AI-Driven Mobile Application for Early Detection of Skin Cancer Lesions

Thesis Abstract

Skin cancer represents one of the most prevalent and yet potentially treatable forms of malignancy if detected early, but many cases remain undiagnosed until advanced stages due to limited access to specialized clinical assessments, especially in low-resource settings. This study aims to develop and evaluate an AI-driven mobile application designed to facilitate the early detection of skin cancer lesions, thereby improving diagnostic accuracy and promoting timely medical intervention. The primary objectives include (1) designing a machine learning model capable of classifying skin lesion images into benign or malignant categories; (2) developing a user-friendly mobile interface for image capture and real-time analysis; and (3) assessing the application's diagnostic performance and user acceptability in a real-world context. The research adopts a mixed-methods approach, integrating quantitative and qualitative procedures. The quantitative component involves a cross-sectional diagnostic accuracy study employing a sample of 1,200 skin lesion images collected from patients at dermatology clinics in metropolitan hospitals. These images, annotated by experienced dermatologists, form the dataset used to train and validate the convolutional neural network (CNN) model. The model's performance is evaluated through metrics such as sensitivity, specificity, positive predictive value, negative predictive value, and area under the receiver operating characteristic (ROC) curve, employing techniques like logistic regression and ROC analysis to fine-tune thresholds. The qualitative component involves semi-structured interviews with 30 users—comprising patients and healthcare practitioners—to assess usability, perceived accuracy, and acceptance of the mobile application, with data analyzed through thematic analysis. Expected findings include a CNN model achieving sensitivity and specificity rates exceeding 85%, with the area under the ROC curve surpassing 0.90, indicating high diagnostic accuracy comparable to specialist evaluations. The mobile application is anticipated to demonstrate substantial usability scores and positive user perceptions, suggesting feasibility for integration into primary healthcare settings. The study will also identify common usability issues and barriers to adoption. This research contributes significantly to existing knowledge by integrating advanced artificial intelligence techniques with mobile health technologies tailored for dermatological diagnostics, filling gaps related to real-world application and user acceptance. The findings offer empirical evidence supporting the validity and reliability of AI-based mobile tools for skin cancer screening, especially in contexts with limited dermatologist availability. The study also extends the theoretical framework of technology acceptance models, emphasizing factors influencing health technology adoption among diverse user groups. The main conclusions affirm that AI-driven mobile applications can serve as effective adjuncts in early skin cancer detection, potentially decreasing diagnostic delays and improving patient outcomes. It underscores the importance of robust training datasets, user-centered design, and careful ethical considerations in deploying AI health tools. The study recommends scaling the application across different demographic groups and integrating AI-supported screening into routine teledermatology services, with suggestions for future research exploring longitudinal impacts, cost-effectiveness, and integration with electronic health records to enhance comprehensive skin cancer management strategies.