Computer-aided diagnosis for pathology image

Abstract

Accurate analysis for pathology image is of great importance in medical diagnosis and treatment. Specifically, nucleus detection is considered as an important prerequisite for this purpose. With the rapid development of computer-aided diagnosis, several computer-aided diagnosis (CAD) models using machine learning and deep learning have been developed fo accurate automatic nucleus detection. In this paper, we propose a nucleus detection method using two layers’ sparse autoencoder (SAE) and transfer learning. First, 26832 image patches of breast cancer are utilized to train the SAE in an unsupervised learning method, which could be regarded as the feature extraction process. Then, the softmax classifier are used to classify that whether an image patch contains a complete nucleus or not. Finally, following transfer learning and sliding window techniques, we use the trained SAE and softmax models for nucleus detection on liver cancer pathology image. Experiments demonstrate that our proposed method could achieve the satisfactory detection results.