Local Residual Attention Network for Classification of Breast Cancer Histopathology Images
Keywords: Local Residual Attention, Breast Histopathology Classification
Abstract: Automatic classification of breast histopathological images is a challenging task, as subtle changes in morphometric features can result in misclassifications. To increase broader adoption and trust in deep-learning based solutions, we need methods that produce the right results for the right reasons, while still maintaining high performance. To make progress toward these goals, we propose a novel Local Residual Attention Network (LRAN), that improves the predictive performance of a base-network by attending to class relevant regions of an image. LRAN follows an encoder-decoder architecture with local attention enforced on the skip connections between the encoder and decoder and global attention on the feature maps of the base-network. Our experiments demonstrate that the inclusion of attention mechanisms increases the classification accuracy by 5-8\% points over the base-network. Our LRAN with ReseNet-18 as the base-network produces a classification accuracy of 91.83\% on the ICIAR 2018 BreAst Cancer Histology (BACH 2018) dataset, which is comparable to the performance on this dataset by a top-performing classification networks.
Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.
Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.
Paper Type: validation/application paper
Primary Subject Area: Interpretability and Explainable AI
Secondary Subject Area: Detection and Diagnosis
Confidentiality And Author Instructions: I read the call for papers and author instructions. I acknowledge that exceeding the page limit and/or altering the latex template can result in desk rejection.
0 Replies
Loading