About My Project
DermaBridge: A Trustworthy Multimodal AI App for Skin Lesion Screening, Explainable Risk Assessment, and Smart Clinical Referral
Problem
Skin cancer can be treated easily if its found in its early stages. Most patients with darker skin tones are usually missed diagnosed and fall under the rader when this stage occurs. 1 in 5 Americans will develop skin cancer in their lifetime with about 60,000 people dying globally. This AI-tool brings a clinical decision and management for early treatment. The DermaBridge addresses the gap using deep learning model-based analysis of dermascopic captured images to identify the patterns and lesions of moles that may be difficult to detect with the naked eye. Not only detection, the AI-tool can assist the origin of the lesion and be able to provide data for personalized care treatment.
Approach
This initiative integrates computer vision, multimodal deep learning, and explainable AI (XAI) to optimize screening accuracy and reliability. Key steps include:
- Training and validating deep learning models using dermatoscopic images and clinical metadata.
- Benchmarking baseline architectures against advanced models like EfficientNetV2.
- Utilizing interpretability frameworks like Grad-CAM and SHAP to provide transparent diagnostic insights.
- Auditing models for fairness and bias across varied demographic groups and skin tones.
- Building a functional prototype equipped with smart referral capabilities. The technical stack leverages Python, PyTorch, OpenCV, CNNs, Vision Transformers, and multimodal architectures, utilizing benchmark datasets like HAM10000 and PAD-UFES-20.
Expected Outcome
The primary objective is to build a functional, AI-powered prototype that provides trustworthy skin lesion screening and interpretable risk feedback. This work will culminate in a research poster and presentation detailing the project’s methodology, results, and broader impact. Additionally, this research will deepen my expertise in machine learning, computer vision, algorithmic fairness, and healthtech innovation.