Skin disease detection with AI-powered image recognition

About the project

A European startup sought to develop an AI solution for early diagnosis of skin disorders, utilizing a cloud-based, computer-aided service. In order to attract major healthcare players and promote the use of AI in healthcare products, the company aimed to build a prototype of an AI application core specifically designed for early diagnosis of common types of skin disorders.

Business challenge

The main objective of this project was to build a prototype of an AI application, with a goal to improve patient outcomes within the diagnostics process and enhance the potential of AI technology in healthcare.

• Analyzing complex skin disorders that vary greatly in symptoms, severity, and underlying causes.
• Delivering value to both patients and specialists by enabling patients to make timely decisions about seeking consultations with doctors and providing specialists with preliminary diagnostics and recommendations from the AI system.
• Ensuring compliance with medical industry regulations and standards.
• Creating a cost-effective solution that saves time and money for both the client and end users.
• Demonstrating the capabilities of the solution to specialists in Al, healthcare, and potential investors.

Technical Challenge

The HW.Tech team faced several technical challenges while developing the prototype for a skin disorder diagnostic solution:

• Selecting the optimal algorithm to analyze four types of skin diseases, with the potential for scaling to more types in future iterations.
• Collecting free medical images from Dermnet.com, an open-source skin diseases library, and processing the images to comply with the model’s needs, including separating the skin and other elements in the photo, highlighting areas with the highest density of problem areas, and scaling to improve the quality of image recognition.
• Deploying the prototype in Jupyter Notebook, a web-based interactive development environment for creating, sharing, and collaborating on code, and providing all documentation and prototype materials to the client.
• Ensuring the solution works in real-time for individual users, with the processes of image recognition and classification model constantly recomputed by an Al engine.

Solution

Our R&D team was in charge of developing a prototype for the skin diagnostic model based on computer vision algorithms, which required advanced knowledge and expertise in AI and machine learning. Here’s a breakdown of what we did:

• Conducted thorough research and analysis to select the most optimal solution to diagnose early skin problems within the prototype.
• Created documentation that estimates the resources and time required to launch this solution into production.
• Designed a prototype with a fully functioning ML model.

Result

During the project, our team:

• Created a fully functional prototype of the AI solution, which was delivered to the customer on time and according to the initial requirements.
• Built a solution that could handle and process thousands of images with disease examples within seconds, and provide primary diagnostics and recommendations to healthcare specialists.
• Ensured the accuracy and effectiveness of the machine learning model for providing personalized and targeted predictions on skin disorders.

Technologies and tools: CNN, Jypyter, Keras, OpenCV, Python, AI, ML, Computer Vision