Gastric Lymph Node Cancer Detection Using Multiple Features Support Vector Machine for Pathology Diagnosis Support System

Abstract

In this paper, an automatic cancer detection method that combines multiple features to support pathologists is presented. Cancer is the most cause of death in Japan, and patients suffering with and who die of cancer are increasing every year, while the number of pathologists is almost constant. Such issues increase the burden on the pathologists and causes service degradation for the patients. One of the ways which resolve these pathologists’ issues is a double checking by pathologists and systems. The method was proposed for detecting cancer in the Pathology diagnosis support system to introduce a double checking. The proposed method combined three image features, Higher-order Local Auto-Correlation (HLAC) feature, Wavelet feature, Delaunay feature, in varying weights. At first, the features was calculated from HE stained gastric lymph node images. We connected each feature into one vector of varying combinations of the features, and discriminate cancer and no cancer by Support Vector Machine (SVM). Cancer detection rates with most combinations of more than two features were better than just one feature. In addition, by changing the scale of Delaunay in 35-order HLAC, Delaunay and Wavelet combination vector, sensitivity was improved. In the best performance, sensitivity and specificity were 95.7% and 82.1% respectively. Therefore, the proposed method can be used for a double check system.