Development of an AI-powered Mobile App for Early Skin Cancer Detection

Thesis Abstract

Skin cancer remains one of the most prevalent and potentially fatal dermatological conditions worldwide, yet early detection remains a significant challenge due to limited access to specialized healthcare and reliance on subjective visual assessments. This study aims to develop an innovative, artificial intelligence (AI)-powered mobile application that facilitates early detection of skin cancer through automated image analysis and pattern recognition. The specific objectives include designing a robust AI algorithm capable of classifying dermoscopic images with high accuracy, developing a user-friendly mobile interface for lay users, and evaluating the app's diagnostic performance in real-world settings. Employing a mixed-methods research design, the study integrates quantitative algorithm development and validation with qualitative user experience assessments. The population comprises 3,000 dermoscopic skin lesion images sourced from publicly accessible datasets such as the International Skin Imaging Collaboration (ISIC) archive, complemented by 500 images collected via clinical collaborations. For the usability evaluation, 200 participants representing diverse demographics will be recruited from dermatology clinics and community health centers. Data collection instruments include a standardized dermoscopic image dataset, surveys assessing user interface usability based on the System Usability Scale (SUS), and semi-structured interviews for contextual insights. The AI model leverages convolutional neural networks (CNNs), specifically fine-tuning pretrained models such as ResNet-50, applying transfer learning techniques. Data analysis encompasses model training using Python TensorFlow/Keras, with performance metrics including accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC). The model's diagnostic capability will be validated against expert dermatologist assessments, and statistical significance will be determined through receiver operating characteristic analysis. User experience data will undergo thematic analysis to identify usability facilitators and barriers. Expected findings predict that the AI model will achieve diagnostic accuracy exceeding 85%, with sensitivity and specificity both surpassing 80%, demonstrating its potential as an effective screening tool. The app's usability is anticipated to score highly on the SUS, indicating suitability for layperson use. The study is expected to reveal critical insights into the integration of AI diagnostics within mobile health platforms and their acceptance among diverse user groups. The development of the app conceptually extends the Technology Acceptance Model (TAM) by incorporating perceived ease of use and perceived usefulness, which will be empirically tested. This research contributes new knowledge at the intersection of dermatology, artificial intelligence, and mobile health technology, offering a scalable, accessible solution for early skin cancer detection. It addresses existing gaps related to AI integration in mobile platforms, user engagement, and validation in community settings. The primary conclusion underscores the feasibility and efficacy of deploying AI-powered mobile applications for dermatological screening, with implications for enhancing early diagnosis, reducing healthcare disparities, and informing policy on digital health implementations. Recommendations emerging from this study include the necessity for further large-scale validation across diverse populations, integration with teledermatology services, and the development of regulatory frameworks to ensure safety, privacy, and efficacy. Future research should explore longitudinal studies assessing app impact on clinical outcomes and behavior change in skin self-monitoring. Overall, this study advocates for the strategic incorporation of AI-driven tools into primary healthcare, aiming to improve early detection rates of skin cancer and ultimately reduce morbidity and mortality associated with this disease.