The ML-based methods were carried out to predict the metastasis in patients with GC in the study.
Some criteria such as sensitivity, spesificity, presicion and AUC can be extracted from ROC curve analysis to assess the ML-based models performance. Among those models, RF can be better predictive model.
Some ML methods have been proposed in patients with GC until now. Some of them have used ML techniques to study the relationship between lncRNAs and complex diseases in Gene datasets, while some others applied them, including SVM, RF, NN and deep learning as predicting metastasis situation [13–16].
A study assesses the performances of seven different ML methods, such as LR, SVM, RF, Lasso regression (LASSO), Sparse neural network (sNNR), Extreme gradient boosting (XGboost) and Stochastic gradient boosting (SGB) to predict GC risk after H. pylori eradication [17]. Based on period of eradication therapy, the data was divided into two train and test datasets. The AUC was obtained to calculate the model performance. The results of the study revealed that the Extreme gradient boosting (XGboost) was considered as the most successful among all seven models; however, the SVM was had the lowest sensitivity, specificity and AUC. In their study, age, smoking, drinking, comorbities, Need of Helicobacter pylory retreatment, medications were significant in both high and low- risk patients. On the contrary, size of tomur and age were considered as essential variables, which was compatible in our study that age in RF act as significant variable.
Akcay et al. (2020) investigated the overall survival (OS) and recurrence patterns by ML algorithms in patients with Radiation Therapy [18]. The goal of the study was to fit the ML approaches, including LR, XGBoost, SVM, RF, multilayer perceptron (MLP) and Gaussian Naive Bayes (GNB) in the assessment of the overall survival (OS), distant metastasis (DM), and peritoneal recurrence (PR) prediction. The best performance models in the prediction of OS, distant metastases, and peritoneal metastases were discoverd to be GNB, XGBoost and RF, respectively. Also, in their study GNB was considered as the better model to evaluate the OS, but all ML-based approaches had ideal performances in our study and it seems that RF act as a better model among six methods.
A clinical study applied ML approaches to predict lymph node metastasis in GC patients [19]. The result of the stugy showed that tomur size, grade of tomur, depth of tomur and age were significant (P < 0.001). Also, their results presented that neural network (NN) had and gradient boosting machine (GBM) methods had the minimum sensitivity and specificity among seven ML-basd algorithms, respectively. In our study, the RF model was regarded as the succeful model among six methods, and tomur size, age were significant that is compatible with the study.
Zhou et al. established the prediction of metastasis of patients with GC using five techniques of MLs [20]. Those methods were LR, RF, DT, Gradient Boosting Machine and Light Gradient Boosting. The most and the least AUC were related to gbm and DT, respectively. Furthermore, the prime variables were tomur size, pathological type and depth of invasio in the study; nevertheless, tomur size, age, grade of tomur, the number of lumph node involved, treatment type, BMI, marital status and history of smoking were significant in our study. Also, the precision of RF model are a little better than other models.