CheXtransfer: Performance and Parameter Efficiency of ImageNet Models for Chest X-Ray Interpretation

TL;DR

This study evaluates the transferability and efficiency of 16 ImageNet architectures for chest X-ray interpretation, revealing that pretraining improves performance especially for smaller models, and that models can be made more parameter-efficient without significant performance loss.

Contribution

It provides the first comprehensive comparison of ImageNet models for chest X-ray tasks, showing that pretraining benefits smaller models and that models can be truncated for efficiency without major performance drops.

Findings

No correlation between ImageNet and CheXpert performance.

Pretraining significantly boosts performance, especially for smaller models.

Models can be truncated to improve efficiency with minimal performance loss.

Abstract

Deep learning methods for chest X-ray interpretation typically rely on pretrained models developed for ImageNet. This paradigm assumes that better ImageNet architectures perform better on chest X-ray tasks and that ImageNet-pretrained weights provide a performance boost over random initialization. In this work, we compare the transfer performance and parameter efficiency of 16 popular convolutional architectures on a large chest X-ray dataset (CheXpert) to investigate these assumptions. First, we find no relationship between ImageNet performance and CheXpert performance for both models without pretraining and models with pretraining. Second, we find that, for models without pretraining, the choice of model family influences performance more than size within a family for medical imaging tasks. Third, we observe that ImageNet pretraining yields a statistically significant boost in performance across architectures, with a higher boost for smaller architectures. Fourth, we examine whether ImageNet architectures are unnecessarily large for CheXpert by truncating final blocks from pretrained models, and find that we can make models 3.25x more parameter-efficient on average without a statistically significant drop in performance. Our work contributes new experimental evidence about the relation of ImageNet to chest x-ray interpretation performance.