The Statistical Methods for Health Equity Series is a monthly online series co-hosted by the Data Science for Health Equity community, the Alan Turing Institute Health Equity Interest Group, and the Department of Statistical Science at University College London.

---

We are excited to announce the next instalment of our webinar series with speaker Brandon Dominique - (Northeastern University)

Title: Post-Hoc Methods for Detecting and Correcting Subgroup Bias in Medical AI Systems

Abstract: This talk examines post-hoc calibration as a unifying lens for understanding and addressing subgroup-level biases in AI models for skin cancer detection. We present two complementary perspectives on this problem. First, we implement a score-based CUSUM testing framework for auditing strong calibration in deployed models, demonstrating that state-of-the-art melanoma detection algorithms, despite strong average discriminative performance, exhibit systematic miscalibration across demographic subgroups defined by age, sex, and Fitzpatrick skin tone, with age emerging as a consistent driver of both over- and under-estimation of risk. Second, we implement a post-hoc clustering method aimed at not only discovering subgroups but correcting their calibration, leveraging Vision-Language Model embeddings to learn interpretable, semantically meaningful groups and applying group-specific corrections. Across multiple VLMs and dermatology datasets, this method achieves up to 79% reduction in Expected Calibration Error over competitive baselines while discovering clinically coherent groupings that align with known epidemiological risk patterns. Together, these works establish post-hoc calibration as both a diagnostic and corrective framework for building more equitable and trustworthy medical AI systems.

Speaker Bio: Brandon Dominique is a PhD candidate in Computer Engineering at Northeastern University, where he works with Dr. Jennifer Dy in the Signal Processing, Imaging, Reasoning, and Learning (SPIRAL) group. His research sits at the intersection of fairness, transparency, and auditing, with a focus on skin cancer detection and the development of tools that evaluate and improve how AI models perform across diverse demographic subgroups. He is a recipient of the GEM Fellowship as well as the LSAMP STARS Fellowship. Prior to Northeastern, he completed his B.S. in Computer Engineering at the New Jersey Institute of Technology.

--

Please direct any questions about the webinar series to info@datascienceforhealthequity.com.