Why does my medical AI look at pictures of birds? Exploring the efficacy of transfer learning across domain boundaries

Abstract

It is an open secret that ImageNet is treated as the panacea of pretraining. Particularly in medical machine learning, models not trained from scratch are often finetuned based on ImageNet-pretrained models. We posit that pretraining on data from the domain of the downstream task should almost always be preferred instead. We leverage RadNet-12M, a dataset containing more than 12 million computed tomography (CT) image slices, to explore the efficacy of self-supervised pretraining on medical and natural images. Our experiments cover intra- and cross-domain transfer scenarios, varying data scales, finetuning vs. linear evaluation, and feature space analysis. We observe that intra-domain transfer compares favorably to cross-domain transfer, achieving comparable or improved performance (0.44% - 2.07% performance increase using RadNet pretraining, depending on the experiment) and demonstrate the existence of a domain boundary-related generalization gap and domain-specific learned features.

Code available from https://github.com/TIO-IKIM/Transfer-learning-across-domain-boundaries/ - Paper, code, and contents are subject to the CC-BY-NC 4.0 license

References in corpus (5)

• An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
• Efficiently Modeling Long Sequences with Structured State Spaces
• Medical Deep Learning -- A systematic Meta-Review
• Fully-automated Body Composition Analysis in Routine CT Imaging Using 3D Semantic Segmentation Convolutional Neural Networks
• QU-BraTS: MICCAI BraTS 2020 Challenge on Quantifying Uncertainty in Brain Tumor Segmentation - Analysis of Ranking Scores and Benchmarking Results