Unsupervised SAM segmentation of zebrafish body: Application to melanin analysis

Zebrafish have always been a valuable model for studies on human health. Their transparency makes it highly suitable for observing melanin synthesis. Moreover, their high genetic similarity to humans facilitates the study of human diseases, including pigmentation disorders. However, accurate quantification of melanin in zebrafish is essential for determining the efficacy of inhibitors. This study investigated melanin content in zebrafish exposed to varying concentrations of 1-phenyl 2-thiourea (PTU) and α-Arbutin. To improve the accuracy and reduce the manual effort associated with melanin quantification, we employed the emerging Segment Anything Model (SAM) for unsupervised image segmentation of zebrafish. The model demonstrated high accuracy, achieving 100 % in generating body masks that delineate the zebrafish body and 96.7 % in automatically selecting appropriate masks. Melanin content was then calculated by pixel-level integration within these masks. Overall, melanin synthesis is inhibited with the increase in inhibitor concentration. This approach enables more precise quantification while minimizing manual effort. The proposed method produces quantitative results comparable to existing methods, while offering a simpler and more precise approach.