Google-MedGemma Based Abnormality Detection in Musculoskeletal radiographs

TL;DR

This paper introduces a MedGemma-based framework for automatic abnormality detection in musculoskeletal radiographs, leveraging a medical foundation model to improve classification accuracy and generalization over traditional methods.

Contribution

The study presents a novel MedGemma-driven approach utilizing a pretrained vision encoder and transfer learning for enhanced abnormality detection in radiographs, surpassing conventional models.

Findings

MedGemma-based classifier outperforms traditional convolutional and autoencoder models.

Transfer learning with MedGemma improves generalization in medical image classification.

Modular training strategies enable efficient domain adaptation.

Abstract

This paper proposes a MedGemma-based framework for automatic abnormality detection in musculoskeletal radiographs. Departing from conventional autoencoder and neural network pipelines, the proposed method leverages the MedGemma foundation model, incorporating a SigLIP-derived vision encoder pretrained on diverse medical imaging modalities. Preprocessed X-ray images are encoded into high-dimensional embeddings using the MedGemma vision backbone, which are subsequently passed through a lightweight multilayer perceptron for binary classification. Experimental assessment reveals that the MedGemma-driven classifier exhibits strong performance, exceeding conventional convolutional and autoencoder-based metrics. Additionally, the model leverages MedGemma's transfer learning capabilities, enhancing generalization and optimizing feature engineering. The integration of a modern medical foundation model not only enhances representation learning but also facilitates modular training strategies such as selective encoder block unfreezing for efficient domain adaptation. The findings suggest that MedGemma-powered classification systems can advance clinical radiograph triage by providing scalable and accurate abnormality detection, with potential for broader applications in automated medical image analysis. Keywords: Google MedGemma, MURA, Medical Image, Classification.