AI-Based Mobile Application for Early Detection of Melanoma Skin Cancer

Thesis Abstract

The escalating incidence of melanoma skin cancer globally underscores the urgent need for accessible and accurate early detection methods to improve patient outcomes and reduce mortality rates. Despite advances in dermatological diagnostics, many cases remain undetected during the early stages due to limited access to specialist care and the variability in clinical interpretation of pigmented skin lesions. This study aims to develop and evaluate an artificial intelligence (AI)-powered mobile application designed to facilitate early detection of melanoma by leveraging image recognition and machine learning techniques. The specific objectives include (1) designing a user-friendly mobile interface integrated with an image capture system; (2) training a convolutional neural network (CNN) model on a comprehensive dataset of dermoscopic images to accurately classify benign and malignant lesions; (3) assessing the diagnostic accuracy, sensitivity, and specificity of the application in real-world conditions; and (4) evaluating user acceptance and potential barriers to adoption among targeted user groups. The research adopts a mixed-methods approach, combining quantitative experimental design with qualitative user surveys. The quantitative component involves a cross-sectional diagnostic accuracy study utilizing a sample of 300 dermoscopic lesion images sourced from publicly available, validated dermatological image repositories, supplemented by data from 150 dermatologist-verified images collected in collaborating clinics. The CNN model will be trained using TensorFlow and Keras, applying transfer learning with a ResNet-50 architecture, followed by validation through k-fold cross-validation to prevent overfitting. The performance metrics—accuracy, sensitivity, specificity, and area under the receiver operating characteristic (ROC) curve—will be computed and analyzed using R statistical software. Additionally, a pilot deployment with 50 consenting users, comprising adults aged 18-65 with pre-existing skin lesions, will be conducted to assess usability, perceived accuracy, and willingness to adopt the technology, analyzed via thematic qualitative analysis. Expected findings include high diagnostic accuracy (anticipated above 85%), with sensitivity and specificity comparable to dermatologists, confirming the potential of AI to augment early melanoma detection. The integration of user feedback is projected to reveal critical usability features and barriers, such as trust in AI diagnosis, ease of use, and concerns about privacy. The study will also identify limitations, including potential biases due to dataset diversity and the need for continuous model updating to account for varied skin types. The contribution to knowledge centers on demonstrating the feasibility and efficacy of a mobile-based AI diagnostic tool tailored for widespread public use, particularly in resource-limited settings where dermatology expertise is scarce. This research bridges the gap between technological advancements in AI and practical application in dermatological screening, providing a scalable, cost-effective solution for early melanoma detection. It further contributes to the theoretical understanding of user acceptance of health technology, applying the Technology Acceptance Model (TAM) to elucidate factors influencing user engagement. In conclusion, this study affirms the potential of AI-driven mobile applications to serve as an adjunct in early melanoma detection, with implications for public health strategies aimed at reducing skin cancer-related morbidity and mortality. Recommendations include integrating the application into existing health systems, conducting large-scale longitudinal validations, and establishing continuous AI model refinement processes. Future research avenues are suggested to examine long-term impact, integration with teledermatology, and expansion to other dermatological conditions, thereby advancing the role of ICT in personalized skin health management.