Androgenetic alopecia is a progressive loss of hair, now affecting almost 20% of the world's population. Clinical monitoring relies heavily on the analysis of trichoscopic images to assess changes in hair density and structure over time. Manual annotation, while accurate, is often time-consuming and difficult to apply consistently to multiple clinical points or high volumes.
In this context, a collaboration emerged between a digital clinic and LIS around the design and implementation of a workflow and software solution to streamline these tasks, enabling the accurate extraction of capillary measurements to help clinicians diagnose and monitor treatment over time.
After examining a range of generic object segmentation models, the experts opted for Meta's Segment Anything Model (SAM) to speed up the image annotation process, and progressively refined several versions of the SAM using a Human-in-the-loop (HITL) strategy.
To this end, LIS then developed a suite of AI tools, analytics and utilities, integrated into a dedicated Napari plugin, to help human annotators improve both annotation accuracy and speed. This semi-automated workflow efficiently extracts individual and collective hair measurements (hair number, width and density, follicle number and density) while retaining expert supervision. The combined use of AI and image analysis tools significantly reduces annotation time without compromising quality, enabling more scalable and reproducible monitoring of hair loss progression.