Discriminative Multi-feature Representation for Renal Cancer Detection based on Histopathology Images

Renal cancer is one of the most common cancers in the world, and early diagnosis can increase the possibility of successful treatment and survival rate. However, manual detection is time-consuming and relies heavily on the experience of pathologists. Therefore, it is desirable to employ a computer-aided approach to automate the diagnostic process thereby saving time and labor. To date, a substantial amount of research with common deep learning methods have been applied to address this issue. However, deep learning methods require large numbers of images to train the model. Alternatively, traditional machine learning methods such as texture feature extractors can reach a reasonable result with a smaller computing cost. In this paper, we extensively study the efficiency of texture features extracted from histopathology images at detecting kidney cancer by adopting a weighted fusion method of HOG and GLCM, which includes both local structural features and full texture information from the histopathology images. We applied the proposed method on a histopathology image data set containing 93 patients with renal cancer and 150 patients with normal kidneys. The experimental results indicate that our method can achieve a similar outcome to deep learning methods, while reducing the computing time.