Jingwei Zhang, Ph.D. Research Proficiency Presentation: 'Classification of Huge Sized Images using Deep Neural Networks'

Friday, June 11, 2021 - 1:00pm to 3:00pm
Zoom - contact events@cs.stonybrook.edu for Zoom info.
Event Description: 
Abstract: Deep neural networks (DNN) have achieved significant success in a variety of computer vision tasks including digital pathology. Huge sized images are becoming more and more pervasive especially in the digital pathology field. Yet traditional DNNs usually can not scale well on these images due to computational and memory constraints. In this report, we conduct a comprehensive literature review on modern deep learning based methods dealing with classification problems on huge sized images. Based on the similarity in the approach of typical works, we categorize the literature into three categories: naive patch based method, methods classifying huge images at full size and MIL based methods. We then propose a novel spatial and magnification based attention sampling strategy. First, we use a down-sampled large size image to estimate an attention map that represents a spatial probability distribution of informative patches at different magnifications. Then a small number of patches are cropped from the large size medical image at certain magnifications based on the obtained attention. The final label of the large size image is predicted solely by these patches using an end-to-end training strategy. Our experiments show that our approach on two different histopathology datasets, the publicly available BACH and a subset of the TCGA-PRAD dataset, demonstrate that the proposed method runs 2.5 times faster with automatic magnification selection in training and at least 1.6 times faster than using all patches in inference as the most of state-of-the-art methods do, without loosing in performance.
Computed Event Type: