Tumour mutation burden status detection method based on multiple examples combined with clustering algorithm

Traditional tumour mutation burden (TMB) detection methods suffer from high cost, complex operation and time consuming problems. In this paper, a TMB detection method based on multiple instance learning (MIL) combined with a clustering algorithm is proposed. Multiple instance learning is used to solve the problem of difficult labelling of pathology images after high-resolution segmentation; the introduction of multiple feature extractors enhances the diversity of image features. The clustering algorithm is used to cluster the pathology images and randomly extract features to combine them into a new training set, which maintains high prediction performance even with a small amount of data. Comparison experiments with other models show that the method exhibits better performance in TMB state detection.