Meniscus injury prediction model based on metric learning

Abstract

A meniscus injury is a prevalent condition affecting the knee joint. The construction of a subjective prediction model for meniscus injury represents a potentially invaluable diagnostic tool for physicians. Nevertheless, given the variability of pathological manifestations among individual patients, machine learning-based models may produce errors when attempting to predict specific medical records. In order to mitigate this issue, the present study suggests the incorporation of metric learning within the machine learning (ML) modelling process, with the aim of reducing the intra-class spacing of comparable samples and thereby enhancing the classification accuracy of individual medical records. This work has not yet been attempted in the field of knee joint prediction. The findings demonstrate that the adoption of metric learning produces better optimal outcomes. Compared to machine learning baseline models, F1 was increased by 2%.