AI-Powered Mobile App for Early Skin Cancer Detection and Monitoring

Thesis Abstract

Skin cancer remains one of the most prevalent and potentially lethal forms of malignancy worldwide, yet early detection and continuous monitoring remain significant challenges due to limited access to specialized dermatological services and reliance on manual visual assessments. This study aims to develop and evaluate an AI-powered mobile application designed to facilitate early skin cancer detection and ongoing monitoring of suspicious skin lesions. The specific objectives include (1) designing a robust machine learning model capable of accurately classifying benign versus malignant skin lesions; (2) integrating the model into a user-friendly mobile platform; (3) assessing the app’s diagnostic performance in a real-world clinical setting; and (4) examining user acceptability and compliance among at-risk populations. Employing a mixed-method research design, the study combines quantitative algorithm development and validation with qualitative usability evaluation. The quantitative component involves collecting a dataset of 5,000 dermoscopic and clinical images of skin lesions—sourced from dermatology clinics and public repositories—split into training, validation, and testing subsets. The core analytical technique is convolutional neural network (CNN) modeling, with transfer learning applied to enhance accuracy, followed by statistical evaluation through receiver operating characteristic (ROC) curve analysis, sensitivity, specificity, and F1 scores to assess the model’s diagnostic performance. For the qualitative phase, 30 participants from high-risk groups (including diverse age, gender, and skin type) are interviewed using thematic analysis to gauge usability, perceived accuracy, and behavioral intentions regarding app adoption. Expected findings include the development of a CNN-based classifier achieving at least 90% accuracy, 95% sensitivity, and 85% specificity in distinguishing malignant from benign lesions, corroborated by strong ROC-AUC values exceeding 0.92. The app’s usability assessment is anticipated to demonstrate high acceptability, with participants expressing increased confidence in early detection and willingness to incorporate regular self-monitoring into health routines. Additionally, the study hypothesizes that user engagement correlates positively with perceived benefits, as explained by the Technology Acceptance Model (TAM) framework. This research contributes novel insights into the application of artificial intelligence in mobile health interventions for skin cancer, offering a scalable, accessible, and cost-effective tool for early diagnosis. It extends existing literature by integrating deep learning with user-centered design and real-world validation, filling notable gaps in deploying AI-driven diagnostics in resource-constrained settings. The work advances knowledge on the operationalization of AI models within mobile platforms, emphasizing clinical accuracy, user engagement, and health behavior change. The study concludes that the AI-powered mobile app holds significant potential to improve early detection rates and facilitate remote monitoring, ultimately reducing morbidity and mortality associated with skin cancer. Recommendations include rigorous clinical validation, integration with electronic health records, and greater emphasis on user education to maximize adoption. Future research is suggested to explore longitudinal studies to assess long-term behavioral impacts and scalability of the technology across diverse healthcare systems.