LEVERAGING AUGMENTED REALITY AND MACHINE LEARNING TO SUGGEST DIAGNOSES FOR COMMON SKIN DISEASES IN BLACK AFRICANS

Abstract

Skin diseases are widespread health conditions that affect people of all skin tones but are often underdiagnosed or misdiagnosed in individuals with darker skin due to limited representation in dermatological research and diagnostic datasets. This thesis presents SkinVista, an mHealth application that integrates Augmented Reality (AR) and Machine Learning (ML) to enhance skin health awareness and support the suggestive diagnosis of common skin conditions among Black African populations. A distinctive strength of this research is the direct collaboration with Black dermatologists, whose clinical expertise informed every stage of SkinVista’s design, model development, and evaluation. This collaboration ensured that the system is contextually appropriate and specifically aligned with the diagnostic nuances of darker skin tones. SkinVista leverages AR-based camera guidance to help users capture high-quality skin images, while an ML model trained on dermatological images representing darker skin tones provides instant diagnostic suggestions for conditions such as acne, eczema, ringworm, and keloids. SkinVista was evaluated in two phases: (1) a pilot study involving six participants, and (2) a main study with seventy-one participants. The pilot study identified considerations and refinements to the study procedures and the application itself. Feedback showed high usability, with a mean SUS score of 85.0, and strong technology acceptance, with participants reporting high perceived usefulness (M = 4.33) and informativeness (M = 4.42) and also confirmed that the AR-guided capture workflow was intuitive and that it was feasible to conduct the study with a larger sample. Building on these refinements, the main study implemented the improved procedures with a larger sample, where participants interacted with the app, completed AR-guided image capture, received ML-based diagnostic suggestions, and evaluated the system using validated scales across usefulness, informativeness, usability, engagement, simplicity, and technology acceptance. Quantitative results showed high perceived usefulness and informativeness, with good usability ratings and positive behavioural intentions toward adoption. Qualitative feedback indicated that participants valued the app’s visual appeal, intuitive navigation, and educational content, while suggesting greater personalization and offline functionality. The findings demonstrate the potential of SkinVista to address diagnostic disparities through intelligent automation. This thesis contributes to Human-Computer Interaction (HCI) and digital health research by providing design insights for creating equitable, AI-driven healthcare tools that empower underrepresented populations and promote accessibility in dermatological care.