Skip to main content

Research Repository

Advanced Search

Skin Deep Unlearning: Artefact and Instrument Debiasing in the Context of Melanoma Classification

Bevan, Peter; Atapour-Abarghouei, Amir

Skin Deep Unlearning: Artefact and Instrument Debiasing in the Context of Melanoma Classification Thumbnail


Authors

Peter Bevan



Contributors

Kamalika Chaudhuri
Editor

Stefanie Jegelka
Editor

Le Song
Editor

Csaba Szepesvari
Editor

Gang Niu
Editor

Sivan Sabato
Editor

Abstract

Convolutional Neural Networks have demonstrated dermatologist-level performance in the classification of melanoma and other skin lesions, but prediction irregularities due to biases seen within the training data are an issue that should be addressed before widespread deployment is possible. In this work, we robustly remove bias and spurious variation from an automated melanoma classification pipeline using two leading bias ‘unlearning’ techniques. We show that the biases introduced by surgical markings and rulers presented in previous studies can be reasonably mitigated using these bias removal methods. We also demonstrate the generalisation benefits of ‘unlearning’ spurious variation relating to the imaging instrument used to capture lesion images. Our experimental results provide evidence that the effects of each of the aforementioned biases are notably reduced, with different debiasing techniques excelling at different tasks.

Citation

Bevan, P., & Atapour-Abarghouei, A. (2022). Skin Deep Unlearning: Artefact and Instrument Debiasing in the Context of Melanoma Classification. In K. Chaudhuri, S. Jegelka, L. Song, C. Szepesvari, G. Niu, & S. Sabato (Eds.), Proceedings of Machine Learning Research (1874-1892)

Conference Name The 39th International Conference on Machine Learning (ICML 2022)
Conference Location Baltimore, MD
Start Date Jul 17, 2022
End Date Jul 23, 2022
Acceptance Date May 15, 2022
Online Publication Date Dec 6, 2022
Publication Date 2022
Deposit Date Jun 10, 2022
Publicly Available Date Mar 29, 2024
Volume 162
Pages 1874-1892
Series Title Proceedings of Machine Learning Research (PMLR)
Series ISSN 2640-3498
Book Title Proceedings of Machine Learning Research
Public URL https://durham-repository.worktribe.com/output/1136127
Publisher URL https://proceedings.mlr.press/v162/bevan22a.html

Files





You might also like



Downloadable Citations