# RobMedNAS: searching robust neural network architectures for medical image synthesis

## TL;DR

This paper introduces RobMedNAS, a method to find robust U-Net architectures for medical image synthesis that resist adversarial attacks.

## Contribution

RobMedNAS is a novel neural architecture search strategy for improving U-Net robustness in medical image synthesis.

## Key findings

- RobMedNAS-optimized models show enhanced resilience to adversarial attacks.
- The optimized models maintain high accuracy as measured by Dice coefficient and mean absolute error.
- Retrospective analysis confirms the effectiveness of RobMedNAS in medical image synthesis.

## 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.