Position Guided Leveraging Vision Transformer for Precise Skin Diagnoses Via Smartphone Imagery.

Skin diagnostics have historically relied on specialized equipment, making assessments both costly and less accessible. In this paper, we introduce a novel methodology that utilizes smartphone imagery in conjunction with deep learning techniques to diagnose various precise skin conditions including pores, rubeosis, wrinkles, and texture. Central to this approach is the implementation of a face-landmark detection algorithm, ensuring precise identification of specific facial regions. Image patches extracted from these identified regions are then processed using our proposed vision transformer enhanced with robust positional guidance. This leads to a final classifier responsible for the diagnostic results. To validate the performance of our proposed method, we utilized smartphone imagery and skin diagnosis scores derived from precision skin measurement equipment from 848 participants. The results affirm the reliability of our method in assessing skin condition severity, suggesting a promising shift towards more accessible skin diagnostics.