Automated acquisition of knowledge beyond pathologists
Yoichiro Yamamoto, Toyonori Tsuzuki, Jun Akatsuka, Masao Ueki, Hiromu Morikawa, Yasushi Numata, Taishi Takahara, Takuji Tsuyuki, Akira Shimizu, Ichiro Maeda, Shinichi Tsuchiya, Hiroyuki Kanno, Yukihiro Kondo, Manabu Fukumoto, Gen Tamiya, Naonori Ueda, and Go Kimura
Received Date: 14th December 18
Deep learning algorithms have been successfully used in medical image classification and cancer detection. In the next stage, the technology of acquiring explainable knowledge from medical images is highly desired. Herein, fully automated acquisition of explainable features from annotation-free histopathological images is achieved via revealing statistical distortions in datasets by introducing the way of pathologists’ examination into a set of deep neural networks. As validation, we compared the prediction accuracy of prostate cancer recurrence using our algorithm-generated features with that of diagnosis by an expert pathologist using established criteria on 13,188 whole-mount pathology images. Our method found not only the findings established by humans but also features that have not been recognized so far, and showed higher accuracy than human in prognostic prediction. This study provides a new field to the deep learning approach as a novel tool for discovering uncharted knowledge, leading to effective treatments and drug discovery.
Read in full at bioRxiv.
This is an abstract of a preprint hosted on an independent third party site. It has not been peer reviewed but is currently under consideration at Nature Communications.