RobMedNAS: searching robust neural network architectures for medical image synthesis
Open Access
- 13 August 2024
- journal article
- research article
- Published by IOP Publishing in Biomedical Physics & Engineering Express
- Vol. 10 (5) , 055029
Abstract
Investigating U-Net model robustness in medical image synthesis against adversarial perturbations, this study introduces RobMedNAS, a neural architecture search strategy for identifying resilient U-Net configurations. Through retrospective analysis of synthesized CT from MRI data, employing Dice coefficient and mean absolute error metrics across critical anatomical areas, the study evaluates traditional U-Net models and RobMedNAS-optimized models under adversarial attacks. Findings demonstrate RobMedNAS’s efficacy in enhancing U-Net resilience without compromising on accuracy, proposing a novel pathway for robust medical image processing.Keywords
Funding Information
- U.S. National Library of Medicine (NIH R01 LM013151)
- Science and Technology Facilities Council
- Engineering and Physical Sciences Research Council (EP/T022159/1)
- Cambridge Service for Data Driven Discovery
- National Science Foundation (DMS-1749857)
This publication has 22 references indexed in Scilit:
- Generating synthetic computed tomography for radiotherapy: SynthRAD2023 challenge reportMedical Image Analysis, 2024
- Current and future developments of synthetic computed tomography generation for radiotherapyPhysics and Imaging in Radiation Oncology, 2023
- ResViT: Residual Vision Transformers for Multimodal Medical Image SynthesisIEEE Transactions on Medical Imaging, 2022
- Recent advances and clinical applications of deep learning in medical image analysisMedical Image Analysis, 2022
- Artificial Intelligence–Based Data Corrections for Attenuation and Scatter in Position Emission Tomography and Single-Photon Emission Computed TomographyPET Clinics, 2021
- Robustifying Deep Networks for Medical Image SegmentationJournal of Digital Imaging, 2021
- Synthetic Computed Tomography Generation from 0.35T Magnetic Resonance Images for Magnetic Resonance–Only Radiation Therapy Planning Using Perceptual Loss ModelsPractical Radiation Oncology, 2021
- Deep learning based synthetic‐CT generation in radiotherapy and PET: A reviewMedical Physics, 2021
- Partially-Connected Neural Architecture Search for Reduced Computational RedundancyIEEE Transactions on Pattern Analysis and Machine Intelligence, 2021
- A Taxonomy of Global Optimization Methods Based on Response SurfacesJournal of Global Optimization, 2001