUROC of 0.886. Conclusion To the best of our knowledge, this is the first deep learning model specifically developed for the detection of residual or recurrent disease following in W&W patients during endoscopic examination. This tool has the potential to enhance early lesion detection, guide clinical decision-making, and increase opportunities for salvage, curative treatment strategies. Rectal Cancer Complete Clinical response Wait & Watch and Artificial Intelligence Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction Colorectal cancer (CRC) is the third most frequently diagnosed cancer globally and the second leading cause of cancer-related mortality. Rectal cancer comprises about 30–35% of all CRC cases, representing a major portion of the global disease impact [ 1 ]. Management of locally advanced rectal cancer (LARC) presents considerable clinical challenges and typically necessitates multimodal therapeutic approaches to optimize patient outcomes. Evidence demonstrates that neoadjuvant chemoradiotherapy (nCRT) substantially improves local disease control and overall survival, establishing it as a fundamental component of contemporary rectal cancer treatment protocols [ 2 , 3 ]. A key therapeutic milestone following nCRT is the attainment of a complete clinical response (cCR), characterized by the absence of detectable malignancy based on clinical, radiological, and endoscopic assessments [ 4 ]. The emergence of the cCR subgroup has led to a paradigm shift in rectal cancer management, enabling the adoption of organ-preserving “wait and watch” (W&W) strategies. This approach circumvents the morbidity and functional deficits associated with total mesorectal excision. Patients in this cohort are closely monitored through rigorous follow-up involving serial endoscopic examinations, MRI, CT, and CEA levels [ 5 ]. While this strategy enhances quality of life and mitigates surgical risks, it carries the inherent risk that approximately 20–25% of patients with initial cCR will develop luminal recurrence, predominantly within the first two years. Therefore, early and accurate detection of residual or recurrent disease during surveillance is critical to ensure oncologic safety and facilitate timely salvage interventions. Despite regular follow-up, endoscopic detection of recurrence remains challenging. Radiation-induced mucosal alterations including fibrosis, friability, and neovascularization can obscure visualization and simulate recurrent lesions, resulting in diagnostic ambiguity. These factors contribute to substantial inter-observer variability among endoscopists, potentially causing both over and underdiagnosis. Given these diagnostic challenges, there is a pressing need for objective, reliable, and reproducible tools to enhance detection accuracy in post nCRT surveillance. Artificial intelligence (AI), particularly deep learning (DL), has recently shown significant promise in advancing medical image analysis. Among DL architectures, convolutional neural networks (CNNs) have proven especially proficient in interpreting visual data such as endoscopic and radiologic images. Within gastroenterology, CNN-based algorithms have been successfully applied to detect colonic polyps and to distinguish neoplastic from non-neoplastic lesions with notable accuracy [ 6 ]. Despite these advances, the utilization of AI in the post nCRT surveillance of LARC remains insufficiently explored. Very few studies have focused on developing AI-driven models to detect residual or recurrent disease in patients exhibiting complete cCR, representing a notable gap in oncologic care [ 7 ]. AI’s capability to detect subtle mucosal architectural changes often overlooked by human observers could markedly improve early detection of recurrence, thereby minimizing the need for invasive biopsies or additional imaging. Moreover, AI tools can provide consistent, scalable support as adjunct readers, augmenting clinical decision-making without supplanting expert judgment. The aim of this study is to develop a novel AI model for analysis of endoscopic images following complete clinical response in rectal cancer. Methods Study Design This retrospective, observational study was conducted at University Hospitals Liverpool and affiliated Clatterbridge Cancer Centre, tertiary referral centres specializing in the management of patients with LARC. Study population included individuals who were on W&W protocol following cCR to nCRT. Eligibility criteria for patient inclusion were aligned with standards established in previous multicentre studies, including those by [ 8 ], to maintain methodological consistency. Inclusion criteria required patients to be over 18 years of age, have histologically confirmed rectal adenocarcinoma, have completed nCRT, and exhibit a cCR as determined by both MRI and endoscopy. Furthermore, availability of both pre-treatment and high-resolution post nCRT endoscopic images suitable for deep learning analysis was mandatory for inclusion. The study design is consistent with approaches previously adopted by researchers who employed temporally stratified cohorts to develop and validate deep learning models using surveillance imaging data [ 8 ]. Figure 1 presents a visual pipeline in which endoscopic images of rectal cancer are processed through a convolutional neural network (CNN) model. The system is designed to automatically exclude suboptimal or inadequate frames, ensuring only high-quality inputs are analysed. The trained AI model then classifies the remaining frames to support clinical decision-making by distinguishing between post-treatment fibrotic scarring and true tumour recurrence, thereby enhancing diagnostic accuracy in the surveillance of patients following neoadjuvant therapy. Image Acquisition and Preprocessing High-resolution white-light proctosigmoidoscopy was employed to and regions of interest (ROI) were selected based on optimal mucosal visualization, excluding frames compromised by artefacts such as bleeding, motion blur, or poor focus. A standardized preprocessing pipeline was implemented to enhance image quality and consistency across the dataset. This included histogram equalization to normalize lighting, spatial rescaling to ensure uniform resolution, and the application of denoising filters to reduce background noise. Distinct from studies primarily adapting techniques from natural image processing, our methodology also incorporated targeted artefact removal and image stabilization procedures to further improve image fidelity for deep learning analysis. Data Annotation and Reference Standard Image annotations were validated against histopathological findings, serving as the diagnostic gold standard. Frames corresponding to confirmed residual or recurrent malignancy were categorized as “tumour present,” while those obtained from patients who maintained a cCR without evidence of recurrence for a minimum follow-up period of 24 months were labelled as “no tumour.” This annotation strategy is consistent with the classification framework utilized in other CNNs [ 6 ]. The annotated endoscopic image dataset was used for training and optimizing the deep learning–based AI model at University of Porto. Deep Learning Architecture The model architecture was built upon the Wide ResNet-101-2 framework, a deep residual network variant optimized for extracting complex features from high-dimensional medical imaging data. Initial pretraining on the ImageNet dataset enabled effective transfer learning, facilitating improved convergence and performance on the relatively limited endoscopic dataset, an approach consistent with established practices in AI-based endoscopic analysis [ 9 ]. To enhance generalizability and reduce overfitting, the network incorporated batch normalization and dropout layers. Additionally, spatial attention mechanisms were integrated to focus on critical mucosal textures and vascular patterns relevant to the detection of residual or recurrent disease. The dataset was split at the subject level using a 90/10 ratio, allocating 87 patients (1,615 frames) to the training set and 10 patients (180 frames) to the test set. To prevent data leakage, patient-level separation was ensured, an approach consistent with the methodology used by [ 7 ]. To enhance model generalizability, image augmentation techniques such as rotation, flipping, and brightness adjustment were applied, in line with practices adopted in multicentre endoscopic research. Testing and Performance Evaluation The evaluation framework incorporated key performance metrics including sensitivity, specificity, accuracy, positive predictive value (PPV), negative predictive value (NPV), and area under the receiver operating characteristic curve (AUROC), with 95% confidence intervals estimated for AUROC. Confusion matrix analysis was conducted to assess misclassification rates, and statistical reliability was ensured using binomial exact tests [ 10 ]. Statistical Analysis Descriptive statistics summarized the patient count, frame distribution per class, and the ratio of normal to abnormal cases. Primary classification metrics such as accuracy, sensitivity (recall), specificity, PPV, NPV, and AUROC were computed at the frame level by comparing model outputs against expert-annotated ground truth. To ensure statistical reliability, 95% confidence intervals (CIs) were estimated using a non-parametric bootstrapping approach with 1,000 iterations. A confusion matrix was constructed to visualize true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN), serving as the basis for calculating sensitivity, specificity, PPV and NPV. The statistical significance of the association between actual and predicted classifications was assessed using a chi-square test of independence, with a two-tailed p-value < 0.05 considered significant. Additionally, the receiver operating characteristic (ROC) curve was plotted, and the AUROC was computed to evaluate the model’s discriminative capacity across varying thresholds. AUROC values between 0.8 and 0.9 were interpreted as indicative of excellent diagnostic performance. All analyses were performed using Python (v3.9). Results Cohort Overview and Image Data Distribution Between January 2020 and December 2024, 97 patients were classified to have a cCR after nCRT. From these, 1,795 annotated frames were collected. Twenty-four patients (363 frames) had histologically confirmed residual or recurrent disease, while 73 patients (1,432 frames) exhibited normal rectal mucosa indicative of a complete response. Model Performance Metrics Figure 2 demonstrate the model's strong ability to differentiate between normal mucosa and recurrent disease with high overall precision. The model achieved an accuracy of 92.8%, indicating that approximately 1,667 out of 1,795 frames were correctly classified. This metric reflects the model’s general effectiveness, even in the presence of class imbalance. Sensitivity was 80% indicating the model accurately flagged 4 out of every 5 disease frames. While some false negatives remain, this sensitivity level is considered clinically acceptable for surveillance purposes. The model exhibited a high specificity of 97.3%, which is particularly important in the context of surveillance, as it reduces false positives, thereby minimizing unnecessary procedures and patient distress. A positive predictive value (PPV) of 90.9% indicates that 9 out of 10 frames classified as recurrent disease were truly positive, reducing the likelihood of overtreatment. Similarly, the negative predictive value (NPV) of 94.0% suggests that frames predicted as negative were reliably free of recurrence, supporting confident decisions to avoid further invasive investigation. Finally, the area under the receiver operating characteristic curve (AUROC) was 0.886, reflecting excellent discriminative performance. AUROC values in the 0.8–0.9 range are indicative of strong diagnostic capability, confirming the model’s ability to maintain a balanced trade-off between sensitivity and specificity. The high specificity further reinforces the model’s reliability in accurately identifying patients with true complete clinical response, thereby reducing the risk of overdiagnosis. The Receiver Operating Characteristic (ROC) curve serves as a critical tool for assessing the diagnostic performance of the AI model in differentiating residual or recurrent disease from normal mucosa during endoscopic surveillance of locally advanced rectal cancer. By plotting the true positive rate (sensitivity) against the false positive rate (1 – specificity) across a range of classification thresholds, the ROC curve offers a comprehensive evaluation of the model’s discriminative ability. In this study, the model achieved an area under the ROC curve (AUROC) of 0.886, indicating a high level of diagnostic accuracy. This elevated AUROC underscores the model’s effectiveness in maintaining an optimal balance between sensitivity and specificity, even in complex cases where post-treatment mucosal alterations challenge visual interpretation. Accordingly, ROC analysis supports the model’s reliability in accurately distinguishing pathological from normal tissue, reinforcing its potential utility in enhancing clinical decision-making within the watch-and-wait surveillance framework. Confusion Matrix and Classification Breakdown The confusion matrix analysis further elaborates the model’s performance at the frame level shown on Fig. 4. The AI model exhibited strong classification performance, as demonstrated by the confusion matrix outcomes. It accurately detected 290 frames containing residual or recurrent disease (true positives), reflecting its effectiveness in identifying cases requiring clinical intervention. However, 73 disease frames were misclassified as normal (false negatives), indicating some limitations in sensitivity where a subset of pathological cases went undetected. Additionally, the model misclassified 39 frames of normal mucosa as diseased (false positives), which could potentially result in unnecessary diagnostic follow-up. Notably, it correctly identified 1,393 frames of normal mucosa (true negatives), highlighting its high specificity and reliability in distinguishing healthy tissue. Collectively, these results suggest the model excels at differentiating between diseased and normal mucosa, with a particular strength in reducing false positives while maintaining reasonable sensitivity an essential balance for effective clinical decision-making [ 11 ]. Statistical Significance and Confidence Intervals To assess the robustness of the AI model, 95% confidence intervals (CIs) were calculated for the primary performance metrics using bootstrapping techniques. The model achieved an accuracy of 92.8% (95% CI: 91.4–94.1%), sensitivity of 80.0% (95% CI: 75.2–84.6%), specificity of 97.3% (95% CI: 96.1–98.4%), positive predictive value (PPV) of 90.9% (95% CI: 87.2–94.6%), negative predictive value (NPV) of 94.0% (95% CI: 92.5–95.5%), and an area under the receiver operating characteristic curve (AUROC) of 0.886 (95% CI: 0.863–0.910). The narrow confidence intervals indicate high stability and reliability of the model’s performance. Furthermore, chi-square testing confirmed significant class discrimination ability (p < 0.001), validating that the model’s predictive power is statistically significant and exceeds random classification. Discussion The performance metrics from our model indicate that the proposed AI framework holds significant potential as a supportive diagnostic tool during endoscopic surveillance for rectal cancer patients on W&W pathway. Timely and accurate identification of residual or recurrent lesions is crucial for initiating appropriate salvage treatments while avoiding unnecessary surgical procedures when no disease is present. The model’s high NPV of 94.0% is particularly meaningful in the context of watch-and-wait strategies, as it suggests that frames predicted as normal are indeed likely to be disease-free. This reduces the need for additional investigations, thereby minimizing patient and financial burden. Conversely, the PPV of 90.9% provides reassurance that when the AI flags a lesion as suspicious, it is very likely to be a true positive. This enhances clinical confidence in leveraging the AI model to support decision-making, particularly in borderline cases where visual assessments may be subjective or unclear. Moreover, the high specificity of 97.3% highlights the model’s strong ability to avoid false positives, which are common in traditional evaluations due to post-treatment changes such as inflammation, mucosal irregularities, or scarring. This further reinforces the AI tool's utility in improving diagnostic precision in endoscopic surveillance. Our model’s higher accuracy (92.8%) and AUROC (0.886) indicate more precise discrimination of recurrence within a single-centre dataset. The current model’s performance sits between these extremes, suggesting strong internal validity with potential for broader generalisability. Previous studies have reported mixed outcomes in detecting rectal cancer recurrence following nCRT using conventional imaging techniques. Studies report considerable interobserver variability in post-treatment assessments and limitations of MRI in accurately distinguishing post-treatment scarring from residual tumour tissue [ 12 ]. In contrast, the AI model developed in the present study demonstrated enhanced performance, particularly in interpreting endoscopic images, achieving greater accuracy and fewer false positives compared to conventional approaches [ 13 – 16 ], as summarized in Table 2 . When compared with existing AI-based approaches for endoscopic surveillance in colorectal and rectal cancer, the current model shows clear improvements. Previous studies have reported variable sensitivity (70–90%) and specificity (80–95%) across models. In contrast, the present study's AUROC of 0.886, along with superior PPV and specificity, indicates stronger performance in detecting subtle post-nCRT mucosal abnormalities. These advancements may be attributed to the use of the WideResNet-101-2 architecture, which is well-suited for extracting deep, abstract features in complex post-treatment imaging data. Despite encouraging results, the study has several limitations. The dataset, while enriched, originates from a single tertiary centre, potentially limiting external applicability. Additionally, the relatively small number of recurrence-positive cases (24 out of 97 patients) may restrict the model’s generalizability to broader and more diverse populations. The presence of 48 false negatives also presents a concern, emphasizing the importance of integrating additional diagnostic modalities, such as MRI and CEA levels, to support final clinical decisions. Analysis of false positives revealed that many incorrectly classified normal frames exhibited features such as scarring or radiation-induced mucosal changes, which closely resemble early recurrence. This underscores the need for incorporating complementary diagnostic modalities, such as MRI or histopathology, to improve diagnostic accuracy and confidence. Ongoing work focuses on incorporating temporal data from serial endoscopic exams to improve the model’s predictive consistency over time. Moreover, the inclusion of multimodal inputs, such as radiomics, serum biomarkers, and MRI findings—is expected to enhance diagnostic accuracy beyond that of image-only system [ 20 ]. Additionally, despite strong overall accuracy and AUROC, the model’s generalizability across different clinical settings remains unproven. Variations in endoscopic equipment, lighting, mucosal staining techniques, and patient demographics could impact performance. Therefore, prospective validation across multiple centres is essential to confirm broader applicability. By delivering standardized, reproducible assessments, the model has the potential to improve the efficiency of endoscopic surveillance and reduce clinician workload. Early detection of recurrence supports timely therapeutic action, potentially improving clinical outcomes. In addition, widespread adoption of such technology may reduce healthcare expenditures by limiting unnecessary investigations. Collectively, this AI model offers a meaningful improvement to existing W&W surveillance protocols, supporting more effective and patient-centred management. The W&W approach, which emphasizes organ preservation in patients with cCR, depends on accurate and reproducible detection of recurrence. Studies have reported reported accuracy ranging from 53% to 90% and significant variability between observers using traditional assessment methods [ 21 ]. The current CNN model’s combination of high specificity and acceptable sensitivity may provide a more standardized and dependable alternative, contributing to safer and more consistent surveillance workflows. The choice of the Wide ResNet-101-2 architecture, characterized by increased width rather than depth, facilitates deeper feature extraction and mitigates vanishing gradient issues. This architecture is particularly well-suited for capturing complex mucosal and vascular patterns in endoscopic imagery offering potential advantages over alternative frameworks such as EfficientNet [ 22 ] or DenseNet [ 23 ]. For effective clinical translation, future integration of the model into endoscopic systems with real-time inference capabilities, saliency map visualization, and user-friendly interfaces will be essential. The most effective deployment strategy involves augmenting, rather than replacing, the clinician's role. Future research should include prospective video-based validation studies and the incorporation of clinician feedback to facilitate clinical acceptance and workflow integration. On the other hand, despite satisfactory accuracy, the black-box nature of deep learning models reduces clinician confidence in model decisions, thereby decreasing their applicability. In fact, the development of models with increased complexity is often associated with a decreased clinician understanding of the reasons behind such decisions [ 24 ]. In this context, explainable AI (XAI) aims to address this limitation by providing specific tools to help clinicians understand the reasons behind model outputs. Future studies will therefore focus on developing XAI tools for detecting specific regions of an image that may indicate residual lesions or recurrence. As validated in previous studies, the use of heatmaps or bounding boxes would be of great interest [ 25 ]. These mechanisms would not only increase confidence in model decisions but also potentially improve the selection of biopsy sites, ultimately transforming patient outcomes. In conclusion, this tool has the potential to enhance early lesion detection, guide clinical decision-making, and increase opportunities for salvage, curative treatment strategies. To strengthen future deployment, efforts should focus on multicentre external validation, integration of multimodal inputs (e.g., MRI, serum biomarkers), and prospective clinical trials comparing standard care with AI-augmented workflows. Emphasis on explainability, through tools such as attention or saliency maps, as well as hardware optimization and regulatory compliance, will be vital to achieving safe and effective implementation in clinical practice [ 26 ]. Declarations Funding: This study received no external funding. Institutional Review Board Statement: This non-interventional study adhered to the Declaration of Helsinki’s ethical principles and received approval from the institutional review boards. To ensure data privacy, all potentially identifying information was removed, and each patient’s data was assigned a random, anonymized number before being incorporated into the dataset. A certified legal team with data protection expertise further guaranteed that the data remained untraceable, fully complying with General Data Protection Regulation (GDPR) standards. Informed Consent Statement: Informed consent was waived as this was a retrospective non interventional study. Data Availability Statement: Additional data are available upon reasonable request. Conflicts of Interest: The authors declare no conflicts of interest. Author Contributions Conceptualization: MAJ (Muhammad Ahsan Javed), MM (Miguel Mascarenhas) and SA (Shakil Ahmed); Methodology : MM, FM (Francisco Mendes), EC (Eduardo Carvalho) and AS (André Santos); Formal analysis and investigation : FM, AS, EC, RR (Reshma Rajan), ZK (Zarnigar Khan) and IB (Iain Blake); Writing - original draft preparation : MAJ, MM, RR, FM; Writing - review and editing : all authors; Resources: RR, IB, ZK; Supervision: MM, MAJ and SA Acknowledgments: We would like to thank Professor Sun Myint and Dr Ngy Wah Than for contributing to images from Clatterbridge Cancer Centre and Mr Shay Willoughby for his support in this project References Sung, H., Ferlay, J., Siegel, R. L., Laversanne, M., Soerjomataram, I., Jemal, A. and Bray, F. (2021) Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA: a Cancer Journal for Clinicians 71, 209–249. Sauer, R., Becker, H., Hohenberger, W., Rödel, C., Wittekind, C., Fietkau, R., Martus, P., Tschmelitsch, J., Hager, E., Hess, C. F., et al. (2004) Preoperative versus Postoperative Chemoradiotherapy for Rectal Cancer. New England Journal of Medicine 351, 1731–1740. Maas, M., Nelemans, P. J., Valentini, V., Das, P., Rödel, C., Kuo, L.-J., Calvo, F. A., García-Aguilar, J., Glynne-Jones, R., Haustermans, K., et al. (2010) Long-term outcome in patients with a pathological complete response after chemoradiation for rectal cancer: a pooled analysis of individual patient data. The Lancet Oncology 11, 835–844. Habr-Gama, A., Perez, R. O., Nadalin, W., Sabbaga, J., Ribeiro, U., Silva e Sousa, A. H., Campos, F. G., Kiss, D. R. and Gama-Rodrigues, J. (2004) Operative Versus Nonoperative Treatment for Stage 0 Distal Rectal Cancer Following Chemoradiation Therapy. Annals of Surgery 240, 711–718. Fokas, E., Appelt, A., et al. (2021) International consensus recommendations on key outcome measures for organ preservation after (chemo)radiotherapy in patients with rectal cancer. Nat Rev Clin Oncol 18, 805–816. Spadaccini, M., Massimi, D., Mori, Y., Ludovico Alfarone, Alessandro Fugazza, Maselli, R., Sharma, P., Facciorusso, A., Hassan, C. and Repici, A. (2023) Artificial Intelligence-Aided Endoscopy and Colorectal Cancer Screening. Diagnostics, Multidisciplinary Digital Publishing Institute 13, 1102–1102. Zhang, J., Liu, R., Wang, X., Zhang, S., Shao, L., Liu, J., Zhao, J., Wang, Q., Tian, J. and Lu, Y. (2024) Deep learning model based on endoscopic images predicting treatment response in locally advanced rectal cancer undergo neoadjuvant chemoradiotherapy: a multicenter study. Journal of Cancer Research and Clinical Oncology 150. Shen, H., Jin, Z., Chen, Q., Zhang, L., You, J., Zhang, S. and Zhang, B. (2024) Image-based artificial intelligence for the prediction of pathological complete response to neoadjuvant chemoradiotherapy in patients with rectal cancer: a systematic review and meta-analysis. La radiologia medica, Springer Science + Business Media 129, 598–614. Sarwinda, D., Paradisa, R. H., Bustamam, A. and Anggia, P. (2021) Deep Learning in Image Classification using Residual Network (ResNet) Variants for Detection of Colorectal Cancer. Procedia Computer Science 179, 423–431. Letizia Monguzzi, Ippolito, D., Davide Paolo Bernasconi, Chiara Trattenero, Galimberti, S. and Sandro Sironi. (2013) Locally advanced rectal cancer: Value of ADC mapping in prediction of tumor response to radiochemotherapy. European Journal of Radiology, Elsevier BV 82, 234–240. Zhou, X., Dai, J., Lu, Y., Zhao, Q., Liu, Y., Wang, C., Zhao, Z., Wang, C., Gao, Z., Yu, Y., et al. (2024) Whole slide image based prognosis prediction in rectal cancer using unsupervised artificial intelligence. BMC Cancer, Springer Science and Business Media LLC 24. Horvat, N., Harini Veeraraghavan, Rizkallah, S., David, de, F., Zheng, J., Marinela Capanu, Fuqua, J. L., Maria Clara Fernandes, Sosa, R. E., et al. (2022) Combined artificial intelligence and radiologist model for predicting rectal cancer treatment response from magnetic resonance imaging: an external validation study. Abdominal Radiology, Springer Nature 47, 2770–2782 Kim, M., Park, T., Oh, B. Y., Kim, M. J., Cho, B.-J. and Son, I. T. (2024) Performance reporting design in artificial intelligence studies using image-based TNM staging and prognostic parameters in rectal cancer: a systematic review. Annals of Coloproctology 40, 13–26. İsmail Mendi, B., Kose, K., Fleshner, L., Adam, R., Safai, B., Farabi, B. and Atak, M. F. (2024) Artificial Intelligence in the Non-Invasive Detection of Melanoma. Life, MDPI AG 14, 1602. Williams, H., Thompson, H. M., Lee, C., Rangnekar, A., Gomez, J. T., Widmar, M., Wei, I. H., Pappou, E. P., Nash, G. M., Weiser, M. R., et al. (2024) Assessing Endoscopic Response in Locally Advanced Rectal Cancer Treated with Total Neoadjuvant Therapy: Development and Validation of a Highly Accurate Convolutional Neural Network. Annals of Surgical Oncology, Springer Science and Business Media LLC 31, 6443–6451. Wang, A., Mo, J., Zhong, C., Wu, S., Wei, S., Tu, B., Liu, C., Chen, D., Xu, Q., Cai, M., et al. (2021) Artificial intelligence-assisted detection and classification of colorectal polyps under colonoscopy: a systematic review and meta-analysis. Annals of Translational Medicine, AME Publishing Company 9, 1662–1662. Esteva, A., Kuprel, B., Novoa, R. A., Ko, J., Swetter, S. M., Blau, H. M., & Thrun, S. (2017). Dermatologist-level classification of skin cancer with deep neural networks. Nature, 542(7639), 115–118. Smith, S.C., Siau, K., Cannatelli, R., Antonelli, G., Shivaji, U.N., Ghosh, S., Saltzman, J.R., Hassan, C. and Iacucci, M., 2021. Training methods in optical diagnosis and characterization of colorectal polyps: a systematic review and meta-analysis. Endoscopy international open, 9(05), pp.E716-E726. Zhang, Y., Zhang, X., Wu, Q., Gu, C. and Wang, Z., 2021. Artificial intelligence-aided colonoscopy for polyp detection: a systematic review and meta-analysis of randomized clinical trials. Journal of Laparoendoscopic & Advanced Surgical Techniques, 31(10), pp.1143–1149. Zhang, Y., He, K., Guo, Y., Liu, X., Yang, Q., Zhang, C., Xie, Y., Mu, S., Guo, Y., Fu, Y., et al. (2020) A Novel Multimodal Radiomics Model for Preoperative Prediction of Lymphovascular Invasion in Rectal Cancer. Frontiers in Oncology 10. Maas, M., Beets-Tan, R. G. H., Lambregts, D. M. J., Lammering, G., Nelemans, P. J., Engelen, S. M. E., van Dam, R. M., Jansen, R. L. H., Sosef, M., Leijtens, J. W. A., et al. (2011) Wait-and-See Policy for Clinical Complete Responders After Chemoradiation for Rectal Cancer. Journal of Clinical Oncology 29, 4633–4640 Abd El-Ghany, S., Mahmood, M. A. and Abd El-Aziz, A. A. (2024) Adaptive Dynamic Learning Rate Optimization Technique for Colorectal Cancer Diagnosis Based on Histopathological Image Using EfficientNet-B0 Deep Learning Model. Electronics 13, 3126. Hasan, M., Islam, J., Ahmed, M. and Hasan, M. M. (2023) Prediction of Colon Cancer using DenseNet121, CNN, and REsNET50 Machine Learning Models and using Image Processing Techniques. 2023 International Conference on Artificial Intelligence Robotics, Signal and Image Processing (AIRoSIP), IEEE 296–301. Mascarenhas M, Mendes F, Martins M, et al. Explainable AI in Digestive Healthcare and Gastrointestinal Endoscopy. J Clin Med. 2025;14(2). Mascarenhas M, Mendes F, Fonseca F, et al. A Novel Deep Learning Model for Predicting Colorectal Anastomotic Leakage: A Pioneer Multicenter Transatlantic Study. Journal of Clinical Medicine. 2025;14(15):5462. Stanzione, A., Verde, F., Romeo, V., Boccadifuoco, F., Mainenti, P. P. and Maurea, S. (2021) Radiomics and machine learning applications in rectal cancer: Current update and future perspectives. World Journal of Gastroenterology 27, 5306–5321. Tables Table 1 Key components of the deep learning model architecture and training parameters used in the study. The Wide ResNet-101-2 model was trained on resized endoscopic images (224 × 224 pixels) using categorical cross-entropy loss. Regularization techniques included dropout and batch normalization. Data augmentation strategies were applied to enhance generalization. A patient-level 90:10 training and validation split was used, with early stopping triggered after 10 consecutive validation epochs without improvement. Component Description Model Architecture Wide ResNet-101-2 Input Image Dimensions 224 × 224 pixels (resized) Loss Function Categorical Cross-Entropy Regularization Dropout (rate = 0.3), Batch Normalization Data Augmentation Rotation, horizontal/vertical flip, brightness/contrast shift Epochs 50 (with early stopping if no improvement in 10 consecutive validation epochs) Training–Validation–Test Split 90/10% patient level split Table 2 Benchmark Comparison with Published AI Studies. Study & Year Method Modality AUROC Accuracy Sensitivity Specificity Esteva et al. [ 17 ] (2017) Deep CNN Dermoscopy 0.911 89.5% 76.0% 94.0% Smith et al. [ 18 ] (2021) DL Endoscopy Anal neoplasia 0.793 85.1% 70.0% 90.2% Zhang et al. [ 19 ] (2021) AI - Colonoscopy CRC detection 0.831 88.3% 75.2% 91.8% Current Study (2025) Wide ResNet-101-2 Procto- sigmoidoscopy 0.886 92.8% 80.0% 97.3% Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 14 Nov, 2025 Reviews received at journal 10 Nov, 2025 Reviews received at journal 19 Oct, 2025 Reviewers agreed at journal 11 Oct, 2025 Reviewers agreed at journal 09 Oct, 2025 Reviewers invited by journal 09 Oct, 2025 Editor assigned by journal 09 Oct, 2025 Submission checks completed at journal 04 Oct, 2025 First submitted to journal 03 Oct, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7775722","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":531947828,"identity":"7c76b2de-28a2-40c1-bbd7-ac8ad69a0d97","order_by":0,"name":"Muhammad Ahsan Javed","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA7klEQVRIie2RsQrCMBCGIwd1+bVrwFJfoVJwUXwWoVCXDo7ugi59AEEfplLoVHQN6KAITg5CF8HFhCqIYOrokG+444aP+y9hzGD4R7gq6DvWc3bKllQqIV4KflRYitdcrdjL+bEYO1s0eZAV40kfrJ4eCLlmyT73WwvsYPEwWC7yEDKlRxCaYCJiLSgFuU+NmUoYMcL1u9EWo9Md2EhlW5SKfdErnhh25ZYEVj2mUuFqiyZYR0TdHhBIZeaTusXiZ2+90pzvymC7Rjxw21M6kXwx17aD4+GSac5X1OKykyrqT5MKQXJ7UwwGg8HwyQOPPUKLgsw8GAAAAABJRU5ErkJggg==","orcid":"","institution":"The University of Liverpool","correspondingAuthor":true,"prefix":"","firstName":"Muhammad","middleName":"Ahsan","lastName":"Javed","suffix":""},{"id":531947829,"identity":"151b479d-754b-4ee9-b02c-f6bfbbbd98e7","order_by":1,"name":"Miguel Mascarenhas","email":"","orcid":"","institution":"Unidade Local de Saúde","correspondingAuthor":false,"prefix":"","firstName":"Miguel","middleName":"","lastName":"Mascarenhas","suffix":""},{"id":531947830,"identity":"0534e904-b1c8-4598-a8b4-593046297d28","order_by":2,"name":"Francisco Mendes","email":"","orcid":"","institution":"Unidade Local de Saúde","correspondingAuthor":false,"prefix":"","firstName":"Francisco","middleName":"","lastName":"Mendes","suffix":""},{"id":531947831,"identity":"1edd0627-9b2a-4d60-bd57-36a813a9db5e","order_by":3,"name":"Eduardo Carvalho","email":"","orcid":"","institution":"INEGI - Instituto de Ciência e Inovação em Engenharia Mecânica e Engenharia Industrial","correspondingAuthor":false,"prefix":"","firstName":"Eduardo","middleName":"","lastName":"Carvalho","suffix":""},{"id":531947832,"identity":"e04856ae-f9b9-4847-929c-622595fd8612","order_by":4,"name":"Reshma Rajan","email":"","orcid":"","institution":"The University of Liverpool","correspondingAuthor":false,"prefix":"","firstName":"Reshma","middleName":"","lastName":"Rajan","suffix":""},{"id":531947833,"identity":"d56fd17f-5068-4e62-97c1-d620d5bfd990","order_by":5,"name":"André Santos","email":"","orcid":"","institution":"University of Porto","correspondingAuthor":false,"prefix":"","firstName":"André","middleName":"","lastName":"Santos","suffix":""},{"id":531947834,"identity":"d0c8cc78-c53f-4ae1-b3e3-6a4527546b45","order_by":6,"name":"Zarnigar Khan","email":"","orcid":"","institution":"Liverpool University Hospital NHS Foundation Trust","correspondingAuthor":false,"prefix":"","firstName":"Zarnigar","middleName":"","lastName":"Khan","suffix":""},{"id":531947835,"identity":"0e4832d1-52dc-42d2-8e76-4db2197e1519","order_by":7,"name":"Iain Blake","email":"","orcid":"","institution":"The University of Liverpool","correspondingAuthor":false,"prefix":"","firstName":"Iain","middleName":"","lastName":"Blake","suffix":""},{"id":531947836,"identity":"9082cf3e-e47d-4f7a-9cfc-cb60d7cb0a7e","order_by":8,"name":"Shakil Ahmed","email":"","orcid":"","institution":"Liverpool University Hospital NHS Foundation Trust","correspondingAuthor":false,"prefix":"","firstName":"Shakil","middleName":"","lastName":"Ahmed","suffix":""}],"badges":[],"createdAt":"2025-10-03 18:38:16","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7775722/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7775722/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":94208282,"identity":"10462517-5d8d-48e6-b1f0-42f1dea780df","added_by":"auto","created_at":"2025-10-23 15:06:57","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":847171,"visible":true,"origin":"","legend":"","description":"","filename":"AIincCRsubmittediA.docx","url":"https://assets-eu.researchsquare.com/files/rs-7775722/v1/f1780870a7c00fe2adadd851.docx"},{"id":94209205,"identity":"6bcae4db-6e04-4195-bd59-1f5013376980","added_by":"auto","created_at":"2025-10-23 15:14:57","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10398,"visible":true,"origin":"","legend":"","description":"","filename":"80595888e6574b80b7fbbe22f814d6c9.json","url":"https://assets-eu.researchsquare.com/files/rs-7775722/v1/9638929581ff6634bfb533d9.json"},{"id":94207642,"identity":"36aa62d2-f26b-47d1-9249-3cf17d167afe","added_by":"auto","created_at":"2025-10-23 14:58:57","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":94276,"visible":true,"origin":"","legend":"","description":"","filename":"80595888e6574b80b7fbbe22f814d6c91enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7775722/v1/6ff234114161d60a21a52d3f.xml"},{"id":94207647,"identity":"c7accacc-06f7-458d-84bc-1f7292a0a9ed","added_by":"auto","created_at":"2025-10-23 14:58:57","extension":"eps","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":101826,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage1.eps","url":"https://assets-eu.researchsquare.com/files/rs-7775722/v1/e56547ea85e25fc0b0a42645.eps"},{"id":94207641,"identity":"85424606-523e-458f-acc8-15c3d551557b","added_by":"auto","created_at":"2025-10-23 14:58:57","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":410415,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7775722/v1/4abd8924f458d81802473320.png"},{"id":94207645,"identity":"8082f3b7-5798-4b18-b91a-32a973cd1549","added_by":"auto","created_at":"2025-10-23 14:58:57","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":368984,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7775722/v1/7fdfe5fdba514a7c94d2c43c.png"},{"id":94207646,"identity":"e4b6df40-a9dd-48ff-b1eb-01121756bcde","added_by":"auto","created_at":"2025-10-23 14:58:57","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":41735,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7775722/v1/a0bee6583bebf42b8f214603.png"},{"id":94207644,"identity":"f765d2db-827d-4930-935a-cb6179662d2b","added_by":"auto","created_at":"2025-10-23 14:58:57","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":73009,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7775722/v1/0e9fec0b8e24dd76db329b7f.png"},{"id":94207648,"identity":"2fb02c62-f695-49d4-aa58-003d31b5e1f2","added_by":"auto","created_at":"2025-10-23 14:58:57","extension":"xml","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":93563,"visible":true,"origin":"","legend":"","description":"","filename":"80595888e6574b80b7fbbe22f814d6c91structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7775722/v1/b17b5bc58256d1b9ebcde37d.xml"},{"id":94207649,"identity":"5f1a9fcb-9275-48a1-960b-e8140f05200d","added_by":"auto","created_at":"2025-10-23 14:58:57","extension":"html","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":102659,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7775722/v1/3518ca714656de257fe834de.html"},{"id":94207637,"identity":"7abf804e-97bd-4ef5-abb9-de1e77feda32","added_by":"auto","created_at":"2025-10-23 14:58:57","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":508227,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eVisual pipeline of endoscopic images processed through a CNN model. Representative endoscopic frames analyzed by the AI model, illustrating pre- and post-treatment appearances. The system correctly identified complete response (top right) and residual disease (bottom right).\u003c/em\u003e\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7775722/v1/0b2be73f6d511108b7fa6963.png"},{"id":94207635,"identity":"d6106f6f-520c-448a-a72a-110fceef5fb7","added_by":"auto","created_at":"2025-10-23 14:58:56","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":56001,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eDiagnostic performance metrics of the AI model, including accuracy, sensitivity, specificity, predictive values, and AUROC with 95% confidence interval.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7775722/v1/c1b8e0d07d4e7c1b7f3d56e3.png"},{"id":94208280,"identity":"3431ed55-a5a9-4cf6-9a2e-194789a1456a","added_by":"auto","created_at":"2025-10-23 15:06:57","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":36090,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eReceiver‑operating‑characteristic curve of the AI model denotes 95 % confidence intervals.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7775722/v1/76200ff788635b43c833e9e9.png"},{"id":94207640,"identity":"23be9053-529c-4f4a-a7fb-74b1b54a819d","added_by":"auto","created_at":"2025-10-23 14:58:57","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":95050,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eConfusion matrix illustrating the AI model’s performance in classifying endoscopic frames. The model accurately identified cases of residual or recurrent disease (true positives) and normal mucosa consistent with complete clinical response (true negatives), with few misclassifications, reflecting strong specificity and overall diagnostic reliability.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-7775722/v1/b67093747ab56d655e570fd8.png"},{"id":94209648,"identity":"8b0608b1-d1f2-4140-b1a9-2d18437e4088","added_by":"auto","created_at":"2025-10-23 15:22:57","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1559305,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7775722/v1/d27c86db-5e87-4033-bf9d-99da056f07c3.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Use of Artificial Intelligence in Analysis of Endoscopic Images Following Complete Clinical Response in Rectal Cancer","fulltext":[{"header":"Introduction","content":"\u003cp\u003eColorectal cancer (CRC) is the third most frequently diagnosed cancer globally and the second leading cause of cancer-related mortality. Rectal cancer comprises about 30\u0026ndash;35% of all CRC cases, representing a major portion of the global disease impact [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Management of locally advanced rectal cancer (LARC) presents considerable clinical challenges and typically necessitates multimodal therapeutic approaches to optimize patient outcomes. Evidence demonstrates that neoadjuvant chemoradiotherapy (nCRT) substantially improves local disease control and overall survival, establishing it as a fundamental component of contemporary rectal cancer treatment protocols [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. A key therapeutic milestone following nCRT is the attainment of a complete clinical response (cCR), characterized by the absence of detectable malignancy based on clinical, radiological, and endoscopic assessments [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThe emergence of the cCR subgroup has led to a paradigm shift in rectal cancer management, enabling the adoption of organ-preserving \u0026ldquo;wait and watch\u0026rdquo; (W\u0026amp;W) strategies. This approach circumvents the morbidity and functional deficits associated with total mesorectal excision. Patients in this cohort are closely monitored through rigorous follow-up involving serial endoscopic examinations, MRI, CT, and CEA levels [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. While this strategy enhances quality of life and mitigates surgical risks, it carries the inherent risk that approximately 20\u0026ndash;25% of patients with initial cCR will develop luminal recurrence, predominantly within the first two years. Therefore, early and accurate detection of residual or recurrent disease during surveillance is critical to ensure oncologic safety and facilitate timely salvage interventions.\u003c/p\u003e\u003cp\u003eDespite regular follow-up, endoscopic detection of recurrence remains challenging. Radiation-induced mucosal alterations including fibrosis, friability, and neovascularization can obscure visualization and simulate recurrent lesions, resulting in diagnostic ambiguity. These factors contribute to substantial inter-observer variability among endoscopists, potentially causing both over and underdiagnosis. Given these diagnostic challenges, there is a pressing need for objective, reliable, and reproducible tools to enhance detection accuracy in post nCRT surveillance.\u003c/p\u003e\u003cp\u003eArtificial intelligence (AI), particularly deep learning (DL), has recently shown significant promise in advancing medical image analysis. Among DL architectures, convolutional neural networks (CNNs) have proven especially proficient in interpreting visual data such as endoscopic and radiologic images. Within gastroenterology, CNN-based algorithms have been successfully applied to detect colonic polyps and to distinguish neoplastic from non-neoplastic lesions with notable accuracy [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eDespite these advances, the utilization of AI in the post nCRT surveillance of LARC remains insufficiently explored. Very few studies have focused on developing AI-driven models to detect residual or recurrent disease in patients exhibiting complete cCR, representing a notable gap in oncologic care [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. AI\u0026rsquo;s capability to detect subtle mucosal architectural changes often overlooked by human observers could markedly improve early detection of recurrence, thereby minimizing the need for invasive biopsies or additional imaging. Moreover, AI tools can provide consistent, scalable support as adjunct readers, augmenting clinical decision-making without supplanting expert judgment. The aim of this study is to develop a novel AI model for analysis of endoscopic images following complete clinical response in rectal cancer.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eStudy Design\u003c/p\u003e\u003cp\u003eThis retrospective, observational study was conducted at University Hospitals Liverpool and affiliated Clatterbridge Cancer Centre, tertiary referral centres specializing in the management of patients with LARC. Study population included individuals who were on W\u0026amp;W protocol following cCR to nCRT. Eligibility criteria for patient inclusion were aligned with standards established in previous multicentre studies, including those by [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e], to maintain methodological consistency. Inclusion criteria required patients to be over 18 years of age, have histologically confirmed rectal adenocarcinoma, have completed nCRT, and exhibit a cCR as determined by both MRI and endoscopy. Furthermore, availability of both pre-treatment and high-resolution post nCRT endoscopic images suitable for deep learning analysis was mandatory for inclusion. The study design is consistent with approaches previously adopted by researchers who employed temporally stratified cohorts to develop and validate deep learning models using surveillance imaging data [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e presents a visual pipeline in which endoscopic images of rectal cancer are processed through a convolutional neural network (CNN) model. The system is designed to automatically exclude suboptimal or inadequate frames, ensuring only high-quality inputs are analysed. The trained AI model then classifies the remaining frames to support clinical decision-making by distinguishing between post-treatment fibrotic scarring and true tumour recurrence, thereby enhancing diagnostic accuracy in the surveillance of patients following neoadjuvant therapy.\u003c/p\u003e\u003cp\u003eImage Acquisition and Preprocessing\u003c/p\u003e\u003cp\u003eHigh-resolution white-light proctosigmoidoscopy was employed to and regions of interest (ROI) were selected based on optimal mucosal visualization, excluding frames compromised by artefacts such as bleeding, motion blur, or poor focus. A standardized preprocessing pipeline was implemented to enhance image quality and consistency across the dataset. This included histogram equalization to normalize lighting, spatial rescaling to ensure uniform resolution, and the application of denoising filters to reduce background noise. Distinct from studies primarily adapting techniques from natural image processing, our methodology also incorporated targeted artefact removal and image stabilization procedures to further improve image fidelity for deep learning analysis.\u003c/p\u003e\u003cp\u003eData Annotation and Reference Standard\u003c/p\u003e\u003cp\u003eImage annotations were validated against histopathological findings, serving as the diagnostic gold standard. Frames corresponding to confirmed residual or recurrent malignancy were categorized as \u0026ldquo;tumour present,\u0026rdquo; while those obtained from patients who maintained a cCR without evidence of recurrence for a minimum follow-up period of 24 months were labelled as \u0026ldquo;no tumour.\u0026rdquo; This annotation strategy is consistent with the classification framework utilized in other CNNs [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. The annotated endoscopic image dataset was used for training and optimizing the deep learning\u0026ndash;based AI model at University of Porto.\u003c/p\u003e\u003cp\u003eDeep Learning Architecture\u003c/p\u003e\u003cp\u003eThe model architecture was built upon the Wide ResNet-101-2 framework, a deep residual network variant optimized for extracting complex features from high-dimensional medical imaging data. Initial pretraining on the ImageNet dataset enabled effective transfer learning, facilitating improved convergence and performance on the relatively limited endoscopic dataset, an approach consistent with established practices in AI-based endoscopic analysis [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. To enhance generalizability and reduce overfitting, the network incorporated batch normalization and dropout layers. Additionally, spatial attention mechanisms were integrated to focus on critical mucosal textures and vascular patterns relevant to the detection of residual or recurrent disease.\u003c/p\u003e\u003cp\u003eThe dataset was split at the subject level using a 90/10 ratio, allocating 87 patients (1,615 frames) to the training set and 10 patients (180 frames) to the test set. To prevent data leakage, patient-level separation was ensured, an approach consistent with the methodology used by [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. To enhance model generalizability, image augmentation techniques such as rotation, flipping, and brightness adjustment were applied, in line with practices adopted in multicentre endoscopic research.\u003c/p\u003e\u003cp\u003eTesting and Performance Evaluation\u003c/p\u003e\u003cp\u003eThe evaluation framework incorporated key performance metrics including sensitivity, specificity, accuracy, positive predictive value (PPV), negative predictive value (NPV), and area under the receiver operating characteristic curve (AUROC), with 95% confidence intervals estimated for AUROC. Confusion matrix analysis was conducted to assess misclassification rates, and statistical reliability was ensured using binomial exact tests [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e].\u003c/p\u003e\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eStatistical Analysis\u003c/h2\u003e\u003cp\u003eDescriptive statistics summarized the patient count, frame distribution per class, and the ratio of normal to abnormal cases. Primary classification metrics such as accuracy, sensitivity (recall), specificity, PPV, NPV, and AUROC were computed at the frame level by comparing model outputs against expert-annotated ground truth.\u003c/p\u003e\u003cp\u003eTo ensure statistical reliability, 95% confidence intervals (CIs) were estimated using a non-parametric bootstrapping approach with 1,000 iterations. A confusion matrix was constructed to visualize true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN), serving as the basis for calculating sensitivity, specificity, PPV and NPV.\u003c/p\u003e\u003cp\u003eThe statistical significance of the association between actual and predicted classifications was assessed using a chi-square test of independence, with a two-tailed p-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 considered significant. Additionally, the receiver operating characteristic (ROC) curve was plotted, and the AUROC was computed to evaluate the model\u0026rsquo;s discriminative capacity across varying thresholds. AUROC values between 0.8 and 0.9 were interpreted as indicative of excellent diagnostic performance. All analyses were performed using Python (v3.9).\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003eCohort Overview and Image Data Distribution\u003c/h2\u003e\u003cp\u003eBetween January 2020 and December 2024, 97 patients were classified to have a cCR after nCRT. From these, 1,795 annotated frames were collected. Twenty-four patients (363 frames) had histologically confirmed residual or recurrent disease, while 73 patients (1,432 frames) exhibited normal rectal mucosa indicative of a complete response.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eModel Performance Metrics\u003c/h3\u003e\n\u003cp\u003eFigure \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e demonstrate the model's strong ability to differentiate between normal mucosa and recurrent disease with high overall precision.\u003c/p\u003e\u003cp\u003eThe model achieved an accuracy of 92.8%, indicating that approximately 1,667 out of 1,795 frames were correctly classified. This metric reflects the model\u0026rsquo;s general effectiveness, even in the presence of class imbalance. Sensitivity was 80% indicating the model accurately flagged 4 out of every 5 disease frames. While some false negatives remain, this sensitivity level is considered clinically acceptable for surveillance purposes. The model exhibited a high specificity of 97.3%, which is particularly important in the context of surveillance, as it reduces false positives, thereby minimizing unnecessary procedures and patient distress.\u003c/p\u003e\u003cp\u003eA positive predictive value (PPV) of 90.9% indicates that 9 out of 10 frames classified as recurrent disease were truly positive, reducing the likelihood of overtreatment. Similarly, the negative predictive value (NPV) of 94.0% suggests that frames predicted as negative were reliably free of recurrence, supporting confident decisions to avoid further invasive investigation.\u003c/p\u003e\u003cp\u003eFinally, the area under the receiver operating characteristic curve (AUROC) was 0.886, reflecting excellent discriminative performance. AUROC values in the 0.8\u0026ndash;0.9 range are indicative of strong diagnostic capability, confirming the model\u0026rsquo;s ability to maintain a balanced trade-off between sensitivity and specificity. The high specificity further reinforces the model\u0026rsquo;s reliability in accurately identifying patients with true complete clinical response, thereby reducing the risk of overdiagnosis.\u003c/p\u003e\u003cp\u003eThe Receiver Operating Characteristic (ROC) curve serves as a critical tool for assessing the diagnostic performance of the AI model in differentiating residual or recurrent disease from normal mucosa during endoscopic surveillance of locally advanced rectal cancer. By plotting the true positive rate (sensitivity) against the false positive rate (1 \u0026ndash; specificity) across a range of classification thresholds, the ROC curve offers a comprehensive evaluation of the model\u0026rsquo;s discriminative ability. In this study, the model achieved an area under the ROC curve (AUROC) of 0.886, indicating a high level of diagnostic accuracy. This elevated AUROC underscores the model\u0026rsquo;s effectiveness in maintaining an optimal balance between sensitivity and specificity, even in complex cases where post-treatment mucosal alterations challenge visual interpretation. Accordingly, ROC analysis supports the model\u0026rsquo;s reliability in accurately distinguishing pathological from normal tissue, reinforcing its potential utility in enhancing clinical decision-making within the watch-and-wait surveillance framework.\u003c/p\u003e\n\u003ch3\u003eConfusion Matrix and Classification Breakdown\u003c/h3\u003e\n\u003cp\u003eThe confusion matrix analysis further elaborates the model\u0026rsquo;s performance at the frame level shown on Fig.\u0026nbsp;4.\u003c/p\u003e\u003cp\u003eThe AI model exhibited strong classification performance, as demonstrated by the confusion matrix outcomes. It accurately detected 290 frames containing residual or recurrent disease (true positives), reflecting its effectiveness in identifying cases requiring clinical intervention. However, 73 disease frames were misclassified as normal (false negatives), indicating some limitations in sensitivity where a subset of pathological cases went undetected. Additionally, the model misclassified 39 frames of normal mucosa as diseased (false positives), which could potentially result in unnecessary diagnostic follow-up. Notably, it correctly identified 1,393 frames of normal mucosa (true negatives), highlighting its high specificity and reliability in distinguishing healthy tissue. Collectively, these results suggest the model excels at differentiating between diseased and normal mucosa, with a particular strength in reducing false positives while maintaining reasonable sensitivity an essential balance for effective clinical decision-making [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003eStatistical Significance and Confidence Intervals\u003c/h2\u003e\u003cp\u003eTo assess the robustness of the AI model, 95% confidence intervals (CIs) were calculated for the primary performance metrics using bootstrapping techniques. The model achieved an accuracy of 92.8% (95% CI: 91.4\u0026ndash;94.1%), sensitivity of 80.0% (95% CI: 75.2\u0026ndash;84.6%), specificity of 97.3% (95% CI: 96.1\u0026ndash;98.4%), positive predictive value (PPV) of 90.9% (95% CI: 87.2\u0026ndash;94.6%), negative predictive value (NPV) of 94.0% (95% CI: 92.5\u0026ndash;95.5%), and an area under the receiver operating characteristic curve (AUROC) of 0.886 (95% CI: 0.863\u0026ndash;0.910). The narrow confidence intervals indicate high stability and reliability of the model\u0026rsquo;s performance. Furthermore, chi-square testing confirmed significant class discrimination ability (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), validating that the model\u0026rsquo;s predictive power is statistically significant and exceeds random classification.\u003c/p\u003e\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe performance metrics from our model indicate that the proposed AI framework holds significant potential as a supportive diagnostic tool during endoscopic surveillance for rectal cancer patients on W\u0026amp;W pathway. Timely and accurate identification of residual or recurrent lesions is crucial for initiating appropriate salvage treatments while avoiding unnecessary surgical procedures when no disease is present. The model\u0026rsquo;s high NPV of 94.0% is particularly meaningful in the context of watch-and-wait strategies, as it suggests that frames predicted as normal are indeed likely to be disease-free. This reduces the need for additional investigations, thereby minimizing patient and financial burden. Conversely, the PPV of 90.9% provides reassurance that when the AI flags a lesion as suspicious, it is very likely to be a true positive. This enhances clinical confidence in leveraging the AI model to support decision-making, particularly in borderline cases where visual assessments may be subjective or unclear. Moreover, the high specificity of 97.3% highlights the model\u0026rsquo;s strong ability to avoid false positives, which are common in traditional evaluations due to post-treatment changes such as inflammation, mucosal irregularities, or scarring. This further reinforces the AI tool's utility in improving diagnostic precision in endoscopic surveillance.\u003c/p\u003e\u003cp\u003eOur model\u0026rsquo;s higher accuracy (92.8%) and AUROC (0.886) indicate more precise discrimination of recurrence within a single-centre dataset. The current model\u0026rsquo;s performance sits between these extremes, suggesting strong internal validity with potential for broader generalisability.\u003c/p\u003e\u003cp\u003ePrevious studies have reported mixed outcomes in detecting rectal cancer recurrence following nCRT using conventional imaging techniques. Studies report considerable interobserver variability in post-treatment assessments and limitations of MRI in accurately distinguishing post-treatment scarring from residual tumour tissue [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. In contrast, the AI model developed in the present study demonstrated enhanced performance, particularly in interpreting endoscopic images, achieving greater accuracy and fewer false positives compared to conventional approaches [\u003cspan additionalcitationids=\"CR14 CR15\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e], as summarized in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\u003cp\u003eWhen compared with existing AI-based approaches for endoscopic surveillance in colorectal and rectal cancer, the current model shows clear improvements. Previous studies have reported variable sensitivity (70\u0026ndash;90%) and specificity (80\u0026ndash;95%) across models. In contrast, the present study's AUROC of 0.886, along with superior PPV and specificity, indicates stronger performance in detecting subtle post-nCRT mucosal abnormalities. These advancements may be attributed to the use of the WideResNet-101-2 architecture, which is well-suited for extracting deep, abstract features in complex post-treatment imaging data.\u003c/p\u003e\u003cp\u003eDespite encouraging results, the study has several limitations. The dataset, while enriched, originates from a single tertiary centre, potentially limiting external applicability. Additionally, the relatively small number of recurrence-positive cases (24 out of 97 patients) may restrict the model\u0026rsquo;s generalizability to broader and more diverse populations. The presence of 48 false negatives also presents a concern, emphasizing the importance of integrating additional diagnostic modalities, such as MRI and CEA levels, to support final clinical decisions. Analysis of false positives revealed that many incorrectly classified normal frames exhibited features such as scarring or radiation-induced mucosal changes, which closely resemble early recurrence. This underscores the need for incorporating complementary diagnostic modalities, such as MRI or histopathology, to improve diagnostic accuracy and confidence.\u003c/p\u003e\u003cp\u003eOngoing work focuses on incorporating temporal data from serial endoscopic exams to improve the model\u0026rsquo;s predictive consistency over time. Moreover, the inclusion of multimodal inputs, such as radiomics, serum biomarkers, and MRI findings\u0026mdash;is expected to enhance diagnostic accuracy beyond that of image-only system [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Additionally, despite strong overall accuracy and AUROC, the model\u0026rsquo;s generalizability across different clinical settings remains unproven. Variations in endoscopic equipment, lighting, mucosal staining techniques, and patient demographics could impact performance. Therefore, prospective validation across multiple centres is essential to confirm broader applicability.\u003c/p\u003e\u003cp\u003eBy delivering standardized, reproducible assessments, the model has the potential to improve the efficiency of endoscopic surveillance and reduce clinician workload. Early detection of recurrence supports timely therapeutic action, potentially improving clinical outcomes. In addition, widespread adoption of such technology may reduce healthcare expenditures by limiting unnecessary investigations. Collectively, this AI model offers a meaningful improvement to existing W\u0026amp;W surveillance protocols, supporting more effective and patient-centred management.\u003c/p\u003e\u003cp\u003eThe W\u0026amp;W approach, which emphasizes organ preservation in patients with cCR, depends on accurate and reproducible detection of recurrence. Studies have reported reported accuracy ranging from 53% to 90% and significant variability between observers using traditional assessment methods [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. The current CNN model\u0026rsquo;s combination of high specificity and acceptable sensitivity may provide a more standardized and dependable alternative, contributing to safer and more consistent surveillance workflows.\u003c/p\u003e\u003cp\u003eThe choice of the Wide ResNet-101-2 architecture, characterized by increased width rather than depth, facilitates deeper feature extraction and mitigates vanishing gradient issues. This architecture is particularly well-suited for capturing complex mucosal and vascular patterns in endoscopic imagery offering potential advantages over alternative frameworks such as EfficientNet [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] or DenseNet [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eFor effective clinical translation, future integration of the model into endoscopic systems with real-time inference capabilities, saliency map visualization, and user-friendly interfaces will be essential. The most effective deployment strategy involves augmenting, rather than replacing, the clinician's role. Future research should include prospective video-based validation studies and the incorporation of clinician feedback to facilitate clinical acceptance and workflow integration.\u003c/p\u003e\u003cp\u003eOn the other hand, despite satisfactory accuracy, the \u003cem\u003eblack-box\u003c/em\u003e nature of deep learning models reduces clinician confidence in model decisions, thereby decreasing their applicability. In fact, the development of models with increased complexity is often associated with a decreased clinician understanding of the reasons behind such decisions [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. In this context, explainable AI (XAI) aims to address this limitation by providing specific tools to help clinicians understand the reasons behind model outputs. Future studies will therefore focus on developing XAI tools for detecting specific regions of an image that may indicate residual lesions or recurrence. As validated in previous studies, the use of heatmaps or bounding boxes would be of great interest [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. These mechanisms would not only increase confidence in model decisions but also potentially improve the selection of biopsy sites, ultimately transforming patient outcomes.\u003c/p\u003e\u003cp\u003eIn conclusion, this tool has the potential to enhance early lesion detection, guide clinical decision-making, and increase opportunities for salvage, curative treatment strategies. To strengthen future deployment, efforts should focus on multicentre external validation, integration of multimodal inputs (e.g., MRI, serum biomarkers), and prospective clinical trials comparing standard care with AI-augmented workflows. Emphasis on explainability, through tools such as attention or saliency maps, as well as hardware optimization and regulatory compliance, will be vital to achieving safe and effective implementation in clinical practice [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e].\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eFunding: This study received no external funding.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eInstitutional Review Board Statement: This non-interventional study adhered to the Declaration of Helsinki\u0026rsquo;s ethical principles and received approval from the institutional review boards. To ensure data privacy, all potentially identifying information was removed, and each patient\u0026rsquo;s data was assigned a random, anonymized number before being incorporated into the dataset. A certified legal team with data protection expertise further guaranteed that the data remained untraceable, fully complying with General Data Protection Regulation (GDPR) standards.\u003c/p\u003e\n\u003cp\u003eInformed Consent Statement: Informed consent was waived as this was a retrospective non interventional study.\u003c/p\u003e\n\u003cp\u003eData Availability Statement: Additional data are available upon reasonable request.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eConflicts of Interest: The authors declare no conflicts of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConceptualization:\u0026nbsp;\u003c/strong\u003eMAJ (Muhammad Ahsan Javed), MM (Miguel Mascarenhas) and SA (Shakil Ahmed); \u003cstrong\u003eMethodology\u003c/strong\u003e: MM, FM (Francisco Mendes), EC (Eduardo Carvalho) and AS (Andr\u0026eacute; Santos); \u003cstrong\u003eFormal analysis and investigation\u003c/strong\u003e: FM, AS, EC, RR (Reshma Rajan), ZK (Zarnigar Khan) and IB (Iain Blake); \u003cstrong\u003eWriting - original draft preparation\u003c/strong\u003e: MAJ, MM, RR, FM; \u003cstrong\u003eWriting - review and editing\u003c/strong\u003e: all authors; \u003cstrong\u003eResources:\u003c/strong\u003e RR, IB, ZK; \u003cstrong\u003eSupervision:\u003c/strong\u003e MM, MAJ and SA\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments:\u0026nbsp;\u003c/strong\u003eWe would like to thank Professor Sun Myint and Dr Ngy Wah Than for contributing to images from Clatterbridge Cancer Centre and Mr Shay Willoughby for his support in this project\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eSung, H., Ferlay, J., Siegel, R. L., Laversanne, M., Soerjomataram, I., Jemal, A. and Bray, F. (2021) Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA: a Cancer Journal for Clinicians 71, 209\u0026ndash;249.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSauer, R., Becker, H., Hohenberger, W., R\u0026ouml;del, C., Wittekind, C., Fietkau, R., Martus, P., Tschmelitsch, J., Hager, E., Hess, C. F., et al. (2004) Preoperative versus Postoperative Chemoradiotherapy for Rectal Cancer. New England Journal of Medicine 351, 1731\u0026ndash;1740.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMaas, M., Nelemans, P. J., Valentini, V., Das, P., R\u0026ouml;del, C., Kuo, L.-J., Calvo, F. A., Garc\u0026iacute;a-Aguilar, J., Glynne-Jones, R., Haustermans, K., et al. (2010) Long-term outcome in patients with a pathological complete response after chemoradiation for rectal cancer: a pooled analysis of individual patient data. The Lancet Oncology 11, 835\u0026ndash;844.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHabr-Gama, A., Perez, R. O., Nadalin, W., Sabbaga, J., Ribeiro, U., Silva e Sousa, A. H., Campos, F. G., Kiss, D. R. and Gama-Rodrigues, J. (2004) Operative Versus Nonoperative Treatment for Stage 0 Distal Rectal Cancer Following Chemoradiation Therapy. Annals of Surgery 240, 711\u0026ndash;718.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFokas, E., Appelt, A., et al. (2021) International consensus recommendations on key outcome measures for organ preservation after (chemo)radiotherapy in patients with rectal cancer. Nat Rev Clin Oncol 18, 805\u0026ndash;816.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSpadaccini, M., Massimi, D., Mori, Y., Ludovico Alfarone, Alessandro Fugazza, Maselli, R., Sharma, P., Facciorusso, A., Hassan, C. and Repici, A. (2023) Artificial Intelligence-Aided Endoscopy and Colorectal Cancer Screening. Diagnostics, Multidisciplinary Digital Publishing Institute 13, 1102\u0026ndash;1102.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang, J., Liu, R., Wang, X., Zhang, S., Shao, L., Liu, J., Zhao, J., Wang, Q., Tian, J. and Lu, Y. (2024) Deep learning model based on endoscopic images predicting treatment response in locally advanced rectal cancer undergo neoadjuvant chemoradiotherapy: a multicenter study. Journal of Cancer Research and Clinical Oncology 150.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShen, H., Jin, Z., Chen, Q., Zhang, L., You, J., Zhang, S. and Zhang, B. (2024) Image-based artificial intelligence for the prediction of pathological complete response to neoadjuvant chemoradiotherapy in patients with rectal cancer: a systematic review and meta-analysis. La radiologia medica, Springer Science\u0026thinsp;+\u0026thinsp;Business Media 129, 598\u0026ndash;614.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSarwinda, D., Paradisa, R. H., Bustamam, A. and Anggia, P. (2021) Deep Learning in Image Classification using Residual Network (ResNet) Variants for Detection of Colorectal Cancer. Procedia Computer Science 179, 423\u0026ndash;431.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLetizia Monguzzi, Ippolito, D., Davide Paolo Bernasconi, Chiara Trattenero, Galimberti, S. and Sandro Sironi. (2013) Locally advanced rectal cancer: Value of ADC mapping in prediction of tumor response to radiochemotherapy. European Journal of Radiology, Elsevier BV 82, 234\u0026ndash;240.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhou, X., Dai, J., Lu, Y., Zhao, Q., Liu, Y., Wang, C., Zhao, Z., Wang, C., Gao, Z., Yu, Y., et al. (2024) Whole slide image based prognosis prediction in rectal cancer using unsupervised artificial intelligence. BMC Cancer, Springer Science and Business Media LLC 24.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHorvat, N., Harini Veeraraghavan, Rizkallah, S., David, de, F., Zheng, J., Marinela Capanu, Fuqua, J. L., Maria Clara Fernandes, Sosa, R. E., et al. (2022) Combined artificial intelligence and radiologist model for predicting rectal cancer treatment response from magnetic resonance imaging: an external validation study. Abdominal Radiology, Springer Nature 47, 2770\u0026ndash;2782\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKim, M., Park, T., Oh, B. Y., Kim, M. J., Cho, B.-J. and Son, I. T. (2024) Performance reporting design in artificial intelligence studies using image-based TNM staging and prognostic parameters in rectal cancer: a systematic review. Annals of Coloproctology 40, 13\u0026ndash;26.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eİsmail Mendi, B., Kose, K., Fleshner, L., Adam, R., Safai, B., Farabi, B. and Atak, M. F. (2024) Artificial Intelligence in the Non-Invasive Detection of Melanoma. Life, MDPI AG 14, 1602.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWilliams, H., Thompson, H. M., Lee, C., Rangnekar, A., Gomez, J. T., Widmar, M., Wei, I. H., Pappou, E. P., Nash, G. M., Weiser, M. R., et al. (2024) Assessing Endoscopic Response in Locally Advanced Rectal Cancer Treated with Total Neoadjuvant Therapy: Development and Validation of a Highly Accurate Convolutional Neural Network. Annals of Surgical Oncology, Springer Science and Business Media LLC 31, 6443\u0026ndash;6451.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang, A., Mo, J., Zhong, C., Wu, S., Wei, S., Tu, B., Liu, C., Chen, D., Xu, Q., Cai, M., et al. (2021) Artificial intelligence-assisted detection and classification of colorectal polyps under colonoscopy: a systematic review and meta-analysis. Annals of Translational Medicine, AME Publishing Company 9, 1662\u0026ndash;1662.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEsteva, A., Kuprel, B., Novoa, R. A., Ko, J., Swetter, S. M., Blau, H. M., \u0026amp; Thrun, S. (2017). Dermatologist-level classification of skin cancer with deep neural networks. Nature, 542(7639), 115\u0026ndash;118.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSmith, S.C., Siau, K., Cannatelli, R., Antonelli, G., Shivaji, U.N., Ghosh, S., Saltzman, J.R., Hassan, C. and Iacucci, M., 2021. Training methods in optical diagnosis and characterization of colorectal polyps: a systematic review and meta-analysis. Endoscopy international open, 9(05), pp.E716-E726.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang, Y., Zhang, X., Wu, Q., Gu, C. and Wang, Z., 2021. Artificial intelligence-aided colonoscopy for polyp detection: a systematic review and meta-analysis of randomized clinical trials. Journal of Laparoendoscopic \u0026amp; Advanced Surgical Techniques, 31(10), pp.1143\u0026ndash;1149.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang, Y., He, K., Guo, Y., Liu, X., Yang, Q., Zhang, C., Xie, Y., Mu, S., Guo, Y., Fu, Y., et al. (2020) A Novel Multimodal Radiomics Model for Preoperative Prediction of Lymphovascular Invasion in Rectal Cancer. Frontiers in Oncology 10.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMaas, M., Beets-Tan, R. G. H., Lambregts, D. M. J., Lammering, G., Nelemans, P. J., Engelen, S. M. E., van Dam, R. M., Jansen, R. L. H., Sosef, M., Leijtens, J. W. A., et al. (2011) Wait-and-See Policy for Clinical Complete Responders After Chemoradiation for Rectal Cancer. Journal of Clinical Oncology 29, 4633\u0026ndash;4640\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAbd El-Ghany, S., Mahmood, M. A. and Abd El-Aziz, A. A. (2024) Adaptive Dynamic Learning Rate Optimization Technique for Colorectal Cancer Diagnosis Based on Histopathological Image Using EfficientNet-B0 Deep Learning Model. Electronics 13, 3126.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHasan, M., Islam, J., Ahmed, M. and Hasan, M. M. (2023) Prediction of Colon Cancer using DenseNet121, CNN, and REsNET50 Machine Learning Models and using Image Processing Techniques. 2023 International Conference on Artificial Intelligence Robotics, Signal and Image Processing (AIRoSIP), IEEE 296\u0026ndash;301.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMascarenhas M, Mendes F, Martins M, et al. Explainable AI in Digestive Healthcare and Gastrointestinal Endoscopy. J Clin Med. 2025;14(2).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMascarenhas M, Mendes F, Fonseca F, et al. A Novel Deep Learning Model for Predicting Colorectal Anastomotic Leakage: A Pioneer Multicenter Transatlantic Study. Journal of Clinical Medicine. 2025;14(15):5462.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eStanzione, A., Verde, F., Romeo, V., Boccadifuoco, F., Mainenti, P. P. and Maurea, S. (2021) Radiomics and machine learning applications in rectal cancer: Current update and future perspectives. World Journal of Gastroenterology 27, 5306\u0026ndash;5321.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003e\u003cem\u003eKey components of the deep learning model architecture and training parameters used in the study. The Wide ResNet-101-2 model was trained on resized endoscopic images (224 \u0026times; 224 pixels) using categorical cross-entropy loss. Regularization techniques included dropout and batch normalization. Data augmentation strategies were applied to enhance generalization. A patient-level 90:10 training and validation split was used, with early stopping triggered after 10 consecutive validation epochs without improvement.\u003c/em\u003e\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eComponent\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eDescription\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eModel Architecture\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWide ResNet-101-2\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInput Image Dimensions\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e224 \u0026times; 224 pixels (resized)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLoss Function\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCategorical Cross-Entropy\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRegularization\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDropout (rate\u0026thinsp;=\u0026thinsp;0.3), Batch Normalization\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eData Augmentation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRotation, horizontal/vertical flip, brightness/contrast shift\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEpochs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e50 (with early stopping if no improvement in 10 consecutive validation epochs)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTraining\u0026ndash;Validation\u0026ndash;Test Split\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e90/10% patient level split\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab2\" border=\"1\" class=\"fr-table-selection-hover\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003e\u003cem\u003eBenchmark Comparison with Published AI Studies.\u003c/em\u003e\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eStudy \u0026amp; Year\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eMethod\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eModality\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAUROC\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSensitivity\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEsteva et al. [\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e]\u003c/p\u003e\n \u003cp\u003e(2017)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDeep CNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDermoscopy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.911\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e89.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e76.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e94.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSmith et al. [\u003cspan class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/p\u003e\n \u003cp\u003e(2021)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDL Endoscopy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAnal neoplasia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.793\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e85.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e70.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e90.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eZhang et al. [\u003cspan class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/p\u003e\n \u003cp\u003e(2021)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAI - Colonoscopy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCRC detection\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.831\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e88.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e75.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e91.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eCurrent Study\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(2025)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWide ResNet-101-2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eProcto-\u003c/p\u003e\n \u003cp\u003esigmoidoscopy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.886\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u003cstrong\u003e92.8%\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u003cstrong\u003e80.0%\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u003cstrong\u003e97.3%\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"techniques-in-coloproctology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"tcol","sideBox":"Learn more about [Techniques in Coloproctology](http://link.springer.com/journal/10151)","snPcode":"10151","submissionUrl":"https://submission.nature.com/new-submission/10151/3","title":"Techniques in Coloproctology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Rectal Cancer, Complete Clinical response, Wait \u0026 Watch and Artificial Intelligence","lastPublishedDoi":"10.21203/rs.3.rs-7775722/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7775722/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e\u003cp\u003ePatients with locally advanced rectal cancer (LARC) who have a complete clinical response (cCR) to neoadjuvant chemoradiotherapy (nCRT) may opt for organ preservation, wait and watch \u0026ndash; (W\u0026amp;W). This consists of an intense surveillance program including serial endoscopies, pelvic MRI, CEA, and CT scans to detect recurrent disease at an early stage. However, identifying residual or recurrent lesions endoscopically in these cases can be challenging due to mucosal changes such as friability and neovascularization. We developed a novel deep learning model to assist in the detection of residual or recurrent rectal cancer lesions during proctosigmoidoscopy.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e\u003cp\u003eWe trained a convolutional neural network (Wide ResNet-101-2) on a dataset of 1,795 annotated frames from proctosigmoidoscopy exams of 97 patients treated at a tertiary referral centre. Residual or recurrent disease was defined by histopathological confirmation. The dataset was split into training and testing cohorts using a 90/10% patient-level split.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eOut of 97 patients, 24 (363 frames) had confirmed residual or recurrent disease, while 73 (1,432 frames) presented normal rectal mucosa. The model achieved an overall accuracy of 92.8%, with a sensitivity of 80.0%, specificity of 97.3%, PPV of 90.9%, NPV of 94.0%, and an AUROC of 0.886.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e\u003cp\u003eTo the best of our knowledge, this is the first deep learning model specifically developed for the detection of residual or recurrent disease following in W\u0026amp;W patients during endoscopic examination. This tool has the potential to enhance early lesion detection, guide clinical decision-making, and increase opportunities for salvage, curative treatment strategies.\u003c/p\u003e","manuscriptTitle":"Use of Artificial Intelligence in Analysis of Endoscopic Images Following Complete Clinical Response in Rectal Cancer","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-23 14:58:52","doi":"10.21203/rs.3.rs-7775722/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-11-14T14:11:47+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-10T15:42:22+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-19T19:06:46+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"172034825870725453573032678645706890541","date":"2025-10-11T16:50:18+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"124377330910926584332041143869440489508","date":"2025-10-09T16:02:09+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-10-09T15:34:49+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-10-09T15:32:32+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-10-04T06:18:50+00:00","index":"","fulltext":""},{"type":"submitted","content":"Techniques in Coloproctology","date":"2025-10-03T18:29:04+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"techniques-in-coloproctology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"tcol","sideBox":"Learn more about [Techniques in Coloproctology](http://link.springer.com/journal/10151)","snPcode":"10151","submissionUrl":"https://submission.nature.com/new-submission/10151/3","title":"Techniques in Coloproctology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"b34958cd-2b65-45d6-8e28-01941b2b80ab","owner":[],"postedDate":"October 23rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-03-31T18:39:08+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-23 14:58:52","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7775722","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7775722","identity":"rs-7775722","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}