Ensemble learning of tissue components for prostate histopathology image grading

Abstract

Ensemble learning is an effective machine learning approach to improve the prediction performance by fusing several single classifier models. In computer-aided diagnosis system (CAD), machine learning has become one of the dominant solutions for tissue images diagnosis and grading. One problem in a single classifier model for multi-components of the tissue images combination to construct dense feature vectors is the overfitting. In this paper, an ensemble learning for multi-component tissue images classification approach is proposed. The prostate cancer Hematoxylin and Eosin (H&E) histopathology images from HUKM were used to test the proposed ensemble approach for diagnosing and Gleason grading. The experiments results of several prostate classification tasks, namely, benign vs. Grade 3, benign vs.Grade4, and Grade 3vs.Grade 4 show that the proposed ensemble significantly outperforms the previous typical CAD and the naïve approach that combines the texture features of all tissue component directly in dense feature vectors for a classifier.