Establishment of a machine learning model for the risk assessment of Perineural invasion in gastric cancer

Abstract Background and Aims: More and more studies have proved that Perineural Invasion (PNI)plays an important role in cancer development,but the traditional detection methods are cumbersome pathological examinations and extremely dependent on doctors' experience, can not be applied to all hospitals. Therefore, we aim to build a model that predicts PNI using machine learning. Methods Outliers were removed using the Isolation Forest method and eligible patients were divided into training and testing cohorts using the Isolation Forest algorithm, and the data were subjected to binary tree segmentation, sample selection, feature selection and segmentation point selection, all using randomisation. The distributions of categorical variables were compared using the Chi-squared test and Fisher's exact test. AUC, balanced F Score, confusion matrix, Matthews correlation coefficient and diagnostic odds ratio to compare the predictive power of the models. Results The X-tree (random forest) model is a convenient and reliable tool for predicting PNI status in gastric cancer patients using preoperative clinical indicators. It has demonstrated excellent performance with an AUC of 0.97, precision of 0.93, and recall of 0.84 for the test set. Conclusions PNI is not conducive to the survival of gastric cancer patients, and the study established a model for predicting PNI in patients with gastric cancer based on their preoperative clinical characteristics.