Automated Classification of Human Histological Images, A Multiple-Instance Learning Approach

In this paper, we apply a multiple-instance learning (MIL) method, MILES (multiple-instance learning via embedded instance selection), to human histological image classification. MILES converts a MIL problem to a supervised learning problem by an instance-based feature mapping. 1-norm SVM is then adopted to select features and construct a classifier simultaneously. MILES identifies the sub-images that reflect underlying category concepts, and use them for classification. Experimental validation is provided based on images from different organs and parts of the body. The new approach demonstrates significantly improved performance in comparison with a method based on a Gaussian mixture model