Skin Deep Unlearning: Artefact and Instrument Debiasing in the Context of Melanoma Classification
This work addresses bias issues in automated melanoma classification to improve reliability for medical deployment, but it is incremental as it applies existing unlearning methods to known biases.
The paper tackled biases in melanoma classification from skin lesion images, such as surgical markings and imaging instruments, using two bias unlearning techniques, and showed that these biases were notably reduced with different techniques excelling at specific tasks.
Convolutional Neural Networks have demonstrated dermatologist-level performance in the classification of melanoma from skin lesion images, 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.