Physical foundations for trustworthy medical imaging: a review for artificial intelligence researchers

TL;DR

This survey addresses the problem of trustworthy AI in medical imaging by detailing the physical principles underpinning each modality and outlining how physics-informed machine learning can improve robustness, interpretability, and generalization. It surveys modality-specific physics—from X-ray and MRI to ultrasound and nuclear imaging—while highlighting hardware- and software-based multimodal fusion, and the role of artifacts and technical limitations in shaping AI performance. The authors discuss observational, learning, and inductive biases used to embed physics into learning algorithms (e.g., PINNs, SA-PINNs, physics-based losses and architectures) and address challenges related to data scarcity, domain shifts, and clinical translation. They conclude that a synergistic collaboration among clinicians, physicists, and AI developers is essential to translate AI-powered medical imaging into trustworthy clinical practice, with future directions including foundation models, multimodal integration, and physics-informed generative reconstruction.

Abstract

Artificial intelligence in medical imaging has seen unprecedented growth in the last years, due to rapid advances in deep learning and computing resources. Applications cover the full range of existing medical imaging modalities, with unique characteristics driven by the physics of each technique. Yet, artificial intelligence professionals entering the field, and even experienced developers, often lack a comprehensive understanding of the physical principles underlying medical image acquisition, which hinders their ability to fully leverage its potential. The integration of physics knowledge into artificial intelligence algorithms enhances their trustworthiness and robustness in medical imaging, especially in scenarios with limited data availability. In this work, we review the fundamentals of physics in medical images and their impact on the latest advances in artificial intelligence, particularly, in generative models and reconstruction algorithms. Finally, we explore the integration of physics knowledge into physics-inspired machine learning models, which leverage physics-based constraints to enhance the learning of medical imaging features.

Figures (2)

• Figure 1: Medical imaging modalities in the electromagnetic radiation spectrum.
• Figure 2: Ultrasound imaging in the acoustic spectrum.