Yehuda's develop fairness-aware deep learning and computer vision methods for oncology imaging, with a primary focus on skin cancer detection. His current research investigates data-centric subset and coreset selection strategies to improve model accuracy while reducing performance gaps across skin tone subgroups. The goal of his work is to produce robust, clinically relevant evaluation and more equitable AI systems that can generalize across diverse patient populations.
In this line of work, Yehuda builds and evaluates deep learning classifiers that help distinguish benign from malignant skin lesions using clinical dermatology images. His work is directly related to skin cancer detection since it develops the training and testing workflow used to produce reliable AI models, including how data are selected for training, how models are tuned, and how performance is validated on held-out patients. Rather than reporting only overall accuracy, he examines whether the model performs equally well across different skin tones by comparing key clinical outcomes such as cancer detection sensitivity (how often cancer is correctly identified) and false-alarm rates (how often benign cases are flagged as cancer). As part of this research, Yehuda collaborates with dermatologists, oncologists, and radiologists to ensure the methods and evaluations align with clinical needs and real-world use. This approach supports AI-based skin cancer detection tools that are both accurate and more consistent across diverse patient populations.
Advisor: Prof. Vivek Singh