An AI-powered diagnostic tool for identifying Monkeypox, Measles, Chickenpox, and normal skin conditions.
Problem Statement
In May 2022, an outbreak of mpox appeared suddenly and rapidly spread across Europe, the Americas and then all six WHO regions, with 110 countries reporting about 87 thousand cases and 112 deaths.
The global outbreak of mpox was declared a public health emergency of international concern (PHEIC) on 23 of July 2022. the major issues faced were:
- Difficulty in early diagnosis due to similarity with Chickenpox
- Delayed diagnosis due to time required for PCR tests
- Inaccurate diagnosis due to high volume and human error
An easily accessible and cost-effective solution was required to accurately diagnose the disease instantly and with high accuracy.
Solution Overview
“Diag-Assist” is an AI powered solution, built to predict viral exanthems (such as monkeypox, chickenpox, measles etc) using user provided images by classifying images of skin conditions into appropriate category using a deep learning model. The solution provides following benefits:
- Instant Prediction
- Easy Access
- Cost Effective
- Highly Accurate
In addition, this solution is open source, highly scalable and easy to use for wider adoption. The solution provides a very viable alternative to traditional PCR tests, which may not be available in all regions.
Figure 1: Architecture Diagram
Solution Demo Video
You can try the app yourself.
Future Scope
The current implementation of “Diag-Assist” can be expanded further to explore additional areas into the medical field. Some of them are listed below:
- Expansion to Other Skin Conditions:
- Integration with Telemedicine Platforms:
- Enhanced Diagnostic Accuracy:
- Real-time Monitoring and Alerts:
- Personalized Treatment Recommendations:
- Cross-platform Accessibility
- Integration with Electronic Health Records (EHR):
- AI-driven Research and Data Analysis:
- Partnerships with Healthcare Providers and Pharmacies:
- Global Health Impact:
- Insurance and Financial Integration
By focusing on these future enhancements, the AI-powered diagnostic solution can evolve into a comprehensive healthcare tool that not only aids in immediate diagnosis but also supports long-term health management and disease prevention on a global scale.
Technical Implementation
Dataset Selection and Processing
Data Augmentation techniques like rotation, zooming and flipping were used to enhance the dataset size and diversity leading to a augmented dataset with over 2800 images.
Dataset has been sourced from Kaggle where a total of 770 images was classified into 4 categories : Chickenpox, Monkeypox, Measles and Normal.
Table 1: Augmented Dataset comparison
DL Model Selection
•Exhaustive comparison of ResNet with DenseNet, Xception, VGG16, VGG19.
•Fine-tuned ResNet model has been selected with it’s peak accuracy reaching over 90%.
Table 2: Accuracy table
“AI is the new electricity.”
– Andrew Ng, Co-founder of Google Brain