Harnessing Artificial Intelligence for Diagnosing Androgenetic Alopecia: A Training and Validation Study

The investigators intend to develop and validate artificial intelligence (AI) and machine learning (ML) models in diagnosis of male and female pattern hair loss, and assessment of its severity based on clinical and trichoscopic image using widely available and accessible handheld dermoscopes.

Conventional androgenetic alopecia (AGA) diagnosis and severity assessment are tedious and time-consuming tasks that are prone to human errors. These challenges can be tackled using artificial intelligence (AI), namely leveraging applications of machine learning and artificial neural networks for enhancing the diagnostic accuracy of scalp disease classification systems via dermoscopic image analysis. Computer aided assessment of hair microphotographs was attempted for decades, yet it faced many technical hurdles before the onset of deep learning and neural networks; and currently available software generate inaccurate results compared with visual counting. More accurate methods of analysis are needed for trichoscopic imaging, utilising deep learning image recognition models trained with a large image dataset. A number of deep learning models have been developed in recent years using videodermoscopy that achieved reliable hair density, thickness and severity classification, yet remain limited by small non-inclusive training datasets, need for hair shaving and lack of detailed reporting. Moreover, to our knowledge all previous models depend on image acquisition from expensive standalone videodermoscopy devices that lack widespread availability, rather than handheld dermoscopes that are commonly available.

The study will enroll 400 participants (200 healthy controls and 200 AGA patients). Controls undergo history and trichoscopic exams to exclude hair disorders. Trichoscopic examination will be conducted using a handheld dermoscope (CuTechs DS175) with a specialized field spacer. Patients will be assessed for disease severity using gender-specific scales. Both groups will have standardized digital and trichoscopic images taken for analysis. Images will be used to manually count and classify hairs, assess follicle units, and identify dermoscopic signs. A structured database will store all data and link clinical and image data to support objective diagnosis. AI models, particularly CNNs using transfer learning, will be trained on preprocessed images for classification and severity scoring. Model performance will be evaluated using metrics like accuracy, precision, recall, F1-score, and AUC-ROC compared with metrics reported by expert trichologists to validate accuracy and reliability