Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

On the Role of Calibration in Benchmarking Algorithmic Fairness for Skin Cancer Detection

Brandon Dominique, Prudence Lam, Nicholas Kurtansky, Jochen Weber, Kivanc Kose, Veronica Rotemberg, Jennifer Dy

TL;DR

This work addresses persistent disparities in melanoma detection by arguing that calibration should accompany AUROC-based fairness assessments. It benchmarks the ISIC 2020 Challenge winner ADAE against two rival methods and an ERM baseline on ISIC 2020 and PROVE-AI, using adaptive-score $CUSUM$ calibration and intersectional AUROC analyses across sex, Fitzpatrick Skin Tone, and age. The findings show that while discriminative accuracy improves with top models, calibration frequently degrades on unseen data, with age-driven miscalibration a recurrent issue. The paper highlights the necessity of comprehensive auditing and extensive metadata collection to achieve equitable AI-driven dermatology, providing public code to enable replication and further research.

Abstract

Artificial Intelligence (AI) models have demonstrated expert-level performance in melanoma detection, yet their clinical adoption is hindered by performance disparities across demographic subgroups such as gender, race, and age. Previous efforts to benchmark the performance of AI models have primarily focused on assessing model performance using group fairness metrics that rely on the Area Under the Receiver Operating Characteristic curve (AUROC), which does not provide insights into a model's ability to provide accurate estimates. In line with clinical assessments, this paper addresses this gap by incorporating calibration as a complementary benchmarking metric to AUROC-based fairness metrics. Calibration evaluates the alignment between predicted probabilities and observed event rates, offering deeper insights into subgroup biases. We assess the performance of the leading skin cancer detection algorithm of the ISIC 2020 Challenge on the ISIC 2020 Challenge dataset and the PROVE-AI dataset, and compare it with the second and third place models, focusing on subgroups defined by sex, race (Fitzpatrick Skin Tone), and age. Our findings reveal that while existing models enhance discriminative accuracy, they often over-diagnose risk and exhibit calibration issues when applied to new datasets. This study underscores the necessity for comprehensive model auditing strategies and extensive metadata collection to achieve equitable AI-driven healthcare solutions. All code is publicly available at https://github.com/bdominique/testing_strong_calibration.

On the Role of Calibration in Benchmarking Algorithmic Fairness for Skin Cancer Detection

TL;DR

This work addresses persistent disparities in melanoma detection by arguing that calibration should accompany AUROC-based fairness assessments. It benchmarks the ISIC 2020 Challenge winner ADAE against two rival methods and an ERM baseline on ISIC 2020 and PROVE-AI, using adaptive-score calibration and intersectional AUROC analyses across sex, Fitzpatrick Skin Tone, and age. The findings show that while discriminative accuracy improves with top models, calibration frequently degrades on unseen data, with age-driven miscalibration a recurrent issue. The paper highlights the necessity of comprehensive auditing and extensive metadata collection to achieve equitable AI-driven dermatology, providing public code to enable replication and further research.

Abstract

Artificial Intelligence (AI) models have demonstrated expert-level performance in melanoma detection, yet their clinical adoption is hindered by performance disparities across demographic subgroups such as gender, race, and age. Previous efforts to benchmark the performance of AI models have primarily focused on assessing model performance using group fairness metrics that rely on the Area Under the Receiver Operating Characteristic curve (AUROC), which does not provide insights into a model's ability to provide accurate estimates. In line with clinical assessments, this paper addresses this gap by incorporating calibration as a complementary benchmarking metric to AUROC-based fairness metrics. Calibration evaluates the alignment between predicted probabilities and observed event rates, offering deeper insights into subgroup biases. We assess the performance of the leading skin cancer detection algorithm of the ISIC 2020 Challenge on the ISIC 2020 Challenge dataset and the PROVE-AI dataset, and compare it with the second and third place models, focusing on subgroups defined by sex, race (Fitzpatrick Skin Tone), and age. Our findings reveal that while existing models enhance discriminative accuracy, they often over-diagnose risk and exhibit calibration issues when applied to new datasets. This study underscores the necessity for comprehensive model auditing strategies and extensive metadata collection to achieve equitable AI-driven healthcare solutions. All code is publicly available at https://github.com/bdominique/testing_strong_calibration.

Paper Structure

This paper contains 27 sections, 1 equation, 4 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Fairness in Medical Imaging
  4. Calibration
  5. Fairness Benchmarking
  6. Methodology
  7. Datasets
  8. Selected Methods
  9. All Data are Ext (ADAE) (1st Place, ha2020identifyingmelanomaimagesusing)
  10. 2nd Place, [2]
  11. 3rd Place, [3]
  12. Expected Risk Minimization (ERM) [4]
  13. Metrics and Evaluation
  14. Choice of Calibration Metric
  15. Statistical Analysis
  16. ...and 12 more sections

Figures (4)

  • Figure 1: A control chart and Variable Importance plot to check for subgroups who have their true risk overestimated in the ISIC 2020 dataset.
  • Figure 2: A Control Chart and Variable Importance plot to check for subgroups who have their true risk underestimated in the ISIC 2020 dataset.
  • Figure 3: A control chart and variable importance plot to check for subgroups who have their true risk overestimated in the PROVE-AI dataset.
  • Figure 4: A control chart and variable importance plot to check for subgroups who have their true risk underestimated in the PROVE-AI dataset.