Skin Scan | Devpost

Inspiration

Skin cancer is the most prevalent form of cancer worldwide. According to the World Health Organization (WHO), one in every three cancers diagnosed globally is a skin cancer, and one in five Americans is projected to develop the disease by age 70. In the United States alone, over five million cases are diagnosed annually—exceeding the total for all other cancers combined. This high incidence not only represents a dramatic public health challenge but also places a substantial economic burden on healthcare systems. Furthermore, the escalating cost of healthcare and the limited accessibility of health resources—especially in rural areas—underscore the urgent need for innovative diagnostic solutions.

In light of rapid advancements in artificial intelligence (AI), our project seeks to leverage these technologies to facilitate earlier detection of skin cancer. By improving early diagnosis, we aim to reduce mortality rates and foster more equitable access to quality healthcare services.

What It Does

Our platform is designed to empower healthcare professionals by providing an online interface that enables the monitoring and prompt diagnosis of skin cancer. Key functionalities include:

Real-Time Diagnosis: The platform allows doctors to assess skin lesions and receive immediate feedback and notifications.
Secure Data Management: A robust login system safeguards patient privacy, securely tracking diagnostic histories and demographic data.
Patient Education: Integrated resources offer patients comprehensive information about skin cancer prevention, diagnosis, and management, promoting awareness among individuals and their families.

How We Built It

Front-End Development

Framework: The user interface is built using the Flask web framework, ensuring a responsive and accessible experience.
Data Storage: A SQLite storage layer is implemented to support secure login systems and manage patient data.
Camera Integration: The MediaDevices API facilitates real-time camera access, enabling dynamic detection of skin lesions.
User Interaction: Chat.js is incorporated to present patient information clearly and promote interactive communication.

Back-End Development

Model Development: We optimized our diagnostic models using Jupyter Notebook, ensuring a robust and efficient workflow.
Classification Techniques: Our classification system employs XGBoost along with image vector embedding. This approach enables trinary and multi-type classification to differentiate between benign and malignant lesions, as well as to identify specific types of skin cancer.

Challenges Encountered

Real-Time Image Processing: Developing a model capable of accurately detecting skin cancer from live webcam images posed significant technical challenges.
Limited Training Data: The constraints of a small sample size for training necessitated innovative strategies for data augmentation and model tuning.

Accomplishments

Despite stringent time constraints, we successfully integrated the front-end and back-end systems to develop a platform with the potential to save thousands of lives. The seamless collaboration between components underlines our commitment to creating a tool that can make a meaningful impact in clinical settings.

Lessons Learned

Our project revealed that simpler models can sometimes outperform more complex architectures. Notably, XGBoost demonstrated superior performance compared to convolutional neural networks (CNNs) in our specific application. This finding emphasizes the importance of tailoring model selection to the specific requirements of the task and validates the need for rigorous testing and informed judgment.

Future Directions for Skin Scan

Looking forward, our ambition is to integrate the Skin Scan platform into existing medical infrastructures. By doing so, we hope to enhance early diagnosis, lower mortality rates, and extend the benefits of advanced diagnostic technology to a broader segment of society—thereby contributing to more equitable and accessible healthcare for all.