Machine learning-based gray-level co-occurrence matrix signature for predicting lymph node metastasis in undifferentiated-type early gastric cancer.

Background: The most important consideration in determining treatment strategies for undifferentiated early gastric cancer (UEGC) is the risk of lymph node metastasis (LNM). Therefore, identifying a potential biomarker that predicts LNM is quite useful in determining treatment.

Aim: To develop a machine learning (ML)-based integral procedure to construct the LNM gray-level co-occurrence matrix (GLCM) prediction model.

Methods: We retrospectively selected 526 cases of UEGC confirmed through pathological examination after radical gastrectomy without endoscopic treatment in four tertiary hospitals between January 2015 to December 2021. We extracted GLCM-based features from grayscale images and applied ML to the classification of candidate predictive variables. The robustness and clinical utility of each model were evaluated based on the following factors: Receiver operating characteristic curve (ROC), decision curve analysis, and clinical impact curve.

Results: GLCM-based feature extraction significantly correlated with LNM. The top 7 GLCM-based factors included inertia value 0° (IV_0), inertia value 45° (IV_45), inverse gap 0° (IG_0), inverse gap 45° (IG_45), inverse gap full angle (IG_all), Haralick 30° (Haralick_30), Haralick full angle (Haralick_all), and Entropy. The areas under the ROC curve (AUCs) of the random forest classifier (RFC) model, support vector machine, eXtreme gradient boosting, artificial neural network, and decision tree ranged from 0.805 [95% confidence interval (CI): 0.258-1.352] to 0.925 (95%CI: 0.378-1.472) in the training set and from 0.794 (95%CI: 0.237-1.351) to 0.912 (95%CI: 0.355-1.469) in the testing set, respectively. The RFC (training set: AUC: 0.925, 95%CI: 0.378-1.472; testing set: AUC: 0.912, 95%CI: 0.355-1.469) model that incorporates Entropy, Haralick_all, Haralick_30, IG_all, IG_45, IG_0, and IV_45 had the highest predictive accuracy.

Conclusion: The evaluation results indicate that the method of selecting radiological and textural features becomes more effective in the LNM discrimination against UEGC patients. Additionally, the ML-based prediction model developed using the RFC can be used to derive treatment options and identify LNM, which can hence improve clinical outcomes.