Analise de Desaprendizado de Maquina em Modelos de Classificacao de Imagens Medicas

TL;DR

This paper evaluates the SalUn machine unlearning model for medical image classification, demonstrating its effectiveness in removing sensitive data while maintaining model performance, with implications for privacy in medical AI.

Contribution

First application of machine unlearning in medical image classification, analyzing data augmentation effects and showing SalUn's near-retraining performance.

Findings

SalUn achieves performance close to full retraining

Data augmentation impacts unlearning quality

Effective privacy-preserving model updates in medical imaging

Abstract

Machine unlearning aims to remove private or sensitive data from a pre-trained model while preserving the model's robustness. Despite recent advances, this technique has not been explored in medical image classification. This work evaluates the SalUn unlearning model by conducting experiments on the PathMNIST, OrganAMNIST, and BloodMNIST datasets. We also analyse the impact of data augmentation on the quality of unlearning. Results show that SalUn achieves performance close to full retraining, indicating an efficient solution for use in medical applications.