Endoscopic Image-based Radiomics Classifiers for the Prediction of Neoadjuvant Chemoradiotherapy Response in Rectal Cancer Patients

Endoscopic Image-based Radiomics Classifiers for the Prediction of Neoadjuvant Chemoradiotherapy Response in Rectal Cancer Patients | 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 Endoscopic Image-based Radiomics Classifiers for the Prediction of Neoadjuvant Chemoradiotherapy Response in Rectal Cancer Patients Tsuyoshi Ozawa, Yusuke Saikawa, Tamuro Hayama, Keijiro Nozawa, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-1717256/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Objective The accuracy of identifying a CRT-response by a pre-operative radiological examination is limited, and another approach is necessary. We constructed endoscopic image-based radiomics classifiers to predict the response of locally advanced rectal cancers (LARCs) to neoadjuvant chemotherapy (NA-CRT). Design: We enrolled 90 patients who had undergone NA-CRT followed by surgery. We selected 5,255 pre- and post-CRT endoscopic images of the tumors and extracted 860 texture features from each image. Using the extracted texture features, we applied 12 machine learning models to construct radiomics classifiers. The performances of the radiomics classifiers to predict a pathological complete response (pCR) and a CRT-response (good or not) were evaluated with a double cross-validation technique by calculating the area under the receiver operating characteristics curve (AUROC). Results In the prediction of pCR using pre-CRT images, the cubic support vector machine (SVM) model showed the highest AUROC of 0.816 in the image-basis analysis, and the AUROC of 0.904 in the patient-basis analysis. In the prediction of pCR using post-CRT images, the quadratic SVM model showed the highest AUROC (0.839) in the image-basis analysis, and the AUROC of 0.958 in the patient-basis analysis. The Gaussian SVM model demonstrated the best performances to predict a CRT-response in both pre- and post-CRT images with AUROCs of 0.792 and 0.841 in the image-basis analysis and 0.857 and 0.950 in the patient-basis analysis, respectively. Conclusion Our endoscopic image-based radiomics classifiers demonstrated robust performances to predict the CRT-response of LARCs and may contribute to the patients' management. Chemoradiotherapy (CRT) Endoscopy Machine learning Radiomics Rectal cancer Figures Figure 1 Figure 2 Figure 3 Introduction Approximately 25% of colorectal cancers (CRCs) are located in the rectum, and previous studies have described genetic differences between colon cancers and rectal cancers ( 1 ). The anatomical location of the rectum also yields clinical differences between these cancers; surgeries for rectal cancer, especially lower advanced rectal cancer (LARC), are technically difficult compared to those for colon cancers, and this difficulty leads to the frequent local recurrence of rectal cancers ( 2 ). To address this issue, the standard treatment of LARC in Western countries is neoadjuvant chemoradiotherapy (NA-CRT) followed by surgery ( 3 ). NA-CRT has the potential to achieve down-staging of the tumor, which may reduce the risk of tumor-cell shedding during surgery and/or a positive surgical margin, thus improving the local control of the tumor. Roughly 15–25% of patients with LARC have demonstrated a pathological complete response (pCR) after NA-CRT ( 4 – 6 ). Theoretically, surgery can be avoided for patients with rectal cancers who achieve a pCR after NA-CRT, making it possible to preserve the organs. The "watch-and-wait" strategy is thus one of the available treatment strategies; i.e., when the post-CRT examinations demonstrate no evidence of residual tumor, close follow-up is performed without surgery until tumor recurrence becomes apparent ( 7 , 8 ). However, the accuracy of identifying a pCR by a pre-operative radiological examination is limited, and another approach is necessary to improve the results of the watch-and-wait strategy ( 9 – 12 ). Moreover, NA-CRT may cause severe side effects without having an impact on tumor progression, and some tumors even progress during NA-CRT ( 13 , 14 ). It is thus desirable to predict whether a tumor will or will not respond to NA-CRT. The performance of radiomics has greatly improved recently due to the development of computational methods. Radiomics extract various types of quantitative features of regions of interest (ROIs) from digital images and enable the determination of the relationships between the extracted features and underlying pathophysiological condition of the ROIs( 15 , 16 ). Radiomics may make it possible to evaluate the quality of ROIs systematically and objectively, and radiomics thus have a great potential to surpass human performance ( 17 ). Previous studies demonstrated the robust performance of radiomics for predicting lymph node metastases, chemotherapy response or even mutational status of colorectal cancers based on computed tomography (CT) images ( 18 – 20 ). Regarding the prediction and evaluation of LARCs' responses to NA-CRT, several studies evaluated the performance of radiomics analyses using CT images, magnetic resonance images (MRIs), or positron emission tomography (PET) images ( 21 – 27 ). The results of these studies indicated a promising role of radiomics in the prediction of non-responders before NA-CRT and in the identification of pCRs after NA-CRT. However, the identification of the ROIs can be very difficult and subjective in these radiological examinations, especially after NA-CRT, and this may cause inter-observer differences. In the present study, we used endoscopic images of rectal cancers, as it is fairly easy to recognize the ROI in these cancers and the results may have higher reproducibility, which is an important factor for biomarker development. To our knowledge, this is the first study to construct and evaluate the performance of radiomics using endoscopic images of LARCs. Materials And Methods Patients We retrospectively reviewed the clinicopathological data of 140 patients with middle and low LARCs who had undergone NA-CRT followed by surgery at Teikyo University Hospital from Oct. 2007 to Dec. 2019. The exclusion criteria were ( 1 ) non-adenocarcinoma (n = 8), ( 2 ) anal fistula-related cancer (n = 3), ( 3 ) Inflammatory bowel disease associated cancer (n = 1), ( 4 ) patients who developed distant metastases during NA-CRT (n = 7), ( 5 ) a positive surgical margin (n = 5), ( 6 ) patients with synchronous cancer in the other organs (n = 1), ( 7 ) patients whose colonoscopy images were not available (n = 15), and ( 8 ) patients whose NA-CRT was incomplete (n = 10). A final total of 90 patients were included in the analysis (Fig. 1 ). All patient information was de-identified prior to the data analyses to maintain patient anonymity. Patients' written informed consent for their data to be used was obtained, and this study was approved by the Teikyo University Ethics Committee (No. 19–127). The study protocol conforms to the ethical guidelines of the 1975 Declaration of Helsinki and its later amendments. Neoadjuvant chemoradiotherapy and concomitant surgery NA-CRT was performed as described ( 28 ). In summary, the total dose of radiotherapy was 50.5 Gy or 45 Gy, which was given in 28 fractions or 25 fractions. Treatment planning was done with CT scans, and the target volume included the primary tumor, the anus, and lymph nodes in the mesorectum and in the pelvis (lateral lymph nodes). As the chemotherapy regimen, fluoropyrimidine (tegafur-uracil with leucovorin, capecitabine, S-1) alone, S-1 with oxaliplatin, or S-1 with irinotecan were induced concomitantly with radiotherapy. We defined NA-CRT-completion as cases in which both ≥ 80% of scheduled chemotherapy and the radiotherapy dose of ≥ 45 Gy were used; the other cases were recognized as incomplete-CRT ( 14 ). The surgical procedure in each of the eligible cases was performed at approx. 8 weeks after the completion of NA-CRT. A total mesorectal excision (TME) with regional lymph node dissection was performed under laparotomy or laparoscopically. All of the resected specimens were evaluated by the institutional pathologists, and the TNM classification and CRT grade were determined according to the Japanese Society for Cancer of the Colon and Rectum guidelines (ninth edition) as follows. Grade 0: No evidence of tumor response. Grade 1: A < 2/3 regression of the tumor cells. Grade 2: A ≥ 2/3 regression of the tumor cells. Grade 3: Complete regression. Grades 2 and 3 were regarded as a good response, and a pCR was defined as Grade 3 ( 29 ). Image preparation for training and validation sets Each colonoscopy was performed using standard endoscope equipment (Evis Lucera and PCF type Q260AI and H290I, Olympus Medical Systems, Tokyo). Six weeks after finishing NA-CRT, the second-look colonoscopy was performed. All of the patients' endoscopic images were extracted and reviewed by two board-certified gastroenterologists (T.O and T.H). Only the non-magnified images including primary tumors observed using conventional white-light were selected in the analyses. Pre-CRT images were available from 88 of the 90 patients and included 2,474 images in total. Post-CRT images were available from all 90 patients and included 2,781 images in total (Fig. 1 ). The primary tumors were then marked manually using ImageJ software (imagej.nih.gov/ij/download/), and ROIs were created. Radiomics feature extraction and the development of radiomics classifiers using machine learning models First, the ROIs were converted to grayscale using the grb2 gray function of MATLAB software (MathWorks, Natick, MA, U.S.), and we evaluated the texture features based on the distribution of the intensity of each pixel value. Five different types of texture features were calculated using the Vallières radiomics toolbox of MATLAB software ( 30 ): ( 1 ) the gray level histogram which includes three statistics, ( 2 ) the gray level co-occurrence matrix which includes nine statistics, ( 3 ) the gray level run-length matrix which includes 13 statistics, ( 4 ) the gray level size-zone matrix which includes 13 statistics, and ( 5 ) the neighborhood gray-tone difference matrix which includes five statistics. A total of 43 texture features were thus evaluated ( Suppl. Table S1 ). To amplify the texture features, each texture feature was extracted using 20 different parameters based on previous feature extraction models, resulting in a total of 860 features per image. Then, in order to avoid multicollinearity, we used a bootstrap procedure (bootstrap sample = 1000) based on Spearman's rho to select the most suitable parameters for each of the 43 texture features. Finally, each texture feature was normalized through a z-transform. Using the extracted texture features, we developed radiomics classifiers by applying 12 different machine learning models in the Statistics and Machine Learning toolbox of MATLAB: the linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), linear support vector machine (SVM), cubic SVM, Gaussian SVM, quadratic SVM, k-nearest neighbor algorithm (KNN), cubic KNN, cosine KNN, weighted KNN, logistic regression model, and decision trees. The pre-CRT images and post-CRT images were subjected to each of these models separately, and pre-CRT image-based radiomics classifiers and post-CRT image-based radiomics classifiers were developed (Fig. 2 ). Evaluation of the performances of the radiomics classifiers We evaluated the performance of each radiomics classifier to predict a good response and a pCR by using a 10-fold double cross-validation technique ( Suppl. Fig. S1 ). The receiver operating characteristics curve (ROC) was determined and the area under the ROC (AUROC) was calculated for each radiomics classifier. Because each tumor provided multiple images, the performance of our radiomics was evaluated on both an image basis and a patient basis. To evaluate the per-patient performances of the radiomics classifiers, we used the average score of the images from each patient. Cut-off values for the radiomics classifiers were decided using Youden's index, and the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy were calculated. In the comparison of clinicopathological features, the χ 2 test and Fisher's exact test were used for categorical data, and the Wilcoxon rank sum test was used for un-paired continuous variables. All statistics were performed using MATLAB (MathWorks) and JMP pro 15 (SAS Institute Japan, Tokyo). Results The performances of the radiomics classifiers for predicting a pCR and response to NA-CRT on the image basis Sixty-six patients (73%) underwent fluoropyrimidine (FP)-based NA-CRT, 21 patients (23%) underwent S-1 + oxaliplatin based NA-CRT, and three patients (3%) underwent S-1 + irinotecan-based NA-CRT. The average interval between the end of the day of NA-CRT to the second-look endoscopy and to the surgery were 42 days (SD 13) and 55 (SD 13) days, respectively. The pathological findings demonstrated the CRT-grades 0 or 1 in 43 patients (48%), CRT-grade 2 in 37 patients (41%), and CRT-grade 3 in 10 patients (11%). The detailed clinicopathological features are summarized in Table 1 . Table 1 Clinicopahtological features of the patients included in the study Characteristics N (%) Gender Male / Female 71 (79) / 19 (21) Age (Years) Mean (SD) 61.9 (9.8) Distance from AV (cm) 5 59 (66) / 31 (34) cStage 2 / 3 29 (32) / 61 (68) Chemotherapy FP / S-1+OX / S-1+IRI 66 (73) / 21 (23) / 3 (3) Radiation dose (Gy) 45 / 50.4 3 (3) / 87 (97) ypStage 0-1 / 2 / 3 37 (41) / 26 (29) / 27 (30) Histology Well-Mod / Por-Muc / unavailable 81 (90) / 8 (9) / 1 (1) Pre-CRT CEA (ng/ml) 5 52 (58) / 38 (42) CRT grade 0-1 / 2 / 3 43 (48) / 37 (41) / 10 (11) AV: anal verge, FP: fluoropyrimidine, OX: oxaliplatin, IRI: irinotecan, CRT: chemoradiotherapyCEA: carcinoembryonic antigen The numbers of pre-CRT images of CRT grade 0 or 1, grade 2, and grade 3 were 1,151 (47%), 1,098 (44%), and 225 (9%), respectively. The numbers of post-CRT images of CRT grade 0 or 1, grade 2, and grade 3 were 1,211 (44%), 1,302 (47%), and 268 (9%), respectively (Table 2 ). The results of our comparison of the clinicopathological features between CRT grades 0–2 and CRT grade 3 and between CRT grades 0–1 and CRT grades 2–3 are shown in Supplementary Table S2 . Table 2 The detail of the images included in the study Pre or Post CRT No. of patients CRT grade No. of the images (N, %) Pre 88 0–1 2 3 1151 (47) 1098 (44) 225 (9) Post 90 0–1 2 3 1211 (44) 1302 (47) 268 (9) Figure 3 A is a heatmap showing the per-image performances of each radiomics classifier to predict a pCR using pre- or post-CRT images. The AUCs ranged from 0.529 to 0.816 with the pre-CRT images, and the best radiomics classifier was the cubic SVM model with 84% accuracy (Table 3 ). With the post-CRT images, the AUROCs ranged from 0.619 to 0.839, and the best radiomics classifier was the quadratic SVM model with 83% accuracy (Table 3 ). Table 3 The performances of each classifier to predict pCR and CRT-response Prediction Image Classifier Evaluation SEN SPE PPV NPV ACU pCR Pre-CRT Cubic SVM Per image 63% 87% 32% 96% 84% Per patient 70% 100% 100% 96% 97% Post-CRT Quadratic SVM Per image 71% 84% 33% 96% 83% Per patient 90% 96% 75% 99% 96% CRT- response Pre-CRT Gaussian SVM Per image 71% 74% 76% 69% 73% Per patient 74% 88% 87% 76% 81% Post-CRT Gaussian SVM Per image 84% 71% 79% 77% 78% Per patient 96% 81% 85% 96% 89% pCR: pathological complete response, CRT: chemoradiotherapy, SEN: sensitivity, SPE: specificity, PPV: positive predictive value, NPV: negative predictive value, ACU: accuracy Figure 3 B is a heatmap demonstrating the per-image performances of each radiomics classifier to predict a CRT response (good responder or not) using pre- or post-CRT images. The AUROCs ranged from 0.679 to 0.792 with the pre-CRT images, and the best radiomics classifier was the Gaussian SVM model with 73% accuracy (Table 3 ). The AUROCs ranged from 0.687 to 0.841 with the post-CRT images, and the best radiomics classifier was the Gaussian SVM model with 78% accuracy (Table 3 ). The performances of the radiomics classifiers for predicting a pCR and response to NA-CRT on the patient basis We then evaluated the per-patient performance of each of the best per-image classifiers described above. The per-patient performance of the cubic SVM model to predict a pCR using pre-CRT images was excellent with the AUROC of 0.904 (Fig. 3 C) and 97% accuracy (Table 3 ). The per-patient performance of the quadratic SVM model to predict a pCR using post-CRT images showed the AUROC of 0.958 (Fig. 3 D) and 96% accuracy (Table 3 ). The Gaussian SVM model demonstrated high performances to predict a CRT response in both pre-CRT images and post-CRT images, with the AUROC of 0.857 and 80% accuracy, and the AUROC of 0.950 and 89% accuracy, respectively (Fig. 3 E, F, Table 3 ) For pre-CRT radiomics classifiers, the association between the scores of these radiomics classifiers and pre-CRT T-stage and pre-CRT occupation rate of rectal circumference were evaluated, however, there was no statistical association between them ( Suppl. Fig. S2A, B ). For post-CRT radiomics classifiers, the association between the scores of these radiomics classifiers and pathological T-stage and pathological tumor diameter were evaluated. The post-CRT quadratic SVM model was significantly associated with pathological T-stage (p < 0.0001), and the post-CRT Gaussian SVM model was significantly associated with tumor diameter (p = 0.045) ( Suppl. Fig. S2C, D ). Discussion We have developed and validated radiomics classifiers to predict a pCR and CRT response using pre-CRT and post-CRT endoscopic images. Our radiomics models showed robust performances in predicting a pCR and CRT response with both pre- and post-CRT images. These models are easy to use and effective for decision-making regarding the treatment of rectal cancer patients. The prediction of CRT-response before NA-CRT in particular may make it possible to reduce the unnecessary NA-CRT, and the prediction of a pCR after NA-CRT may contribute to the watch-and-wait strategy. In this regard, our radiomics classifiers demonstrated excellent ability to predict the CRT-response using pre-CRT images with 80% accuracy, and to predict a pCR using post-CRT images with 96% accuracy on a patient basis. Radiomics is used to convert medical images into high-dimensional, mineable data by extracting quantitative features and analyzing the extracted features for decision support ( 15 ). The radiomics method is gradually gaining attention with the recent advances in pattern recognition tools. Other research groups constructed radiomics models to predict a pCR or CRT response using CT, MRI, or PET data ( 21 – 24 , 27 ). Most of those studies showed an AUC of ~ 0.85 to predict a pCR or CRT response. Among the studies, Liu et al. reported that their radiomics signature comprised of 30 selected features in pre- and post-MRI images showed the AUC of 0.9799 to predict a pCR in an independent cohort of 70 rectal cancer patients ( 21 ). However, it is usually difficult to accurately detect the tumor location in MR images, especially after CRT. Liu et al. thus placed the ROI on the primary tumor bed region corresponding to that observed pre-CRT when the tumor bed region could not be detected on post-CRT MRI. This method may be complicated and can also be subjective. In contrast, endoscopy shows the location of the tumor bed fairly clearly, and it is much easier to place ROIs with endoscopic help, leading to objective and reproducible results. Moreover, our radiomics classifiers evaluated surface pattern of the tumors. Pre-CRT radiomics classifiers were independent of conventional T stage or occupation rate of the tumor in the rectum. These findings suggest that the surface pattern of the tumor may include various information of the tumor characters. Thus, it is also intriguing to construct a radiomics classifier using new endoscopic technologies such as narrow band imaging or magnifying endoscopy( 31 ). Specific endoscopic findings of a pCR have been reported in several studies ( 9 , 12 , 32 ). Kawai et al. reported that flattened marginal swelling was an independent and strong predictor of a pCR with 69.1% sensitivity and 73.9% specificity ( 12 ). This endoscopic finding provided better specificity but lower sensitivity compared to CT assessment. Combining these methods will thus increase the prediction ability. Future studies that incorporate various types of radiomics features including CT, MRI, PET, and endoscopy are very promising ( 30 , 33 , 34 ). We constructed our models using conventional machine learning models. Bibault et al. used a deep-learning method to construct a radiomics signature using the T-stage and 28 radiomics features, and the results demonstrated that the model could predict a pCR with 80% accuracy, which is better accuracy than that provided by a conventional SVM model ( 35 ). It could thus be intriguing to construct a model using a deep-learning method and radiomics features. Image recognition using a deep-learning method is also a promising method. A convolutional neural network (CNN) is a one of the deep-learning based architectures that is suitable for image interpretation tasks, and it has the potential to be superior to the conventional machine-learning techniques. Skrede et al. recently constructed a new CNN-based biomarker to predict the outcomes of CR that uses digitally scanned conventional hematoxylin and eosin-stained tumor tissue sections, and they demonstrated promising results ( 17 ). Their findings indicated that the deep-learning classification was superior to the conventional pathological assessment. We acknowledge several limitations in our study. It was a retrospective cohort study, and a large prospective study is necessary to validate the performance of our radiomics. In addition, we used a double cross-validation technique to evaluate the performance of our radiomics classifiers because of the limitation of the number of patients and images included in the study; external validation is necessary in future studies. In conclusion, we have constructed and validated radiomics classifiers to predict a pCR and CRT response using endoscopic images and a machine-learning method in rectal cancer patients who had undergone NA-CRT. Our radiomics classifiers demonstrated robust performances and may contribute to the decision-making regarding the management of patients with LARCs in clinical settings. Declarations Acknowledgement: None Funding information: This study was supported by Grant-in-Aid for Young Scientists from Japan Society for Promotion of Science (JSPS KAKENHI Grant Number JP19K16810) Conflict of interest : The authors declare no conflict of interes Ethics: This study was approved by the Teikyo University Ethics Committee (No. 19-127). Author contributions: T.O was involved in study concept and design, analysis and interpretation of data and drafting of the manuscript. Y.S was involved in analysis and interpretation of data. T.H was involved in acquisition of data. K.M, and K.N were involved in critical revision of the manuscript for important intellectual content and material support. I. S, K.J and Y.H were involved in study concept and study supervision. References Imperial R, Ahmed Z, Toor OM, Erdogan C, Khaliq A, Case P, et al. Comparative proteogenomic analysis of right-sided colon cancer, left-sided colon cancer and rectal cancer reveals distinct mutational profiles. Mol Cancer. 2018;17(1):177. Yun HR, Lee LJ, Park JH, Cho YK, Cho YB, Lee WY, et al. Local recurrence after curative resection in patients with colon and rectal cancers. Int J Colorectal Dis. 2008;23(11):1081–7. Benson AB, Venook AP, Al-Hawary MM, Arain MA, Chen YJ, Ciombor KK, et al. NCCN Guidelines Insights: Rectal Cancer, Version 6.2020. J Natl Compr Canc Netw. 2020;18(7):806–15. Minsky B, Cohen A, Enker W, Kelsen D, Kemeny N, Ilson D, et al. Preoperative 5-fluorouracil, low-dose leucovorin, and concurrent radiation therapy for rectal cancer. Cancer. 1994;73(2):273–80. Gerard JP, Azria D, Gourgou-Bourgade S, Martel-Lafay I, Hennequin C, Etienne PL, et al. Clinical outcome of the ACCORD 12/0405 PRODIGE 2 randomized trial in rectal cancer. J Clin Oncol. 2012;30(36):4558–65. Matsusaka S, Ishihara S, Kondo K, Horie H, Uehara K, Oguchi M, et al. A multicenter phase II study of preoperative chemoradiotherapy with S-1 plus oxaliplatin for locally advanced rectal cancer (SHOGUN trial). Radiother Oncol. 2015;116(2):209–13. Habr-Gama A, Sabbaga J, Gama-Rodrigues J, Sao Juliao GP, Proscurshim I, Bailao Aguilar P, et al. Watch and wait approach following extended neoadjuvant chemoradiation for distal rectal cancer: are we getting closer to anal cancer management? Dis Colon Rectum. 2013;56(10):1109–17. Dossa F, Chesney TR, Acuna SA, Baxter NN. A watch-and-wait approach for locally advanced rectal cancer after a clinical complete response following neoadjuvant chemoradiation: a systematic review and meta-analysis. The Lancet Gastroenterology & Hepatology. 2017;2(7):501–13. Habr-Gama A, Perez RO, Wynn G, Marks J, Kessler H, Gama-Rodrigues J. Complete clinical response after neoadjuvant chemoradiation therapy for distal rectal cancer: characterization of clinical and endoscopic findings for standardization. Dis Colon Rectum. 2010;53(12):1692–8. Watanabe T, Kobunai T, Akiyoshi T, Matsuda K, Ishihara S, Nozawa K. Prediction of response to preoperative chemoradiotherapy in rectal cancer by using reverse transcriptase polymerase chain reaction analysis of four genes. Dis Colon Rectum. 2014;57(1):23–31. Fischer J, Eglinton TW, Richards SJ, Frizelle FA. Predicting pathological response to chemoradiotherapy for rectal cancer: a systematic review. Expert Rev Anticancer Ther. 2021:1–12. Kawai K, Ishihara S, Nozawa H, Hata K, Kiyomatsu T, Morikawa T, et al. Prediction of Pathological Complete Response Using Endoscopic Findings and Outcomes of Patients Who Underwent Watchful Waiting After Chemoradiotherapy for Rectal Cancer. Dis Colon Rectum. 2017;60(4):368–75. Freischlag K, Sun Z, Adam MA, Kim J, Palta M, Czito BG, et al. Association Between Incomplete Neoadjuvant Radiotherapy and Survival for Patients With Locally Advanced Rectal Cancer. JAMA Surg. 2017;152(6):558–64. Diefenhardt M, Ludmir EB, Hofheinz RD, Ghadimi M, Minsky BD, Rodel C, et al. Association of Treatment Adherence With Oncologic Outcomes for Patients With Rectal Cancer: A Post Hoc Analysis of the CAO/ARO/AIO-04 Phase 3 Randomized Clinical Trial. JAMA Oncol. 2020. Limkin EJ, Sun R, Dercle L, Zacharaki EI, Robert C, Reuze S, et al. Promises and challenges for the implementation of computational medical imaging (radiomics) in oncology. Ann Oncol. 2017;28(6):1191–206. Shaikh FA, Kolowitz BJ, Awan O, Aerts HJ, von Reden A, Halabi S, et al. Technical Challenges in the Clinical Application of Radiomics. JCO Clin Cancer Inform. 2017;1:1–8. Skrede O-J, De Raedt S, Kleppe A, Hveem TS, Liestøl K, Maddison J, et al. Deep learning for prediction of colorectal cancer outcome: a discovery and validation study. The Lancet. 2020;395(10221):350–60. Huang YQ, Liang CH, He L, Tian J, Liang CS, Chen X, et al. Development and Validation of a Radiomics Nomogram for Preoperative Prediction of Lymph Node Metastasis in Colorectal Cancer. J Clin Oncol. 2016;34(18):2157–64. Yang L, Dong D, Fang M, Zhu Y, Zang Y, Liu Z, et al. Can CT-based radiomics signature predict KRAS/NRAS/BRAF mutations in colorectal cancer? Eur Radiol. 2018;28(5):2058–67. Dercle L, Lu L, Schwartz LH, Qian M, Tejpar S, Eggleton P, et al. Radiomics Response Signature for Identification of Metastatic Colorectal Cancer Sensitive to Therapies Targeting EGFR Pathway. J Natl Cancer Inst. 2020;112(9):902–12. Liu Z, Zhang XY, Shi YJ, Wang L, Zhu HT, Tang Z, et al. Radiomics Analysis for Evaluation of Pathological Complete Response to Neoadjuvant Chemoradiotherapy in Locally Advanced Rectal Cancer. Clin Cancer Res. 2017;23(23):7253–62. Peng H, Dong D, Fang MJ, Li L, Tang LL, Chen L, et al. Prognostic Value of Deep Learning PET/CT-Based Radiomics: Potential Role for Future Individual Induction Chemotherapy in Advanced Nasopharyngeal Carcinoma. Clin Cancer Res. 2019;25(14):4271–9. Huang CM, Huang MY, Huang CW, Tsai HL, Su WC, Chang WC, et al. Machine learning for predicting pathological complete response in patients with locally advanced rectal cancer after neoadjuvant chemoradiotherapy. Sci Rep. 2020;10(1):12555. Hamerla G, Meyer HJ, Hambsch P, Wolf U, Kuhnt T, Hoffmann KT, et al. Radiomics Model Based on Non-Contrast CT Shows No Predictive Power for Complete Pathological Response in Locally Advanced Rectal Cancer. Cancers (Basel). 2019;11(11). Lovinfosse P, Polus M, Van Daele D, Martinive P, Daenen F, Hatt M, et al. FDG PET/CT radiomics for predicting the outcome of locally advanced rectal cancer. Eur J Nucl Med Mol Imaging. 2018;45(3):365–75. Yang C, Jiang ZK, Liu LH, Zeng MS. Pre-treatment ADC image-based random forest classifier for identifying resistant rectal adenocarcinoma to neoadjuvant chemoradiotherapy. Int J Colorectal Dis. 2020;35(1):101–7. Bulens P, Couwenberg A, Intven M, Debucquoy A, Vandecaveye V, Van Cutsem E, et al. Predicting the tumor response to chemoradiotherapy for rectal cancer: Model development and external validation using MRI radiomics. Radiother Oncol. 2020;142:246–52. Kiyomatsu T, Watanabe T, Muto T, Nagawa H. The 4-portal technique decreases adverse effects in preoperative radiotherapy for advanced rectal cancer: comparison between the 2-portal and the 4-portal techniques. Am J Surg. 2007;194(4):542–8. Hashiguchi Y, Muro K, Saito Y, Ito Y, Ajioka Y, Hamaguchi T, et al. Japanese Society for Cancer of the Colon and Rectum (JSCCR) guidelines 2019 for the treatment of colorectal cancer. Int J Clin Oncol. 2019. Vallieres M, Freeman CR, Skamene SR, El Naqa I. A radiomics model from joint FDG-PET and MRI texture features for the prediction of lung metastases in soft-tissue sarcomas of the extremities. Phys Med Biol. 2015;60(14):5471–96. Chino A, Konishi T, Ogura A, Kawachi H, Osumi H, Yoshio T, et al. Endoscopic criteria to evaluate tumor response of rectal cancer to neoadjuvant chemoradiotherapy using magnifying chromoendoscopy. Eur J Surg Oncol. 2018;44(8):1247–53. Sohn DK, Han KS, Kim BC, Hong CW, Chang HJ, Baek JY, et al. Endoscopic assessment of tumor regression after preoperative chemoradiotherapy as a prognostic marker in locally advanced rectal cancer. Surg Oncol. 2017;26(4):453–9. Iwatate Y, Hoshino I, Yokota H, Ishige F, Itami M, Mori Y, et al. Radiogenomics for predicting p53 status, PD-L1 expression, and prognosis with machine learning in pancreatic cancer. Br J Cancer. 2020;123(8):1253–61. Badic B, Hatt M, Durand S, Jossic-Corcos CL, Simon B, Visvikis D, et al. Radiogenomics-based cancer prognosis in colorectal cancer. Sci Rep. 2019;9(1):9743. Bibault JE, Giraud P, Housset M, Durdux C, Taieb J, Berger A, et al. Deep Learning and Radiomics predict complete response after neo-adjuvant chemoradiation for locally advanced rectal cancer. Sci Rep. 2018;8(1):12611. Additional Declarations No competing interests reported. Supplementary Files Supplementarydata.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-1717256","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":110588782,"identity":"fb8d00a7-e349-417f-996a-01f0bb8b0c81","order_by":0,"name":"Tsuyoshi Ozawa","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA60lEQVRIiWNgGAWjYDADfoYDiQ+ANA8fIZU8MIZk44HHBiABNqK1GBw++EwCxCCoxZ7/jOHDHzWH5RmOHU6r/JpjJ8PGwPzw0Q18tkjkGBtIHDts2NhzLO227LZkoMPYjI1z8GrhMZMwYLvN2CxxJu225DZmoBYeNmm8WvjPmEkk/Ltt3yb//lux5LZ6IrQw5JhJHGy7ndjDcCCN8eO2w0RouZFWbNjY9z95BsOBZGnGbcd52JgJ+IW9//DGhz++pdnuP3Ag8ePPbdX2/OzNDx/j04ICmMGxxEyschBg/EGK6lEwCkbBKBgxAABua0kPRynDewAAAABJRU5ErkJggg==","orcid":"","institution":"Teikyo University School of Medicine","correspondingAuthor":true,"prefix":"","firstName":"Tsuyoshi","middleName":"","lastName":"Ozawa","suffix":""},{"id":110588783,"identity":"ebe6dffd-66e2-4b69-ab8e-17319d5be99a","order_by":1,"name":"Yusuke Saikawa","email":"","orcid":"","institution":"Teikyo University","correspondingAuthor":false,"prefix":"","firstName":"Yusuke","middleName":"","lastName":"Saikawa","suffix":""},{"id":110588784,"identity":"463cd78c-5a52-46b6-b060-5b2e1aea59cd","order_by":2,"name":"Tamuro Hayama","email":"","orcid":"","institution":"Teikyo University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Tamuro","middleName":"","lastName":"Hayama","suffix":""},{"id":110588785,"identity":"c55a0c8d-bcb1-493e-aa9e-46f1d1d67878","order_by":3,"name":"Keijiro Nozawa","email":"","orcid":"","institution":"Teikyo University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Keijiro","middleName":"","lastName":"Nozawa","suffix":""},{"id":110588786,"identity":"a54b1400-b06c-4cf2-90a4-a1df37b4d35a","order_by":4,"name":"Keiji Matsuda","email":"","orcid":"","institution":"Teikyo University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Keiji","middleName":"","lastName":"Matsuda","suffix":""},{"id":110588787,"identity":"b6e6b7fb-6824-4640-ac95-1d1d2abb8df8","order_by":5,"name":"Soichiro Ishihara","email":"","orcid":"","institution":"The University of Tokyo","correspondingAuthor":false,"prefix":"","firstName":"Soichiro","middleName":"","lastName":"Ishihara","suffix":""},{"id":110588788,"identity":"3973f83e-b0b0-47c8-9d01-96e2f4c02ab6","order_by":6,"name":"Jun’ichi Kotoku","email":"","orcid":"","institution":"Teikyo University","correspondingAuthor":false,"prefix":"","firstName":"Jun’ichi","middleName":"","lastName":"Kotoku","suffix":""},{"id":110588789,"identity":"f0d9d1ef-f4f9-4911-938d-d22bb05f9855","order_by":7,"name":"Yojiro Hashiguchi","email":"","orcid":"","institution":"Teikyo University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Yojiro","middleName":"","lastName":"Hashiguchi","suffix":""}],"badges":[],"createdAt":"2022-06-01 21:44:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-1717256/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-1717256/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":22380605,"identity":"deb58934-1ade-46f3-aefc-a31306ac3459","added_by":"auto","created_at":"2022-06-07 20:02:00","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":90592,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFlow chart of the patient selection. \u003cspan class=\"ql-cursor\"\u003e\u003c/span\u003e\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Fig1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-1717256/v1/e8539a0d0f4ea1615bb44fe9.jpg"},{"id":22380307,"identity":"9fd02f71-e25a-4e1d-989f-319c7cc1bdd8","added_by":"auto","created_at":"2022-06-07 19:57:00","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":134988,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFlow chart of the development of radiomic classifiers. \u003c/strong\u003eWe manually marked tumor areas in the endoscopic images and converted them into gray scaling. From each image, 774 texture features were extracted. Using these texture features, the radiomics classifiers were developed by applying 12 different machine-learning models. The performances of these radiomics classifiers were evaluated by receiver operating characteristics curves.\u003c/p\u003e","description":"","filename":"Fig2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-1717256/v1/364cbcc4dc3361ef79f06417.jpg"},{"id":22380604,"identity":"cfadb889-92ec-41cd-ab62-acd305ff4138","added_by":"auto","created_at":"2022-06-07 20:02:00","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":122937,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe performance of each radiomics classifier. A:\u003c/strong\u003e A heatmap showing the AUROC of each classifier in the prediction of a pCR in the image-basis analysis. \u003cstrong\u003eB:\u003c/strong\u003e A heatmap showing the AUROC of each classifier in the prediction of the CRT-response in the image-basis analysis. \u003cstrong\u003eC:\u003c/strong\u003e The AUROC of the cubic SVM model to predict a pCR using pre-CRT images in the patient-basis analysis. \u003cstrong\u003eD:\u003c/strong\u003e The AUROC of the quadratic SVM model to predict a pCR using post-CRT images in the patient-basis analysis. \u003cstrong\u003eE:\u003c/strong\u003e The AUROC of the Gaussian SVM model to predict the CRT-response using pre-CRT images in the patient-basis analysis. \u003cstrong\u003eF:\u003c/strong\u003e The AUROC of the Gaussian SVM model to predict the CRT-response using post-CRT images in the patient-basis analysis.\u003c/p\u003e","description":"","filename":"Fig3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-1717256/v1/df327f56afb866ad9dbe93d5.jpg"},{"id":22380606,"identity":"936e0292-027b-435d-8eba-2ab996eb30af","added_by":"auto","created_at":"2022-06-07 20:02:03","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":592429,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-1717256/v1/402062c7-9f61-4396-8ab7-198d4267cd98.pdf"},{"id":22380310,"identity":"10b7c813-f29f-4872-baf7-4908d6207da3","added_by":"auto","created_at":"2022-06-07 19:57:00","extension":"docx","order_by":8,"title":"","display":"","copyAsset":false,"role":"supplement","size":285756,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarydata.docx","url":"https://assets-eu.researchsquare.com/files/rs-1717256/v1/d6d4ec5bccae35a20d6b661f.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Endoscopic Image-based Radiomics Classifiers for the Prediction of Neoadjuvant Chemoradiotherapy Response in Rectal Cancer Patients","fulltext":[{"header":"Introduction","content":"\u003cp\u003eApproximately 25% of colorectal cancers (CRCs) are located in the rectum, and previous studies have described genetic differences between colon cancers and rectal cancers (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e). The anatomical location of the rectum also yields clinical differences between these cancers; surgeries for rectal cancer, especially lower advanced rectal cancer (LARC), are technically difficult compared to those for colon cancers, and this difficulty leads to the frequent local recurrence of rectal cancers (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e). To address this issue, the standard treatment of LARC in Western countries is neoadjuvant chemoradiotherapy (NA-CRT) followed by surgery (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e). NA-CRT has the potential to achieve down-staging of the tumor, which may reduce the risk of tumor-cell shedding during surgery and/or a positive surgical margin, thus improving the local control of the tumor.\u003c/p\u003e \u003cp\u003eRoughly 15\u0026ndash;25% of patients with LARC have demonstrated a pathological complete response (pCR) after NA-CRT (\u003cspan additionalcitationids=\"CR5\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e). Theoretically, surgery can be avoided for patients with rectal cancers who achieve a pCR after NA-CRT, making it possible to preserve the organs. The \"watch-and-wait\" strategy is thus one of the available treatment strategies; i.e., when the post-CRT examinations demonstrate no evidence of residual tumor, close follow-up is performed without surgery until tumor recurrence becomes apparent (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e). However, the accuracy of identifying a pCR by a pre-operative radiological examination is limited, and another approach is necessary to improve the results of the watch-and-wait strategy (\u003cspan additionalcitationids=\"CR10 CR11\" citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). Moreover, NA-CRT may cause severe side effects without having an impact on tumor progression, and some tumors even progress during NA-CRT (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e). It is thus desirable to predict whether a tumor will or will not respond to NA-CRT.\u003c/p\u003e \u003cp\u003eThe performance of radiomics has greatly improved recently due to the development of computational methods. Radiomics extract various types of quantitative features of regions of interest (ROIs) from digital images and enable the determination of the relationships between the extracted features and underlying pathophysiological condition of the ROIs(\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e). Radiomics may make it possible to evaluate the quality of ROIs systematically and objectively, and radiomics thus have a great potential to surpass human performance (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e). Previous studies demonstrated the robust performance of radiomics for predicting lymph node metastases, chemotherapy response or even mutational status of colorectal cancers based on computed tomography (CT) images (\u003cspan additionalcitationids=\"CR19\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eRegarding the prediction and evaluation of LARCs' responses to NA-CRT, several studies evaluated the performance of radiomics analyses using CT images, magnetic resonance images (MRIs), or positron emission tomography (PET) images (\u003cspan additionalcitationids=\"CR22 CR23 CR24 CR25 CR26\" citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e). The results of these studies indicated a promising role of radiomics in the prediction of non-responders before NA-CRT and in the identification of pCRs after NA-CRT. However, the identification of the ROIs can be very difficult and subjective in these radiological examinations, especially after NA-CRT, and this may cause inter-observer differences. In the present study, we used endoscopic images of rectal cancers, as it is fairly easy to recognize the ROI in these cancers and the results may have higher reproducibility, which is an important factor for biomarker development. To our knowledge, this is the first study to construct and evaluate the performance of radiomics using endoscopic images of LARCs.\u003c/p\u003e"},{"header":"Materials And Methods","content":"\u003cdiv class=\"Section2\" id=\"Sec3\"\u003e\n \u003ch2\u003ePatients\u003c/h2\u003e\n \u003cp\u003eWe retrospectively reviewed the clinicopathological data of 140 patients with middle and low LARCs who had undergone NA-CRT followed by surgery at Teikyo University Hospital from Oct. 2007 to Dec. 2019. The exclusion criteria were (\u003cspan class=\"CitationRef\"\u003e1\u003c/span\u003e) non-adenocarcinoma (n\u0026thinsp;=\u0026thinsp;8), (\u003cspan class=\"CitationRef\"\u003e2\u003c/span\u003e) anal fistula-related cancer (n\u0026thinsp;=\u0026thinsp;3), (\u003cspan class=\"CitationRef\"\u003e3\u003c/span\u003e) Inflammatory bowel disease associated cancer (n\u0026thinsp;=\u0026thinsp;1), (\u003cspan class=\"CitationRef\"\u003e4\u003c/span\u003e) patients who developed distant metastases during NA-CRT (n\u0026thinsp;=\u0026thinsp;7), (\u003cspan class=\"CitationRef\"\u003e5\u003c/span\u003e) a positive surgical margin (n\u0026thinsp;=\u0026thinsp;5), (\u003cspan class=\"CitationRef\"\u003e6\u003c/span\u003e) patients with synchronous cancer in the other organs (n\u0026thinsp;=\u0026thinsp;1), (\u003cspan class=\"CitationRef\"\u003e7\u003c/span\u003e) patients whose colonoscopy images were not available (n\u0026thinsp;=\u0026thinsp;15), and (\u003cspan class=\"CitationRef\"\u003e8\u003c/span\u003e) patients whose NA-CRT was incomplete (n\u0026thinsp;=\u0026thinsp;10). A final total of 90 patients were included in the analysis (Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eAll patient information was de-identified prior to the data analyses to maintain patient anonymity. Patients\u0026apos; written informed consent for their data to be used was obtained, and this study was approved by the Teikyo University Ethics Committee (No. 19\u0026ndash;127). The study protocol conforms to the ethical guidelines of the 1975 Declaration of Helsinki and its later amendments.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv class=\"Section2\" id=\"Sec4\"\u003e\n \u003ch2\u003eNeoadjuvant chemoradiotherapy and concomitant surgery\u003c/h2\u003e\n \u003cp\u003eNA-CRT was performed as described (\u003cspan class=\"CitationRef\"\u003e28\u003c/span\u003e). In summary, the total dose of radiotherapy was 50.5 Gy or 45 Gy, which was given in 28 fractions or 25 fractions. Treatment planning was done with CT scans, and the target volume included the primary tumor, the anus, and lymph nodes in the mesorectum and in the pelvis (lateral lymph nodes). As the chemotherapy regimen, fluoropyrimidine (tegafur-uracil with leucovorin, capecitabine, S-1) alone, S-1 with oxaliplatin, or S-1 with irinotecan were induced concomitantly with radiotherapy. We defined NA-CRT-completion as cases in which both \u0026ge;\u0026thinsp;80% of scheduled chemotherapy and the radiotherapy dose of \u0026ge;\u0026thinsp;45 Gy were used; the other cases were recognized as incomplete-CRT (\u003cspan class=\"CitationRef\"\u003e14\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eThe surgical procedure in each of the eligible cases was performed at approx. 8 weeks after the completion of NA-CRT. A total mesorectal excision (TME) with regional lymph node dissection was performed under laparotomy or laparoscopically.\u003c/p\u003e\n \u003cp\u003eAll of the resected specimens were evaluated by the institutional pathologists, and the TNM classification and CRT grade were determined according to the Japanese Society for Cancer of the Colon and Rectum guidelines (ninth edition) as follows. Grade 0: No evidence of tumor response. Grade 1: A\u0026thinsp;\u0026lt;\u0026thinsp;2/3 regression of the tumor cells. Grade 2: A\u0026thinsp;\u0026ge;\u0026thinsp;2/3 regression of the tumor cells. Grade 3: Complete regression. Grades 2 and 3 were regarded as a good response, and a pCR was defined as Grade 3 (\u003cspan class=\"CitationRef\"\u003e29\u003c/span\u003e).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv class=\"Section2\" id=\"Sec5\"\u003e\n \u003ch2\u003eImage preparation for training and validation sets\u003c/h2\u003e\n \u003cp\u003eEach colonoscopy was performed using standard endoscope equipment (Evis Lucera and PCF type Q260AI and H290I, Olympus Medical Systems, Tokyo). Six weeks after finishing NA-CRT, the second-look colonoscopy was performed.\u003c/p\u003e\n \u003cp\u003eAll of the patients\u0026apos; endoscopic images were extracted and reviewed by two board-certified gastroenterologists (T.O and T.H). Only the non-magnified images including primary tumors observed using conventional white-light were selected in the analyses. Pre-CRT images were available from 88 of the 90 patients and included 2,474 images in total. Post-CRT images were available from all 90 patients and included 2,781 images in total (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eThe primary tumors were then marked manually using ImageJ software (imagej.nih.gov/ij/download/), and ROIs were created.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv class=\"Section2\" id=\"Sec6\"\u003e\n \u003ch2\u003eRadiomics feature extraction and the development of radiomics classifiers using machine learning models\u003c/h2\u003e\n \u003cp\u003eFirst, the ROIs were converted to grayscale using the grb2 gray function of MATLAB software (MathWorks, Natick, MA, U.S.), and we evaluated the texture features based on the distribution of the intensity of each pixel value. Five different types of texture features were calculated using the Valli\u0026egrave;res radiomics toolbox of MATLAB software (\u003cspan class=\"CitationRef\"\u003e30\u003c/span\u003e): (\u003cspan class=\"CitationRef\"\u003e1\u003c/span\u003e) the gray level histogram which includes three statistics, (\u003cspan class=\"CitationRef\"\u003e2\u003c/span\u003e) the gray level co-occurrence matrix which includes nine statistics, (\u003cspan class=\"CitationRef\"\u003e3\u003c/span\u003e) the gray level run-length matrix which includes 13 statistics, (\u003cspan class=\"CitationRef\"\u003e4\u003c/span\u003e) the gray level size-zone matrix which includes 13 statistics, and (\u003cspan class=\"CitationRef\"\u003e5\u003c/span\u003e) the neighborhood gray-tone difference matrix which includes five statistics. A total of 43 texture features were thus evaluated (\u003cstrong\u003eSuppl. Table S1\u003c/strong\u003e). To amplify the texture features, each texture feature was extracted using 20 different parameters based on previous feature extraction models, resulting in a total of 860 features per image.\u003c/p\u003e\n \u003cp\u003eThen, in order to avoid multicollinearity, we used a bootstrap procedure (bootstrap sample\u0026thinsp;=\u0026thinsp;1000) based on Spearman\u0026apos;s rho to select the most suitable parameters for each of the 43 texture features. Finally, each texture feature was normalized through a z-transform.\u003c/p\u003e\n \u003cp\u003eUsing the extracted texture features, we developed radiomics classifiers by applying 12 different machine learning models in the Statistics and Machine Learning toolbox of MATLAB: the linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), linear support vector machine (SVM), cubic SVM, Gaussian SVM, quadratic SVM, k-nearest neighbor algorithm (KNN), cubic KNN, cosine KNN, weighted KNN, logistic regression model, and decision trees. The pre-CRT images and post-CRT images were subjected to each of these models separately, and pre-CRT image-based radiomics classifiers and post-CRT image-based radiomics classifiers were developed (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv class=\"Section2\" id=\"Sec7\"\u003e\n \u003ch2\u003eEvaluation of the performances of the radiomics classifiers\u003c/h2\u003e\n \u003cp\u003eWe evaluated the performance of each radiomics classifier to predict a good response and a pCR by using a 10-fold double cross-validation technique (\u003cstrong\u003eSuppl. Fig. S1\u003c/strong\u003e). The receiver operating characteristics curve (ROC) was determined and the area under the ROC (AUROC) was calculated for each radiomics classifier. Because each tumor provided multiple images, the performance of our radiomics was evaluated on both an image basis and a patient basis.\u003c/p\u003e\n \u003cp\u003eTo evaluate the per-patient performances of the radiomics classifiers, we used the average score of the images from each patient. Cut-off values for the radiomics classifiers were decided using Youden\u0026apos;s index, and the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy were calculated. In the comparison of clinicopathological features, the \u0026chi;\u003csup\u003e2\u003c/sup\u003e test and Fisher\u0026apos;s exact test were used for categorical data, and the Wilcoxon rank sum test was used for un-paired continuous variables. All statistics were performed using MATLAB (MathWorks) and JMP pro 15 (SAS Institute Japan, Tokyo).\u003c/p\u003e\n\u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cem\u003eThe performances of the radiomics classifiers for predicting a pCR and response to NA-CRT on the image basis\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eSixty-six patients (73%) underwent fluoropyrimidine (FP)-based NA-CRT, 21 patients (23%) underwent S-1\u0026thinsp;+\u0026thinsp;oxaliplatin based NA-CRT, and three patients (3%) underwent S-1\u0026thinsp;+\u0026thinsp;irinotecan-based NA-CRT. The average interval between the end of the day of NA-CRT to the second-look endoscopy and to the surgery were 42 days (SD 13) and 55 (SD 13) days, respectively. The pathological findings demonstrated the CRT-grades 0 or 1 in 43 patients (48%), CRT-grade 2 in 37 patients (41%), and CRT-grade 3 in 10 patients (11%). The detailed clinicopathological features are summarized in \u003cstrong\u003eTable\u0026nbsp;1\u003c/strong\u003e.\u003c/p\u003e\n\u003ctable border=\"1\" width=\"0\"\u003e\n \u003ccaption\u003e\n \u003cp\u003eTable 1\u003c/p\u003e\n \u003cp\u003eClinicopahtological features of the patients included in the study\u003c/p\u003e\n \u003c/caption\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"226\"\u003e\n \u003cp\u003e\u003cstrong\u003eCharacteristics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"189\"\u003e\n \u003cp\u003e\u003cstrong\u003eN (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"226\"\u003e\n \u003cp\u003e\u003cstrong\u003eGender \u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u0026nbsp; Male / Female\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"189\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e71 (79) / 19 (21)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"226\"\u003e\n \u003cp\u003e\u003cstrong\u003eAge\u0026nbsp;\u003c/strong\u003e(Years)\u003c/p\u003e\n \u003cp\u003e\u0026nbsp; Mean (SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"189\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e61.9 (9.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"226\"\u003e\n \u003cp\u003e\u003cstrong\u003eDistance from AV\u0026nbsp;\u003c/strong\u003e(cm)\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u0026nbsp;\u003c/strong\u003e\u003cu\u003e\u0026lt;\u003c/u\u003e5 / \u0026gt;5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"189\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e59 (66) / 31 (34)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"226\"\u003e\n \u003cp\u003e\u003cstrong\u003ecStage\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u0026nbsp; 2 / 3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"189\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e29 (32) / 61 (68)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"226\"\u003e\n \u003cp\u003e\u003cstrong\u003eChemotherapy\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u0026nbsp;\u003c/strong\u003eFP / S-1+OX / S-1+IRI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"189\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e66 (73) / 21 (23) / 3 (3)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"226\"\u003e\n \u003cp\u003e\u003cstrong\u003eRadiation dose\u0026nbsp;\u003c/strong\u003e(Gy)\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u0026nbsp;\u003c/strong\u003e45 / 50.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"189\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e3 (3) / 87 (97)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"226\"\u003e\n \u003cp\u003e\u003cstrong\u003eypStage\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u0026nbsp; 0-1 / 2 / 3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"189\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e37 (41) / 26 (29) / 27 (30)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"226\"\u003e\n \u003cp\u003e\u003cstrong\u003eHistology\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u0026nbsp; Well-Mod / Por-Muc / unavailable\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"189\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e81 (90) / 8 (9) / 1 (1)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"226\"\u003e\n \u003cp\u003e\u003cstrong\u003ePre-CRT CEA (ng/ml)\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u0026nbsp;\u003c/strong\u003e\u003cu\u003e\u0026lt;\u003c/u\u003e5 / \u0026gt;5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"189\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e52 (58) / 38 (42)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"226\"\u003e\n \u003cp\u003e\u003cstrong\u003eCRT grade\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u0026nbsp;\u003c/strong\u003e0-1 / 2 / 3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"189\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e43 (48) / 37 (41) / 10 (11)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" width=\"226\"\u003eAV: anal verge, FP: fluoropyrimidine, OX: oxaliplatin, IRI: irinotecan, CRT: chemoradiotherapyCEA: carcinoembryonic antigen\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003eThe numbers of pre-CRT images of CRT grade 0 or 1, grade 2, and grade 3 were 1,151 (47%), 1,098 (44%), and 225 (9%), respectively. The numbers of post-CRT images of CRT grade 0 or 1, grade 2, and grade 3 were 1,211 (44%), 1,302 (47%), and 268 (9%), respectively (Table \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). The results of our comparison of the clinicopathological features between CRT grades 0\u0026ndash;2 and CRT grade 3 and between CRT grades 0\u0026ndash;1 and CRT grades 2\u0026ndash;3 are shown in \u003cstrong\u003eSupplementary Table S2\u003c/strong\u003e.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003ctable border=\"1\" id=\"Tab1\"\u003e\n \u003ccaption\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eThe detail of the images included in the study\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePre or Post CRT\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eNo. of patients\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eCRT grade\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eNo. of the images (N, %)\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\u003ePre\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u0026ndash;1\u003c/p\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1151 (47)\u003c/p\u003e\n \u003cp\u003e1098 (44)\u003c/p\u003e\n \u003cp\u003e225 (9)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u0026ndash;1\u003c/p\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1211 (44)\u003c/p\u003e\n \u003cp\u003e1302 (47)\u003c/p\u003e\n \u003cp\u003e268 (9)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/div\u003e\n\u003cp\u003eFigure \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003eA is a heatmap showing the per-image performances of each radiomics classifier to predict a pCR using pre- or post-CRT images. The AUCs ranged from 0.529 to 0.816 with the pre-CRT images, and the best radiomics classifier was the cubic SVM model with 84% accuracy (Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e). With the post-CRT images, the AUROCs ranged from 0.619 to 0.839, and the best radiomics classifier was the quadratic SVM model with 83% accuracy (Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003ctable border=\"1\" id=\"Tab2\"\u003e\n \u003ccaption\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eThe performances of each classifier to predict pCR and CRT-response\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePrediction\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eImage\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eClassifier\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eEvaluation\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSEN\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSPE\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePPV\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eNPV\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eACU\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\" rowspan=\"4\"\u003e\n \u003cp\u003e\u003cstrong\u003epCR\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003ePre-CRT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eCubic\u003c/p\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePer image\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e63%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e87%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e32%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e96%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e84%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePer patient\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e70%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e100%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e100%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e96%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e97%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003ePost-CRT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eQuadratic\u003c/p\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePer image\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e71%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e84%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e33%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e96%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e83%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePer patient\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e90%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e96%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e75%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e99%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e96%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" rowspan=\"4\"\u003e\n \u003cp\u003e\u003cstrong\u003eCRT-\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eresponse\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003ePre-CRT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eGaussian\u003c/p\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePer image\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e71%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e74%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e76%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e69%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e73%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePer patient\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e74%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e88%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e87%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e76%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e81%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003ePost-CRT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eGaussian\u003c/p\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePer image\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e84%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e71%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e79%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e77%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e78%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePer patient\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e96%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e81%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e85%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e96%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e89%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003ctfoot\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"9\"\u003epCR: pathological complete response, CRT: chemoradiotherapy, SEN: sensitivity, SPE: specificity, PPV: positive predictive value, NPV: negative predictive value, ACU: accuracy\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tfoot\u003e\n \u003c/table\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/div\u003e\n\u003cp\u003eFigure \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003eB is a heatmap demonstrating the per-image performances of each radiomics classifier to predict a CRT response (good responder or not) using pre- or post-CRT images. The AUROCs ranged from 0.679 to 0.792 with the pre-CRT images, and the best radiomics classifier was the Gaussian SVM model with 73% accuracy (Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e). The AUROCs ranged from 0.687 to 0.841 with the post-CRT images, and the best radiomics classifier was the Gaussian SVM model with 78% accuracy (Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eThe performances of the radiomics classifiers for predicting a pCR and response to NA-CRT on the patient basis\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eWe then evaluated the per-patient performance of each of the best per-image classifiers described above. The per-patient performance of the cubic SVM model to predict a pCR using pre-CRT images was excellent with the AUROC of 0.904 (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003eC) and 97% accuracy (Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e). The per-patient performance of the quadratic SVM model to predict a pCR using post-CRT images showed the AUROC of 0.958 (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003eD) and 96% accuracy (Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e). The Gaussian SVM model demonstrated high performances to predict a CRT response in both pre-CRT images and post-CRT images, with the AUROC of 0.857 and 80% accuracy, and the AUROC of 0.950 and 89% accuracy, respectively (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003eE, F, Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e)\u003c/p\u003e\n\u003cp\u003eFor pre-CRT radiomics classifiers, the association between the scores of these radiomics classifiers and pre-CRT T-stage and pre-CRT occupation rate of rectal circumference were evaluated, however, there was no statistical association between them (\u003cstrong\u003eSuppl. Fig. S2A, B\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eFor post-CRT radiomics classifiers, the association between the scores of these radiomics classifiers and pathological T-stage and pathological tumor diameter were evaluated. The post-CRT quadratic SVM model was significantly associated with pathological T-stage (p\u0026thinsp;\u0026lt;\u0026thinsp;0.0001), and the post-CRT Gaussian SVM model was significantly associated with tumor diameter (p\u0026thinsp;=\u0026thinsp;0.045) (\u003cstrong\u003eSuppl. Fig. S2C, D\u003c/strong\u003e).\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eWe have developed and validated radiomics classifiers to predict a pCR and CRT response using pre-CRT and post-CRT endoscopic images. Our radiomics models showed robust performances in predicting a pCR and CRT response with both pre- and post-CRT images. These models are easy to use and effective for decision-making regarding the treatment of rectal cancer patients. The prediction of CRT-response before NA-CRT in particular may make it possible to reduce the unnecessary NA-CRT, and the prediction of a pCR after NA-CRT may contribute to the watch-and-wait strategy. In this regard, our radiomics classifiers demonstrated excellent ability to predict the CRT-response using pre-CRT images with 80% accuracy, and to predict a pCR using post-CRT images with 96% accuracy on a patient basis.\u003c/p\u003e \u003cp\u003eRadiomics is used to convert medical images into high-dimensional, mineable data by extracting quantitative features and analyzing the extracted features for decision support (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e). The radiomics method is gradually gaining attention with the recent advances in pattern recognition tools. Other research groups constructed radiomics models to predict a pCR or CRT response using CT, MRI, or PET data (\u003cspan additionalcitationids=\"CR22 CR23\" citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e). Most of those studies showed an AUC of ~\u0026thinsp;0.85 to predict a pCR or CRT response. Among the studies, Liu et al. reported that their radiomics signature comprised of 30 selected features in pre- and post-MRI images showed the AUC of 0.9799 to predict a pCR in an independent cohort of 70 rectal cancer patients (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e). However, it is usually difficult to accurately detect the tumor location in MR images, especially after CRT. Liu et al. thus placed the ROI on the primary tumor bed region corresponding to that observed pre-CRT when the tumor bed region could not be detected on post-CRT MRI. This method may be complicated and can also be subjective. In contrast, endoscopy shows the location of the tumor bed fairly clearly, and it is much easier to place ROIs with endoscopic help, leading to objective and reproducible results. Moreover, our radiomics classifiers evaluated surface pattern of the tumors. Pre-CRT radiomics classifiers were independent of conventional T stage or occupation rate of the tumor in the rectum. These findings suggest that the surface pattern of the tumor may include various information of the tumor characters. Thus, it is also intriguing to construct a radiomics classifier using new endoscopic technologies such as narrow band imaging or magnifying endoscopy(\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSpecific endoscopic findings of a pCR have been reported in several studies (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e). Kawai et al. reported that flattened marginal swelling was an independent and strong predictor of a pCR with 69.1% sensitivity and 73.9% specificity (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). This endoscopic finding provided better specificity but lower sensitivity compared to CT assessment. Combining these methods will thus increase the prediction ability. Future studies that incorporate various types of radiomics features including CT, MRI, PET, and endoscopy are very promising (\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eWe constructed our models using conventional machine learning models. Bibault et al. used a deep-learning method to construct a radiomics signature using the T-stage and 28 radiomics features, and the results demonstrated that the model could predict a pCR with 80% accuracy, which is better accuracy than that provided by a conventional SVM model (\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e). It could thus be intriguing to construct a model using a deep-learning method and radiomics features. Image recognition using a deep-learning method is also a promising method.\u003c/p\u003e \u003cp\u003eA convolutional neural network (CNN) is a one of the deep-learning based architectures that is suitable for image interpretation tasks, and it has the potential to be superior to the conventional machine-learning techniques. Skrede et al. recently constructed a new CNN-based biomarker to predict the outcomes of CR that uses digitally scanned conventional hematoxylin and eosin-stained tumor tissue sections, and they demonstrated promising results (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e). Their findings indicated that the deep-learning classification was superior to the conventional pathological assessment.\u003c/p\u003e \u003cp\u003eWe acknowledge several limitations in our study. It was a retrospective cohort study, and a large prospective study is necessary to validate the performance of our radiomics. In addition, we used a double cross-validation technique to evaluate the performance of our radiomics classifiers because of the limitation of the number of patients and images included in the study; external validation is necessary in future studies.\u003c/p\u003e \u003cp\u003eIn conclusion, we have constructed and validated radiomics classifiers to predict a pCR and CRT response using endoscopic images and a machine-learning method in rectal cancer patients who had undergone NA-CRT. Our radiomics classifiers demonstrated robust performances and may contribute to the decision-making regarding the management of patients with LARCs in clinical settings.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgement:\u0026nbsp;\u003c/strong\u003eNone\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding information:\u0026nbsp;\u003c/strong\u003eThis study was supported by Grant-in-Aid for Young Scientists from Japan Society for Promotion of Science (JSPS KAKENHI Grant Number JP19K16810)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of interest\u003c/strong\u003e: The authors declare no conflict of interes\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics:\u003c/strong\u003e This study was approved by the Teikyo University Ethics Committee (No. 19-127).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions:\u0026nbsp;\u003c/strong\u003eT.O was involved in study concept and design, analysis and interpretation of data and drafting of the manuscript. Y.S was involved in analysis and interpretation of data. T.H was involved in acquisition of data. K.M, and K.N were involved in critical revision of the manuscript for important intellectual content and material support. I. S, K.J and Y.H were involved in study concept and study supervision.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eImperial R, Ahmed Z, Toor OM, Erdogan C, Khaliq A, Case P, et al. Comparative proteogenomic analysis of right-sided colon cancer, left-sided colon cancer and rectal cancer reveals distinct mutational profiles. Mol Cancer. 2018;17(1):177.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYun HR, Lee LJ, Park JH, Cho YK, Cho YB, Lee WY, et al. Local recurrence after curative resection in patients with colon and rectal cancers. Int J Colorectal Dis. 2008;23(11):1081\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBenson AB, Venook AP, Al-Hawary MM, Arain MA, Chen YJ, Ciombor KK, et al. NCCN Guidelines Insights: Rectal Cancer, Version 6.2020. J Natl Compr Canc Netw. 2020;18(7):806\u0026ndash;15.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMinsky B, Cohen A, Enker W, Kelsen D, Kemeny N, Ilson D, et al. Preoperative 5-fluorouracil, low-dose leucovorin, and concurrent radiation therapy for rectal cancer. Cancer. 1994;73(2):273\u0026ndash;80.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGerard JP, Azria D, Gourgou-Bourgade S, Martel-Lafay I, Hennequin C, Etienne PL, et al. Clinical outcome of the ACCORD 12/0405 PRODIGE 2 randomized trial in rectal cancer. J Clin Oncol. 2012;30(36):4558\u0026ndash;65.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMatsusaka S, Ishihara S, Kondo K, Horie H, Uehara K, Oguchi M, et al. A multicenter phase II study of preoperative chemoradiotherapy with S-1 plus oxaliplatin for locally advanced rectal cancer (SHOGUN trial). Radiother Oncol. 2015;116(2):209\u0026ndash;13.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHabr-Gama A, Sabbaga J, Gama-Rodrigues J, Sao Juliao GP, Proscurshim I, Bailao Aguilar P, et al. Watch and wait approach following extended neoadjuvant chemoradiation for distal rectal cancer: are we getting closer to anal cancer management? Dis Colon Rectum. 2013;56(10):1109\u0026ndash;17.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDossa F, Chesney TR, Acuna SA, Baxter NN. A watch-and-wait approach for locally advanced rectal cancer after a clinical complete response following neoadjuvant chemoradiation: a systematic review and meta-analysis. The Lancet Gastroenterology \u0026amp; Hepatology. 2017;2(7):501\u0026ndash;13.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHabr-Gama A, Perez RO, Wynn G, Marks J, Kessler H, Gama-Rodrigues J. Complete clinical response after neoadjuvant chemoradiation therapy for distal rectal cancer: characterization of clinical and endoscopic findings for standardization. Dis Colon Rectum. 2010;53(12):1692\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWatanabe T, Kobunai T, Akiyoshi T, Matsuda K, Ishihara S, Nozawa K. Prediction of response to preoperative chemoradiotherapy in rectal cancer by using reverse transcriptase polymerase chain reaction analysis of four genes. Dis Colon Rectum. 2014;57(1):23\u0026ndash;31.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFischer J, Eglinton TW, Richards SJ, Frizelle FA. Predicting pathological response to chemoradiotherapy for rectal cancer: a systematic review. Expert Rev Anticancer Ther. 2021:1\u0026ndash;12.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKawai K, Ishihara S, Nozawa H, Hata K, Kiyomatsu T, Morikawa T, et al. Prediction of Pathological Complete Response Using Endoscopic Findings and Outcomes of Patients Who Underwent Watchful Waiting After Chemoradiotherapy for Rectal Cancer. Dis Colon Rectum. 2017;60(4):368\u0026ndash;75.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFreischlag K, Sun Z, Adam MA, Kim J, Palta M, Czito BG, et al. Association Between Incomplete Neoadjuvant Radiotherapy and Survival for Patients With Locally Advanced Rectal Cancer. JAMA Surg. 2017;152(6):558\u0026ndash;64.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDiefenhardt M, Ludmir EB, Hofheinz RD, Ghadimi M, Minsky BD, Rodel C, et al. Association of Treatment Adherence With Oncologic Outcomes for Patients With Rectal Cancer: A Post Hoc Analysis of the CAO/ARO/AIO-04 Phase 3 Randomized Clinical Trial. JAMA Oncol. 2020.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLimkin EJ, Sun R, Dercle L, Zacharaki EI, Robert C, Reuze S, et al. Promises and challenges for the implementation of computational medical imaging (radiomics) in oncology. Ann Oncol. 2017;28(6):1191\u0026ndash;206.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShaikh FA, Kolowitz BJ, Awan O, Aerts HJ, von Reden A, Halabi S, et al. Technical Challenges in the Clinical Application of Radiomics. JCO Clin Cancer Inform. 2017;1:1\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSkrede O-J, De Raedt S, Kleppe A, Hveem TS, Liest\u0026oslash;l K, Maddison J, et al. Deep learning for prediction of colorectal cancer outcome: a discovery and validation study. The Lancet. 2020;395(10221):350\u0026ndash;60.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang YQ, Liang CH, He L, Tian J, Liang CS, Chen X, et al. Development and Validation of a Radiomics Nomogram for Preoperative Prediction of Lymph Node Metastasis in Colorectal Cancer. J Clin Oncol. 2016;34(18):2157\u0026ndash;64.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang L, Dong D, Fang M, Zhu Y, Zang Y, Liu Z, et al. Can CT-based radiomics signature predict KRAS/NRAS/BRAF mutations in colorectal cancer? Eur Radiol. 2018;28(5):2058\u0026ndash;67.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDercle L, Lu L, Schwartz LH, Qian M, Tejpar S, Eggleton P, et al. Radiomics Response Signature for Identification of Metastatic Colorectal Cancer Sensitive to Therapies Targeting EGFR Pathway. J Natl Cancer Inst. 2020;112(9):902\u0026ndash;12.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu Z, Zhang XY, Shi YJ, Wang L, Zhu HT, Tang Z, et al. Radiomics Analysis for Evaluation of Pathological Complete Response to Neoadjuvant Chemoradiotherapy in Locally Advanced Rectal Cancer. Clin Cancer Res. 2017;23(23):7253\u0026ndash;62.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePeng H, Dong D, Fang MJ, Li L, Tang LL, Chen L, et al. Prognostic Value of Deep Learning PET/CT-Based Radiomics: Potential Role for Future Individual Induction Chemotherapy in Advanced Nasopharyngeal Carcinoma. Clin Cancer Res. 2019;25(14):4271\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang CM, Huang MY, Huang CW, Tsai HL, Su WC, Chang WC, et al. Machine learning for predicting pathological complete response in patients with locally advanced rectal cancer after neoadjuvant chemoradiotherapy. Sci Rep. 2020;10(1):12555.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHamerla G, Meyer HJ, Hambsch P, Wolf U, Kuhnt T, Hoffmann KT, et al. Radiomics Model Based on Non-Contrast CT Shows No Predictive Power for Complete Pathological Response in Locally Advanced Rectal Cancer. Cancers (Basel). 2019;11(11).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLovinfosse P, Polus M, Van Daele D, Martinive P, Daenen F, Hatt M, et al. FDG PET/CT radiomics for predicting the outcome of locally advanced rectal cancer. Eur J Nucl Med Mol Imaging. 2018;45(3):365\u0026ndash;75.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang C, Jiang ZK, Liu LH, Zeng MS. Pre-treatment ADC image-based random forest classifier for identifying resistant rectal adenocarcinoma to neoadjuvant chemoradiotherapy. Int J Colorectal Dis. 2020;35(1):101\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBulens P, Couwenberg A, Intven M, Debucquoy A, Vandecaveye V, Van Cutsem E, et al. Predicting the tumor response to chemoradiotherapy for rectal cancer: Model development and external validation using MRI radiomics. Radiother Oncol. 2020;142:246\u0026ndash;52.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKiyomatsu T, Watanabe T, Muto T, Nagawa H. The 4-portal technique decreases adverse effects in preoperative radiotherapy for advanced rectal cancer: comparison between the 2-portal and the 4-portal techniques. Am J Surg. 2007;194(4):542\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHashiguchi Y, Muro K, Saito Y, Ito Y, Ajioka Y, Hamaguchi T, et al. Japanese Society for Cancer of the Colon and Rectum (JSCCR) guidelines 2019 for the treatment of colorectal cancer. Int J Clin Oncol. 2019.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVallieres M, Freeman CR, Skamene SR, El Naqa I. A radiomics model from joint FDG-PET and MRI texture features for the prediction of lung metastases in soft-tissue sarcomas of the extremities. Phys Med Biol. 2015;60(14):5471\u0026ndash;96.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChino A, Konishi T, Ogura A, Kawachi H, Osumi H, Yoshio T, et al. Endoscopic criteria to evaluate tumor response of rectal cancer to neoadjuvant chemoradiotherapy using magnifying chromoendoscopy. Eur J Surg Oncol. 2018;44(8):1247\u0026ndash;53.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSohn DK, Han KS, Kim BC, Hong CW, Chang HJ, Baek JY, et al. Endoscopic assessment of tumor regression after preoperative chemoradiotherapy as a prognostic marker in locally advanced rectal cancer. Surg Oncol. 2017;26(4):453\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIwatate Y, Hoshino I, Yokota H, Ishige F, Itami M, Mori Y, et al. Radiogenomics for predicting p53 status, PD-L1 expression, and prognosis with machine learning in pancreatic cancer. Br J Cancer. 2020;123(8):1253\u0026ndash;61.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBadic B, Hatt M, Durand S, Jossic-Corcos CL, Simon B, Visvikis D, et al. Radiogenomics-based cancer prognosis in colorectal cancer. Sci Rep. 2019;9(1):9743.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBibault JE, Giraud P, Housset M, Durdux C, Taieb J, Berger A, et al. Deep Learning and Radiomics predict complete response after neo-adjuvant chemoradiation for locally advanced rectal cancer. Sci Rep. 2018;8(1):12611.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Chemoradiotherapy (CRT), Endoscopy, Machine learning, Radiomics, Rectal cancer","lastPublishedDoi":"10.21203/rs.3.rs-1717256/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-1717256/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eObjective\u003c/h2\u003e \u003cp\u003eThe accuracy of identifying a CRT-response by a pre-operative radiological examination is limited, and another approach is necessary. We constructed endoscopic image-based radiomics classifiers to predict the response of locally advanced rectal cancers (LARCs) to neoadjuvant chemotherapy (NA-CRT).\u003c/p\u003e\u003ch2\u003eDesign:\u003c/h2\u003e \u003cp\u003eWe enrolled 90 patients who had undergone NA-CRT followed by surgery. We selected 5,255 pre- and post-CRT endoscopic images of the tumors and extracted 860 texture features from each image. Using the extracted texture features, we applied 12 machine learning models to construct radiomics classifiers. The performances of the radiomics classifiers to predict a pathological complete response (pCR) and a CRT-response (good or not) were evaluated with a double cross-validation technique by calculating the area under the receiver operating characteristics curve (AUROC).\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eIn the prediction of pCR using pre-CRT images, the cubic support vector machine (SVM) model showed the highest AUROC of 0.816 in the image-basis analysis, and the AUROC of 0.904 in the patient-basis analysis. In the prediction of pCR using post-CRT images, the quadratic SVM model showed the highest AUROC (0.839) in the image-basis analysis, and the AUROC of 0.958 in the patient-basis analysis. The Gaussian SVM model demonstrated the best performances to predict a CRT-response in both pre- and post-CRT images with AUROCs of 0.792 and 0.841 in the image-basis analysis and 0.857 and 0.950 in the patient-basis analysis, respectively.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eOur endoscopic image-based radiomics classifiers demonstrated robust performances to predict the CRT-response of LARCs and may contribute to the patients' management.\u003c/p\u003e","manuscriptTitle":"Endoscopic Image-based Radiomics Classifiers for the Prediction of Neoadjuvant Chemoradiotherapy Response in Rectal Cancer Patients","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2022-06-07 19:56:58","doi":"10.21203/rs.3.rs-1717256/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"17ebf678-e07f-4894-8cb9-8cdfdf556509","owner":[],"postedDate":"June 7th, 2022","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-05-18T11:18:41+00:00","versionOfRecord":[],"versionCreatedAt":"2022-06-07 19:56:58","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-1717256","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-1717256","identity":"rs-1717256","version":["v1"]},"buildId":"wkd9cAKh3xipqaVuRTnlI","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}