VICTOR HII JUN SIONG SMK METHODIST SIBU
CR: 0.0000 | 0 Likes |
7 Views |
1 times |
LS: 1.0
Like it? | Support them now!
The global prevalence of skin diseases represents a significant public health challenge, often exacerbated by a lack of awareness and a critical shortage of dermatological resources. In many regions, particularly among underserved and rural communities, delayed diagnoses lead to poor treatment outcomes and prolonged suffering. To address these systemic gaps, Skinalyze was developed as an innovative, AI-driven web application designed to provide accessible, early-stage screening for a wide array of skin conditions. By combining advanced machine learning techniques with a user-centric web interface, the project aims to empower individuals to take proactive control of their dermatological health.
The development of Skinalyze followed a rigorous five-phase methodology. It began with an extensive literature review to identify key visual features of dermatological conditions and evaluate various AI approaches. During the data preparation phase, a robust dataset of over 1,200 labeled images covering 30 different skin conditions was curated. These images underwent preprocessing, including resizing to a standard 224x224 pixel format and normalization, to ensure consistent input for the neural networks.
The core of the system utilizes a Convolutional Neural Network (CNN) built with TensorFlow, supplemented by YOLOv8 for precise image analysis. The model architecture features multiple convolutional, pooling, and fully-connected layers, optimized through an 80/10/10 training-validation-testing split. To make this technology accessible, a responsive web platform was engineered using a Flask backend and a frontend composed of HTML, CSS, and JavaScript. This architecture supports seamless image uploads, rapid result displays, and the delivery of educational content.
Technical validation of the Skinalyze prototype demonstrated high levels of efficacy. The system achieved an average accuracy of 93.94% across 30 skin diseases. Detailed performance metrics further underscore the model’s reliability:
Precision: 91.2%, Recall: 92.5%, F1-Score: 91.8%, Mean Inference Time: 3.3 seconds
Class-wise analysis revealed that common conditions such as acne, eczema, and fungal infections achieved detection accuracies exceeding 95%. While more nuanced distinctions—such as differentiating between rosacea and dermatitis—showed slightly lower accuracies (85-88%), the system remains highly effective for broad-spectrum screening.
Diagnosis History: A personalized dashboard allows users to track past results and monitor condition changes over time.
Moles Tracking: A function that observes changes of moles from time to time for early melanoma detection.
Weather-Responsive Advice: By integrating local weather forecasts, the app offers personalized skincare recommendations tailored to environmental factors.
Intuitive UI: The interface is designed for diverse literacy levels, ensuring that teenagers, parents, and rural populations can navigate the tool with ease.
Beyond simple detection, Skinalyze integrates several features designed to enhance long-term skin health management:Strategic Impact and Future DirectionsSkinalyze aligns with several United Nations Sustainable Development Goals (SDGs), specifically Goal 3 (Good Health & Well-Being), Goal 9 (Industry, Innovation, and Infrastructure), and Goal 10 (Reduced Inequalities). By lowering the barriers to dermatologic screening, the application promotes equitable health access and facilitates STEM learning within the community.While the current prototype is highly successful, challenges remain regarding dependence on image quality and the classification of very similar conditions. Future iterations of Skinalyze will focus on expanding data diversity to include more skin tones and refining the user interface to further enhance accessibility. Ultimately, Skinalyze serves as a powerful proof-of-concept for how AI can bridge the gap between specialized medical expertise and everyday public health needs.
