Development of a Malignancy Potential Binary Prediction Model Based on Deep Learning for the Mitotic Count of Local Primary Gastrointestinal Stromal Tumors

Abstract

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 [

CT and MRI of Gastrointestinal Stromal Tumours: New Trends and Perspectives

“…The AI model yielded 100% sensitivity (95% CI: 65.6-100) and 90% accuracy (95% CI: 75.6-100) in the test cohort for the diagnosis of high mitotic count. 66 Although the test cohort included only 20 patients with a 1:1 ratio of high and low mitotic count GISTs, this study shows a potential for AI in risk assessment of GIST. 66…”

“…65 One study reported the use of a predictive model to discriminate between GISTs with high mitotic count (>5/ 50 mitoses on high-power fields) and those with low mitotic count (≤5/50 mitoses). 66 The model was based on portal venous phase CT images of 108 patients with GISTs that were used to train a deep learning algorithm. The AI model yielded 100% sensitivity (95% CI: 65.6-100) and 90% accuracy (95% CI: 75.6-100) in the test cohort for the diagnosis of high mitotic count.…”

Deep learning and radiomics to predict the mitotic index of gastrointestinal stromal tumors based on multiparametric MRI

“…Previously, a DL model for predicting the mitotic index of GIST was preliminarily established by providing venous images as input into CNN. The results showed that the image-based DL model could evaluate the MI of GIST before surgery ( 40 ). However, the generalization ability of the model proposed in their study was not high, and the area under the curve (AUC) in the internal test set was only 0.771–0.800.…”

“…In addition to the multi-modal image input, the construction process of the proposed hybrid model proposed was different from that of the image-based CNN model reported in previous studies (40). While the hybrid model combined shape features and clinical indicators, in order to ensure the robustness of the model, only shape features in traditional radiomics were selected to establish the model.…”

Gastrointestinal stromal tumors: a comprehensive radiological review

“…Wang et al reported that a machine learning model distinguished gastric schwannomas from GISTs with excellent accuracy (area under the curve: 0.97) [ 100 ]. Yang et al developed a binary prediction model for mitotic count (area under the curve: 0.80) [ 101 ]. Furthermore, Kang et al reported a CT-based deep-learning model that effectively predicted histological risk (low, intermediate, and high risks) with an accuracy of > 0.77 [ 102 ].…”