Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PINNs for Medical Image Analysis: A Survey

Chayan Banerjee, Kien Nguyen, Olivier Salvado, Truyen Tran, Clinton Fookes

TL;DR

This survey advances the field of physics-informed medical image analysis (PIMIA) by providing a unified taxonomy of physics priors, representation strategies, and integration methods across imaging, generation, inverse imaging, registration, segmentation, and predictive modeling. It synthesizes insights from over 80 papers (2018–2024), introduces an Improvement Score to benchmark PI gains, and highlights that inverse imaging tasks such as super-resolution and CT/MRI reconstruction often reap the largest benefits from physics priors. The authors discuss current challenges—most notably selecting suitable priors and establishing standardized benchmarks—and outline future directions that include uncertainty quantification and transformer-based physics-integrated architectures. Overall, the work clarifies when and how physics information can yield more robust, data-efficient, and physically plausible medical image analyses, with practical implications for improved diagnostics and treatment planning.

Abstract

The incorporation of physical information in machine learning frameworks is transforming medical image analysis (MIA). By integrating fundamental knowledge and governing physical laws, these models achieve enhanced robustness and interpretability. In this work, we explore the utility of physics-informed approaches for MIA (PIMIA) tasks such as registration, generation, classification, and reconstruction. We present a systematic literature review of over 80 papers on physics-informed methods dedicated to MIA. We propose a unified taxonomy to investigate what physics knowledge and processes are modelled, how they are represented, and the strategies to incorporate them into MIA models. We delve deep into a wide range of image analysis tasks, from imaging, generation, prediction, inverse imaging (super-resolution and reconstruction), registration, and image analysis (segmentation and classification). For each task, we thoroughly examine and present in a tabular format the central physics-guided operation, the region of interest (with respect to human anatomy), the corresponding imaging modality, the dataset used for model training, the deep network architecture employed, and the primary physical process, equation, or principle utilized. Additionally, we also introduce a novel metric to compare the performance of PIMIA methods across different tasks and datasets. Based on this review, we summarize and distil our perspectives on the challenges, open research questions, and directions for future research. We highlight key open challenges in PIMIA, including selecting suitable physics priors and establishing a standardized benchmarking platform.

PINNs for Medical Image Analysis: A Survey

TL;DR

This survey advances the field of physics-informed medical image analysis (PIMIA) by providing a unified taxonomy of physics priors, representation strategies, and integration methods across imaging, generation, inverse imaging, registration, segmentation, and predictive modeling. It synthesizes insights from over 80 papers (2018–2024), introduces an Improvement Score to benchmark PI gains, and highlights that inverse imaging tasks such as super-resolution and CT/MRI reconstruction often reap the largest benefits from physics priors. The authors discuss current challenges—most notably selecting suitable priors and establishing standardized benchmarks—and outline future directions that include uncertainty quantification and transformer-based physics-integrated architectures. Overall, the work clarifies when and how physics information can yield more robust, data-efficient, and physically plausible medical image analyses, with practical implications for improved diagnostics and treatment planning.

Abstract

The incorporation of physical information in machine learning frameworks is transforming medical image analysis (MIA). By integrating fundamental knowledge and governing physical laws, these models achieve enhanced robustness and interpretability. In this work, we explore the utility of physics-informed approaches for MIA (PIMIA) tasks such as registration, generation, classification, and reconstruction. We present a systematic literature review of over 80 papers on physics-informed methods dedicated to MIA. We propose a unified taxonomy to investigate what physics knowledge and processes are modelled, how they are represented, and the strategies to incorporate them into MIA models. We delve deep into a wide range of image analysis tasks, from imaging, generation, prediction, inverse imaging (super-resolution and reconstruction), registration, and image analysis (segmentation and classification). For each task, we thoroughly examine and present in a tabular format the central physics-guided operation, the region of interest (with respect to human anatomy), the corresponding imaging modality, the dataset used for model training, the deep network architecture employed, and the primary physical process, equation, or principle utilized. Additionally, we also introduce a novel metric to compare the performance of PIMIA methods across different tasks and datasets. Based on this review, we summarize and distil our perspectives on the challenges, open research questions, and directions for future research. We highlight key open challenges in PIMIA, including selecting suitable physics priors and establishing a standardized benchmarking platform.
Paper Structure (30 sections, 6 equations, 14 figures, 3 tables)

This paper contains 30 sections, 6 equations, 14 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Physics-informed Machine Learning (PIML)
  3. Biases as taxonomy in PIML
  4. Physics-Informed Medical Image Analysis (PIMIA)
  5. Physics incorporation in general ML and MIA:
  6. Intuitive introduction to physics priors in MIA:
  7. Physics prior categories with examples (Taxonomy I):
  8. Medical Imaging Physics:
  9. Process dynamics and Equations:
  10. Others:
  11. Medical Image Analysis (MIA) tasks (Taxonomy II)
  12. Typical pipeline
  13. PIMIA Tasks: Detailed Survey
  14. Imaging
  15. Image Generation
  16. ...and 15 more sections

Figures (14)

  • Figure 1: Physics-informed machine models leverage physics information to enhance efficiency and plausibility, bridging data-driven and numerical models, vital for accurate representations across medical domains.
  • Figure 2: Plots showing the latest research trend on the topic of PIMIA, (a) Published work over the years and (b) Research share of PIMIA tasks.
  • Figure 3: The PINN algorithm from karniadakis2021physics
  • Figure 4: Example of PINN application from wang2022dense
  • Figure 5: The three-stage process begins with the collection of random points in k-space, which represents spatial frequency information. These points are gathered without hardware constraints. Subsequently, a TSP (Travelling Salesman Problem) solver is employed to arrange these points into a path that minimizes distance. This optimized trajectory undergoes refinement during training to accommodate machine/ hardware constraints. This enhances MRI performance by improving data collection efficiency and image quality while ensuring operational feasibility.
  • ...and 9 more figures