End-to-end Training for Whole Image Breast Cancer Diagnosis using An All Convolutional Design

TL;DR

This paper introduces an end-to-end convolutional approach for whole-image breast cancer diagnosis from mammograms, reducing annotation reliance and achieving high accuracy on multiple datasets.

Contribution

It presents a novel all convolutional training method that requires only image-level labels after initial lesion annotation, simplifying the training process and improving performance.

Findings

Achieved 0.88 AUC on DDSM with a single model

Improved to 0.91 AUC with three-model averaging on DDSM

Reached 0.96 AUC on INbreast dataset

Abstract

We develop an end-to-end training algorithm for whole-image breast cancer diagnosis based on mammograms. It requires lesion annotations only at the first stage of training. After that, a whole image classifier can be trained using only image level labels. This greatly reduced the reliance on lesion annotations. Our approach is implemented using an all convolutional design that is simple yet provides superior performance in comparison with the previous methods. On DDSM, our best single-model achieves a per-image AUC score of 0.88 and three-model averaging increases the score to 0.91. On INbreast, our best single-model achieves a per-image AUC score of 0.96. Using DDSM as benchmark, our models compare favorably with the current state-of-the-art. We also demonstrate that a whole image model trained on DDSM can be easily transferred to INbreast without using its lesion annotations and using only a small amount of training data. Code availability: https://github.com/lishen/end2end-all-conv