Adversarially Robust Medical Classification via Attentive Convolutional Neural Networks

TL;DR

This paper introduces an attention-based CNN approach for medical image classification that significantly improves robustness against adversarial attacks, enhancing reliability in automated diagnosis systems.

Contribution

It demonstrates that integrating attention mechanisms into CNNs enhances adversarial robustness without sacrificing accuracy, a novel approach in medical imaging.

Findings

Robust accuracy increased by up to 16% in typical scenarios.

Robust accuracy increased by up to 2700% in extreme cases.

Attention mechanisms effectively improve model stability against adversarial attacks.

Abstract

Convolutional neural network-based medical image classifiers have been shown to be especially susceptible to adversarial examples. Such instabilities are likely to be unacceptable in the future of automated diagnoses. Though statistical adversarial example detection methods have proven to be effective defense mechanisms, additional research is necessary that investigates the fundamental vulnerabilities of deep-learning-based systems and how best to build models that jointly maximize traditional and robust accuracy. This paper presents the inclusion of attention mechanisms in CNN-based medical image classifiers as a reliable and effective strategy for increasing robust accuracy without sacrifice. This method is able to increase robust accuracy by up to 16% in typical adversarial scenarios and up to 2700% in extreme cases.