Utilizing Segment Anything Model For Assessing Localization of GRAD-CAM in Medical Imaging

TL;DR

This paper explores using the Segment Anything Model (SAM) to evaluate the localization accuracy of saliency maps in medical imaging, aiming to improve assessment methods beyond human annotations and address domain-specific challenges.

Contribution

It introduces SAM-based metrics for saliency map evaluation in medical imaging, enhancing accuracy and generalization beyond traditional human-annotated benchmarks.

Findings

SAM shows high similarity to existing metrics in localization assessment.

The methodology can operate without human annotations, enabling broader application.

Challenges include image pre-processing and domain adaptation for SAM in medical images.

Abstract

The introduction of saliency map algorithms as an approach for assessing the interoperability of images has allowed for a deeper understanding of current black-box models with Artificial Intelligence. Their rise in popularity has led to these algorithms being applied in multiple fields, including medical imaging. With a classification task as important as those in the medical domain, a need for rigorous testing of their capabilities arises. Current works examine capabilities through assessing the localization of saliency maps upon medical abnormalities within an image, through comparisons with human annotations. We propose utilizing Segment Anything Model (SAM) to both further the accuracy of such existing metrics, while also generalizing beyond the need for human annotations. Our results show both high degrees of similarity to existing metrics while also highlighting the capabilities of this methodology to beyond human-annotation. Furthermore, we explore the applications (and challenges) of SAM within the medical domain, including image pre-processing before segmenting, natural language proposals to SAM in the form of CLIP-SAM, and SAM accuracy across multiple medical imaging datasets.