Interpreting Biomedical VLMs on High-Imbalance Out-of-Distributions: An Insight into BiomedCLIP on Radiology

TL;DR

This paper analyzes the capabilities and limitations of BiomedCLIP, a large vision-language model, in classifying imbalanced radiology datasets, revealing its strengths and weaknesses in different adaptation scenarios.

Contribution

It provides a detailed analysis of BiomedCLIP's embedding space and evaluates its performance on a highly imbalanced medical dataset using various adaptation methods.

Findings

Zero-shot over-predicts labels, reducing precision

Full fine-tuning improves disease classification

Linear probing detects overlapping features

Abstract

In this paper, we construct two research objectives: i) explore the learned embedding space of BiomedCLIP, an open-source large vision language model, to analyse meaningful class separations, and ii) quantify the limitations of BiomedCLIP when applied to a highly imbalanced, out-of-distribution multi-label medical dataset. We experiment on IU-xray dataset, which exhibits the aforementioned criteria, and evaluate BiomedCLIP in classifying images (radiographs) in three contexts: zero-shot inference, full finetuning, and linear probing. The results show that the model under zero-shot settings over-predicts all labels, leading to poor precision and inter-class separability. Full fine-tuning improves classification of distinct diseases, while linear probing detects overlapping features. We demonstrate visual understanding of the model using Grad-CAM heatmaps and compare with 15 annotations by a radiologist. We highlight the need for careful adaptations of the models to foster reliability and applicability in a real-world setting. The code for the experiments in this work is available and maintained on GitHub.