Overcoming data scarcity in biomedical imaging with a foundational multi-task model

TL;DR

This paper introduces UMedPT, a foundational model for biomedical imaging that achieves high performance with limited training data across various tasks and datasets.

Contribution

The novel multi-task learning strategy decouples training tasks from memory requirements, enabling effective training with less data.

Findings

UMedPT outperformed ImageNet pretraining and state-of-the-art models in biomedical imaging tasks.

It maintained performance with only 1% of training data for in-domain tasks and 50% for out-of-domain tasks.

UMedPT demonstrated superior cross-center transferability in an external validation.

Abstract

Foundational models, pretrained on a large scale, have demonstrated substantial success across non-medical domains. However, training these models typically requires large, comprehensive datasets, which contrasts with the smaller and more specialized datasets common in biomedical imaging. Here we propose a multi-task learning strategy that decouples the number of training tasks from memory requirements. We trained a universal biomedical pretrained model (UMedPT) on a multi-task database including tomographic, microscopic and X-ray images, with various labeling strategies such as classification, segmentation and object detection. The UMedPT foundational model outperformed ImageNet pretraining and previous state-of-the-art models. For classification tasks related to the pretraining database, it maintained its performance with only 1% of the original training data and without fine-tuning. For out-of-domain tasks it required only 50% of the original training data. In an external independent validation, imaging features extracted using UMedPT proved to set a new standard for cross-center transferability.

UMedPT, a foundational model for biomedical imaging, has been trained on a variety of medical tasks with different types of label. It has achieved high performance with less training data in various clinical applications.

Figures