Weakly Supervised Object Detection in Chest X-Rays with Differentiable ROI Proposal Networks and Soft ROI Pooling

TL;DR

This paper introduces WSRPN, a novel weakly supervised object detection method for chest X-rays that generates bounding box proposals using a differentiable ROI-attention module, improving disease localization without requiring detailed annotations.

Contribution

The paper presents WSRPN, a new end-to-end trainable approach that effectively localizes diseases in chest X-rays using only image-level labels, addressing limitations of prior methods.

Findings

Outperforms existing WSup-OD methods in chest X-ray disease localization

End-to-end trainable with only image-label supervision

Effective bounding box proposals generated on-the-fly

Abstract

Weakly supervised object detection (WSup-OD) increases the usefulness and interpretability of image classification algorithms without requiring additional supervision. The successes of multiple instance learning in this task for natural images, however, do not translate well to medical images due to the very different characteristics of their objects (i.e. pathologies). In this work, we propose Weakly Supervised ROI Proposal Networks (WSRPN), a new method for generating bounding box proposals on the fly using a specialized region of interest-attention (ROI-attention) module. WSRPN integrates well with classic backbone-head classification algorithms and is end-to-end trainable with only image-label supervision. We experimentally demonstrate that our new method outperforms existing methods in the challenging task of disease localization in chest X-ray images. Code: https://github.com/philip-mueller/wsrpn