Artificial Intelligence in Endoscopy.
Digestive diseases and sciences
Artificial intelligence (AI) is rapidly developing in various medical fields, and there is an increase in research performed in the field of gastrointestinal (GI) endoscopy. In particular, the advent of convolutional neural network, which is a class of deep learning method, has the potential to revolutionize the field of GI endoscopy, including esophagogastroduodenoscopy (EGD), capsule endoscopy (CE), and colonoscopy. A total of 149 original articles pertaining to AI (27 articles in esophagus, 30 articles in stomach, 29 articles in CE, and 63 articles in colon) were identified in this review. The main focuses of AI in EGD are cancer detection, identifying the depth of cancer invasion, prediction of pathological diagnosis, and prediction of Helicobacter pylori infection. In the field of CE, automated detection of bleeding sites, ulcers, tumors, and various small bowel diseases is being investigated. AI in colonoscopy has advanced with several patient-based prospective studies being conducted on the automated detection and classification of colon polyps. Furthermore, research on inflammatory bowel disease has also been recently reported. Most studies of AI in the field of GI endoscopy are still in the preclinical stages because of the retrospective design using still images. Video-based prospective studies are needed to advance the field. However, AI will continue to develop and be used in daily clinical practice in the near future. In this review, we have highlighted the published literature along with providing current status and insights into the future of AI in GI endoscopy.
10.1007/s10620-021-07086-z
Use of Artificial Intelligence in the Prediction of Malignant Potential of Gastric Gastrointestinal Stromal Tumors.
Digestive diseases and sciences
BACKGROUND AND AIMS:This study aimed to investigate whether AI via a deep learning algorithm using endoscopic ultrasonography (EUS) images could predict the malignant potential of gastric gastrointestinal stromal tumors (GISTs). METHODS:A series of patients who underwent EUS before surgical resection for gastric GISTs were included. A total of 685 images of GISTs from 55 retrospectively included patients were used as the training data set for the AI system. Convolutional neural networks were constructed to build a deep learning model. After applying the synthetic minority oversampling technique, 70% of the generated images were used for AI training and 30% were used to test AI diagnoses. Next, validation was performed using 153 EUS images of 15 patients with GISTs. In addition, conventional EUS features of 55 patients in the training cohort were evaluated to predict the malignant potential of GISTs and mitotic index. RESULTS:The overall sensitivity, specificity, and accuracy of the AI system for predicting malignancy risk were 83%, 94%, and 82% in the training dataset, and 75%, 73%, and 66% in the validation cohort, respectively. When patients were divided into low-risk and high-risk groups, sensitivity, specificity, and accuracy increased to 99% in the training dataset and 99.7%, 99.7%, and 99.6%, respectively, in the validation cohort. No conventional EUS features were found to be associated with either malignant potential or mitotic index (P > 0.05). CONCLUSIONS:AI via a deep learning algorithm using EUS images could predict the malignant potential of gastric GISTs with high accuracy.
10.1007/s10620-021-06830-9
Convolutional neural network-based object detection model to identify gastrointestinal stromal tumors in endoscopic ultrasound images.
Oh Chang Kyo,Kim Taewan,Cho Yu Kyung,Cheung Dae Young,Lee Bo-In,Cho Young-Seok,Kim Jin Il,Choi Myung-Gyu,Lee Han Hee,Lee Seungchul
Journal of gastroenterology and hepatology
BACKGROUND AND AIM:We aimed to develop a convolutional neural network (CNN)-based object detection model for the discrimination of gastric subepithelial tumors, such as gastrointestinal stromal tumors (GISTs), and leiomyomas, in endoscopic ultrasound (EUS) images. METHODS:We used 376 images from 114 patients with histologically confirmed gastric GIST or leiomyoma to train the EUS-CNN. We constructed the EUS-CNN using an EfficientNet CNN model for feature extraction and a weighted bi-directional feature pyramid network for object detection. We assessed the performance of our EUS-CNN by calculating its accuracy, sensitivity, specificity, and area under receiver operating characteristic curve (AUC) using a validation set of 170 images from 54 patients. Four EUS experts and 15 EUS trainees were asked to judge the same validation dataset, and the diagnostic yields were compared between the EUS-CNN and human assessments. RESULTS:In the per-image analysis, the sensitivity, specificity, accuracy, and AUC of our EUS-CNN were 95.6%, 82.1%, 91.2%, and 0.9234, respectively. In the per-patient analysis, the sensitivity, specificity, accuracy, and AUC for our object detection model were 100.0%, 85.7%, 96.3%, and 0.9929, respectively. The EUS-CNN outperformed human assessment in terms of accuracy, sensitivity, and negative predictive value. CONCLUSIONS:We developed the EUS-CNN system, which demonstrated high diagnostic ability for gastric GIST prediction. This EUS-CNN system can be helpful not only for less-experienced endoscopists but also for experienced ones. Additional EUS image accumulation and prospective studies are required alongside validation in a large multicenter trial.
10.1111/jgh.15653
Comparison of Computed Tomography Features of Gastric and Small Bowel Gastrointestinal Stromal Tumors With Different Risk Grades.
Tang Bo,Feng Qiu-Xia,Liu Xi-Sheng
Journal of computer assisted tomography
OBJECTIVE:This study aimed to compare the computed tomography (CT) features of gastric and small bowel gastrointestinal stromal tumors (GISTs) and further identify the predictors for risk stratification of them, respectively. METHODS:According to the modified National Institutes of Health criteria, patients were classified into low-malignant potential group and high-malignant potential group. Two experienced radiologists reviewed the CT features including the difference of CT values between arterial phase and portal venous phase (PVPMAP) by consensus. The CT features of gastric and small bowel GISTs were compared, and the association of CT features with risk grades was analyzed, respectively. Determinant CT features were used to construct corresponding models. RESULTS:Univariate analysis showed that small bowel GISTs tended to present with irregular contour, mixed growth pattern, ill-defined margin, severe necrosis, ulceration, tumor vessels, heterogeneous enhancement, larger size, and marked enhancement compared with gastric GISTs. According to multivariate analysis, tumor size (P < 0.001; odds ratio [OR], 3.279), necrosis (P = 0.008; OR, 2.104) and PVPMAP (P = 0.045; OR, 0.958) were the independent influencing factors for risk stratification of gastric GISTs. In terms of small bowel GISTs, the independent predictors were tumor size (P < 0.001; OR, 3.797) and ulceration (P = 0.031; OR, 4.027). Receiver operating characteristic curve indicated that the CT models for risk stratification of gastric and small bowel GISTs both achieved the best predictive performance. CONCLUSIONS:Computed tomography features of gastric and small bowel GISTs are different. Furthermore, the qualitative and quantitative CT features of GISTs may be favorable for preoperative risk stratification.
10.1097/RCT.0000000000001262
Simple Scoring Model Based on Enhanced CT in Preoperative Prediction of Biological Risk of Gastrointestinal Stromal Tumor.
Technology in cancer research & treatment
To construct a simple scoring model for predicting the biological risk of gastrointestinal stromal tumors based on enhanced computed tomography (CT) features. The clinicopathological and imaging data of 149 patients with primary gastrointestinal stromal tumor were retrospectively analyzed in our hospital. According to the risk classification, the patients were divided into low-risk group and high-risk group. The features of enhanced CT were observed and recorded. Univariate and multivariate logistic regression models were used to determine the predictors of high-risk biological behaviors of gastrointestinal stromal tumor, and then a simple scoring model was constructed according to the regression coefficients of each predictor. The receiver operating characteristic curve was used to evaluate the predictive ability of the model. There was no significant difference between the risk classification of gastrointestinal stromal tumor with gender and age ( = .168, .320), while significant difference was found between the tumor size and location ( < .001). Univariate and multivariate logistic regression analyses showed that tumor size, enlarged vessels feeding or draining the mass, peritumoral lymph node enlargement, and venous phase contrast enhancement rate were independent predictors of the biological risk of gastrointestinal stromal tumor ( < .05). The area under the curve value of tumor size, enlarged vessels feeding or draining the mass, peritumoral lymph node enlargement, and venous phase contrast enhancement rate as the high-risk predictor of gastrointestinal stromal tumor were 0.955, 0.729, 0.680, and 0.807, respectively. Receiver operating characteristic curve results showed that the area under the curve of the scoring model constructed based on enhanced CT features was 0.941 (95% confidence interval: 0.891-0.973). When the total score was >1, the sensitivity of the scoring model in diagnosing gastrointestinal stromal tumor was 85.58%, the specificity was 88.89%, the positive predictive value was 88.51%, the negative predictive value was 86.04%, and the accuracy was 86.18%. The results of DeLong test showed that the area under the curve of the scoring model was better than that of the receiver operating characteristic curve of tumor size, enlarged vessels feeding or draining the mass, peritumoral lymph node enlargement, venous phase contrast enhancement rate, and other indicators alone in predicting the high risk of gastrointestinal stromal tumor, and the differences were statistically significant (Z = 26.510, < .001; Z = 3.992, < .001; Z = 6.353, < .001; Z = 4.052, = .013). The simple scoring model based on enhanced CT features is a simple and practical clinical prediction model, which is helpful to make preoperative individualized treatment plan and improve the prognosis of gastrointestinal stromal tumor patients.
10.1177/15330338231194502
Relationship between multi-slice computed tomography features and pathological risk stratification assessment in gastric gastrointestinal stromal tumors.
World journal of gastrointestinal oncology
BACKGROUND:Computed tomography (CT) imaging features are associated with risk stratification of gastric gastrointestinal stromal tumors (GISTs). AIM:To determine the multi-slice CT imaging features for predicting risk stratification in patients with primary gastric GISTs. METHODS:The clinicopathological and CT imaging data for 147 patients with histologically confirmed primary gastric GISTs were retrospectively analyzed. All patients had received dynamic contrast-enhanced CT (CECT) followed by surgical resection. According to the modified National Institutes of Health criteria, 147 lesions were classified into the low malignant potential group (very low and low risk; 101 lesions) and high malignant potential group (medium and high-risk; 46 lesions). The association between malignant potential and CT characteristic features (including tumor location, size, growth pattern, contour, ulceration, cystic degeneration or necrosis, calcification within the tumor, lymphadenopathy, enhancement patterns, unenhanced CT and CECT attenuation value, and enhancement degree) was analyzed using univariate analysis. Multivariate logistic regression analysis was performed to identify significant predictors of high malignant potential. The receiver operating curve (ROC) was used to evaluate the predictive value of tumor size and the multinomial logistic regression model for risk classification. RESULTS:There were 46 patients with high malignant potential and 101 with low-malignant potential gastric GISTs. Univariate analysis showed no significant differences in age, gender, tumor location, calcification, unenhanced CT and CECT attenuation values, and enhancement degree between the two groups ( > 0.05). However, a significant difference was observed in tumor size (3.14 ± 0.94 6.63 ± 3.26 cm, < 0.001) between the low-grade and high-grade groups. The univariate analysis further revealed that CT imaging features, including tumor contours, lesion growth patterns, ulceration, cystic degeneration or necrosis, lymphadenopathy, and contrast enhancement patterns, were associated with risk stratification ( < 0.05). According to binary logistic regression analysis, tumor size [ < 0.001; odds ratio (OR) = 26.448; 95% confidence interval (CI): 4.854-144.099)], contours ( = 0.028; OR = 7.750; 95%CI: 1.253-47.955), and mixed growth pattern ( = 0.046; OR = 4.740; 95%CI: 1.029-21.828) were independent predictors for risk stratification of gastric GISTs. ROC curve analysis for the multinomial logistic regression model and tumor size to differentiate high-malignant potential from low-malignant potential GISTs achieved a maximum area under the curve of 0.919 (95%CI: 0.863-0.975) and 0.940 (95%CI: 0.893-0.986), respectively. The tumor size cutoff value between the low and high malignant potential groups was 4.05 cm, and the sensitivity and specificity were 93.5% and 84.2%, respectively. CONCLUSION:CT features, including tumor size, growth patterns, and lesion contours, were predictors of malignant potential for primary gastric GISTs.
10.4251/wjgo.v15.i6.1073
A CT-based nomogram for predicting the malignant potential of primary gastric gastrointestinal stromal tumors preoperatively.
Li Chang,Fu Wenhao,Huang Li,Chen Yingqian,Xiang Pei,Guan Jian,Sun Canhui
Abdominal radiology (New York)
PURPOSE:To develop and validate a computerized tomography (CT)-based nomogram for predicting the malignant potential of primary gastric gastrointestinal stromal tumors (GISTs). METHODS:The primary and validation cohorts consisted of 167 and 39 patients (single center, different time periods) with histologically confirmed primary gastric GISTs. Clinical data and preoperative CT images were reviewed. The association of CT characteristics with malignant potential was analyzed using univariate and stepwise logistic regression analyses. A nomogram based on significant CT findings was developed for predicting malignant potential. The predictive accuracy of the nomogram was determined by the concordance index (C-index) and calibration curves. External validation was performed with the validation cohort. RESULTS:CT imaging features including tumor size, tumor location, tumor necrosis, growth pattern, ulceration, enlarged vessels feeding or draining the mass (EVFDM), tumor contour, mesenteric fat infiltration, and direct organ invasion showed significant differences between the low- and high-grade malignant potential groups in univariate analysis (P < 0.05). Only tumor size (> 5 cm vs ≤ 5 cm), location (cardiac/pericardial region vs other), EVFDM, and mesenteric fat infiltration (present vs absent) were significantly associated with high malignant potential in multivariate logistic regression analysis. Incorporating these four independent factors into the nomogram model achieved good C-indexes of 0.946 (95% confidence interval [CI] 0.899-0.975) and 0.952 (95% CI 0.913-0.977) in the primary and validation cohorts, respectively. The cutoff point was 0.33, with sensitivity, specificity, and diagnostic accuracy of 0.865, 0.915, and 0.780, respectively. DISCUSSION:Primary gastric GISTs originating in the cardiac/pericardial region appear to be associated with higher malignant potential. The nomogram consisting of CT features, including size, location, EVFDM, and mesenteric fat infiltration, could be used to accurately predict the high malignant potential of primary gastric GISTs.
10.1007/s00261-021-03026-7
Risk stratification for 1- to 2-cm gastric gastrointestinal stromal tumors: visual assessment of CT and EUS high-risk features versus CT radiomics analysis.
European radiology
OBJECTIVES:To investigate the ability of CT and endoscopic sonography (EUS) in predicting the malignant risk of 1-2-cm gastric gastrointestinal stromal tumors (gGISTs) and to clarify whether radiomics could be applied for risk stratification. METHODS:A total of 151 pathologically confirmed 1-2-cm gGISTs from seven institutions were identified by contrast-enhanced CT scans between January 2010 and March 2021. A detailed description of EUS morphological features was available for 73 gGISTs. The association between EUS or CT high-risk features and pathological malignant potential was evaluated. gGISTs were randomly divided into three groups to build the radiomics model, including 74 in the training cohort, 37 in validation cohort, and 40 in testing cohort. The ROIs covering the whole tumor volume were delineated on the CT images of the portal venous phase. The Pearson test and least absolute shrinkage and selection operator (LASSO) algorithm were used for feature selection, and the ROC curves were used to evaluate the model performance. RESULTS:The presence of EUS- and CT-based morphological high-risk features, including calcification, necrosis, intratumoral heterogeneity, irregular border, or surface ulceration, did not differ between very-low and intermediate risk 1-2-cm gGISTs (p > 0.05). The radiomics model consisting of five radiomics features showed favorable performance in discrimination of malignant 1-2-cm gGISTs, with the AUC of the training, validation, and testing cohort as 0.866, 0.812, and 0.766, respectively. CONCLUSIONS:Instead of CT- and EUS-based morphological high-risk features, the CT radiomics model could potentially be applied for preoperative risk stratification of 1-2-cm gGISTs. KEY POINTS:• The presence of EUS- and CT-based morphological high-risk factors, including calcification, necrosis, intratumoral heterogeneity, irregular border, or surface ulceration, did not correlate with the pathological malignant potential of 1-2-cm gGISTs. • The CT radiomics model could potentially be applied for preoperative risk stratification of 1-2-cm gGISTs.
10.1007/s00330-022-09228-x
Preoperative CT-based radiomics and deep learning model for predicting risk stratification of gastric gastrointestinal stromal tumors.
Medical physics
BACKGROUND:Gastrointestinal stromal tumors (GISTs) are clinically heterogeneous with various malignant potential in different individuals. It is crucial to explore a reliable method for preoperative risk stratification of gastric GISTs noninvasively. PURPOSE:To establish and evaluate a machine learning model using the combination of computed tomography (CT) morphology, radiomics, and deep learning features to predict the risk stratification of primary gastric GISTs preoperatively. METHODS:The 193 gastric GISTs lesions were randomly divided into training set, validation set, and test set in a ratio of 6:2:2. The qualitative and quantitative CT morphological features were assessed by two radiologists. The tumors were segmented manually, and then radiomic features were extracted using PyRadiomics and the deep learning features were extracted using pre-trained Resnet50 from arterial phase and venous phase CT images, respectively. Pearson correlation analysis and recursive feature elimination were used for feature selection. Support vector machines were employed to build a classifier for predicting the risk stratification of GISTs. This study compared the performance of models using different pre-trained convolutional neural networks (CNNs) to extract deep features for classification, as well as the performance of modeling features from single-phase and dual-phase images. The arterial phase, venous phase and dual-phase machine learning models were built, respectively, and the morphological features were added to the dual-phase machine learning model to construct a combined model. Receiver operating characteristic (ROC) curves were used to evaluate the efficacy of each model. The clinical application value of the combined model was determined through the decision curve analysis (DCA) and the net reclassification index (NRI) was analyzed. RESULTS:The area under the curve (AUC) of the dual-phase machine learning model was 0.876, which was higher than that of the arterial phase model or venous phase model (0.813, 0.838, respectively). The combined model had best predictive performance than the above models with an AUC of 0.941 (95% CI: 0.887-0.974) (p = 0.012, Delong test). DCA demonstrated that the combined model had good clinical application value with an NRI of 0.575 (95% CI: 0.357-0.891). CONCLUSION:In this study, we established a combined model that incorporated dual-phase morphology, radiomics, and deep learning characteristics, which can be used to predict the preoperative risk stratification of gastric GISTs.
10.1002/mp.17276
Development of a Malignancy Potential Binary Prediction Model Based on Deep Learning for the Mitotic Count of Local Primary Gastrointestinal Stromal Tumors.
Yang Jiejin,Chen Zeyang,Liu Weipeng,Wang Xiangpeng,Ma Shuai,Jin Feifei,Wang Xiaoying
Korean journal of radiology
OBJECTIVE:The mitotic count of gastrointestinal stromal tumors (GIST) is closely associated with the risk of planting and metastasis. The purpose of this study was to develop a predictive model for the mitotic index of local primary GIST, based on deep learning algorithm. MATERIALS AND METHODS:Abdominal contrast-enhanced CT images of 148 pathologically confirmed GIST cases were retrospectively collected for the development of a deep learning classification algorithm. The areas of GIST masses on the CT images were retrospectively labelled by an experienced radiologist. The postoperative pathological mitotic count was considered as the gold standard (high mitotic count, > 5/50 high-power fields [HPFs]; low mitotic count, ≤ 5/50 HPFs). A binary classification model was trained on the basis of the VGG16 convolutional neural network, using the CT images with the training set (n = 108), validation set (n = 20), and the test set (n = 20). The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated at both, the image level and the patient level. The receiver operating characteristic curves were generated on the basis of the model prediction results and the area under curves (AUCs) were calculated. The risk categories of the tumors were predicted according to the Armed Forces Institute of Pathology criteria. RESULTS:At the image level, the classification prediction results of the mitotic counts in the test cohort were as follows: sensitivity 85.7% (95% confidence interval [CI]: 0.834-0.877), specificity 67.5% (95% CI: 0.636-0.712), PPV 82.1% (95% CI: 0.797-0.843), NPV 73.0% (95% CI: 0.691-0.766), and AUC 0.771 (95% CI: 0.750-0.791). At the patient level, the classification prediction results in the test cohort were as follows: sensitivity 90.0% (95% CI: 0.541-0.995), specificity 70.0% (95% CI: 0.354-0.919), PPV 75.0% (95% CI: 0.428-0.933), NPV 87.5% (95% CI: 0.467-0.993), and AUC 0.800 (95% CI: 0.563-0.943). CONCLUSION:We developed and preliminarily verified the GIST mitotic count binary prediction model, based on the VGG convolutional neural network. The model displayed a good predictive performance.
10.3348/kjr.2019.0851
Deep learning analysis for differential diagnosis and risk classification of gastrointestinal tumors.
Scandinavian journal of gastroenterology
OBJECTIVES:Recently, artificial intelligence (AI) has been applied to clinical diagnosis. Although AI has already been developed for gastrointestinal (GI) tract endoscopy, few studies have applied AI to endoscopic ultrasound (EUS) images. In this study, we used a computer-assisted diagnosis (CAD) system with deep learning analysis of EUS images (EUS-CAD) and assessed its ability to differentiate GI stromal tumors (GISTs) from other mesenchymal tumors and their risk classification performance. MATERIALS AND METHODS:A total of 101 pathologically confirmed cases of subepithelial lesions (SELs) arising from the muscularis propria layer, including 69 GISTs, 17 leiomyomas and 15 schwannomas, were examined. A total of 3283 EUS images were used for training and five-fold-cross-validation, and 827 images were independently tested for diagnosing GISTs. For the risk classification of 69 GISTs, including very-low-, low-, intermediate- and high-risk GISTs, 2,784 EUS images were used for training and three-fold-cross-validation. RESULTS:For the differential diagnostic performance of GIST among all SELs, the accuracy, sensitivity, specificity and area under the receiver operating characteristic (ROC) curve were 80.4%, 82.9%, 75.3% and 0.865, respectively, whereas those for intermediate- and high-risk GISTs were 71.8%, 70.2%, 72.0% and 0.771, respectively. CONCLUSIONS:The EUS-CAD system showed a good diagnostic yield in differentiating GISTs from other mesenchymal tumors and successfully demonstrated the GIST risk classification feasibility. This system can determine whether treatment is necessary based on EUS imaging alone without the need for additional invasive examinations.
10.1080/00365521.2024.2368241