Objective quantification of chemotherapy-induced madarosis: a pilot study of an automated computer vision pipeline for eyebrow density assessment.

Chemotherapy-induced madarosis significantly impacts patient quality of life, yet current assessment methods rely heavily on subjective grading, limiting the evaluation of preventive strategies. This study presents a novel automated computer vision pipeline for objective longitudinal quantification of periocular hair density change and assesses its intra-session repeatability in a pilot clinical setting. The methodology integrates facial landmark detection, multi-temporal registration, and trimap deep learning-based hair segmentation with custom morphological filtering to isolate eyebrow hair structures from noise. The framework was applied to breast cancer patients undergoing anthracycline and taxane-based chemotherapy with localized cryotherapy. To evaluate the measurement repeatability, four standardized photographs (two with eyes open and two with eyes closed) were acquired and processed independently at each time point. Results confirmed the system's sensitivity in tracking individual evolution, accurately capturing the contrasting density changes observed between subjects from baseline to follow-up. Although the longitudinal trajectories are presented as proof of concept of temporal tracking capability, the method exhibited relatively high precision, with intra-session standard deviations consistently remaining below 12% regardless of hair density, with slightly higher values observed at baseline in some cases. We conclude that this automated computer vision pipeline provides a robust, operator-independent metric for monitoring madarosis. By overcoming the limitations of manual grading, this tool represents a promising step toward establishing a reliable primary endpoint for future multicentre clinical trials once validated in larger cohorts.