Concept level audits reveal limits in melanoma ai explanations
Testing concept-level audits of melanoma classifiers with diffusion counterfactuals and matched positive controls
Computer Vision and Pattern RecognitionArtificial IntelligenceMachine Learning
Summary
Melanoma classifiers use certain concepts to recognize skin cancer, but their explanations aren't often checked with known examples. The authors created a way to test these explanations using a special method that changes images in controlled ways. They found some explanations worked better than others and identified problems like one concept affecting others. Their approach uses matched controls and new tests to better evaluate these AI explanations.
What this means in practice
- •For medical ai developers: Evaluate the trustworthiness of melanoma detection AI by testing concept explanations against matched controls for known skin cancer features.
- •For medical imaging software teams: Improve melanoma classifier interpretability tools by incorporating diffusion-based counterfactual audits with colour calibration to reduce misleading explanations.
Authors
Md. Shohanoor Rahman, Md. Hasibul Islam, Md Kishor Morol, Moinul Hossain, Dip Nandi, Liew Tze Hui
Abstract
Concept-based explanations of melanoma classifiers are rarely tested against controls with known targets. We built a counterfactual audit from a diffusion model conditioned on seven 7-point checklist concepts and tested it with such controls. Colour-calibrated start noise reduced colour outliers from 78.5-84.8\% to 6.6-13.3\%. Four of seven knobs were target-dominant but also changed other concepts. Quartile-based positive controls were confounded, and the audit failed its pre-specified rule. It recovered the targets of all 16 controls matched on the generator's pseudo-labels and of 12 of 12 controls labelled by a second evaluator, for three concepts; concept activation vectors recovered fewer (11 of 16). In 18 permuted-label classifiers, significant sweep responses and target rankings also arose, whereas standardised effect margins separated the controls from these null classifiers. The pigmentation sweep raised the melanoma logit in 15 of 15 ISIC 2020 classifiers (+3.12 on average) and 6 of 6 HAM10000 classifiers (+4.01), about four times the null response in scale-free units; the regression effect did not transfer. An exploratory decomposition, evaluated against the controls, suggested that knob cross-talk partly hides vascular responses. Matched controls and null classifiers are practical checks of concept audits.