AfriScan

Inspiration

Tuberculosis remains one of the world’s deadliest infectious diseases, especially in regions where access to radiologists is limited. In many clinics, chest X-rays are taken but cannot be reviewed quickly by specialists, which delays diagnosis and treatment. AfriScan was developed to help clinicians quickly triage chest X-rays and identify patients who may require urgent TB testing.By combining computer vision and explainable AI, the system assists healthcare providers in identifying suspicious lung abnormalities and prioritizing patients for further testing.

What it does

AfriScan is an AI-powered chest X-ray triage assistant. The system allows a clinician to upload a chest X-ray image, after which the AI:

• predicts tuberculosis risk
• provides a confidence score
• highlights suspicious lung regions using explainable AI
• generates a structured triage summary with suggested clinical tests The goal is not to replace clinicians but to support them with faster triage and decision assistance. ## How we built it AfriScan consists of three main components. Frontend A React dashboard that allows clinicians to upload chest X-rays and view AI results. Backend A FastAPI server that processes uploaded images and runs a deep learning model. AI Model A ResNet18 convolutional neural network trained to classify chest X-rays for TB risk. Grad-CAM explainable AI is used to highlight lung regions influencing the prediction. The system performs: Image validation to ensure the input is a chest X-ray AI prediction of TB probability Risk classification (Low / Moderate / High) Grad-CAM localization of suspicious lung regions ## Challenges we ran into One major challenge was ensuring the AI model’s predictions were explainable. In healthcare applications, clinicians must be able to understand why the model made a decision. To address this, we implemented Grad-CAM visualization to highlight the image regions influencing the model's prediction. Another challenge was validating uploaded images to ensure that non-medical images are rejected to prevent incorrect predictions. ## Accomplishments that we're proud of One of our biggest accomplishments was successfully building a complete end-to-end AI system within a very short hackathon timeframe. AfriScan Assist can take a chest X-ray image, analyze it using a deep learning model, and return a tuberculosis risk prediction along with visual explanations using Grad-CAM heatmaps. We are also proud of integrating multiple components into a single working platform: an intuitive frontend interface for uploading scans, a backend API that handles image processing and AI inference, and an explainable AI visualization layer that highlights suspicious regions in the lungs. This makes the output more interpretable for clinicians and not just a “black box” prediction. Another accomplishment is that the system is designed with real-world healthcare challenges in mind, particularly for regions with limited access to radiologists. By providing quick triage insights, AfriScan Assist demonstrates how AI can support early detection and improve healthcare accessibility. ## What we learned During development we learned about:
• building AI pipelines for medical imaging
• implementing explainable AI techniques
• integrating machine learning models with web applications
• designing clinical-style dashboards for decision support ## What's next for AfriScan Future improvements could include: The next step is expanding the model training with a much larger and more diverse medical imaging dataset to improve accuracy and robustness. We also plan to integrate the platform with cloud infrastructure to support real-time deployment in clinics and remote healthcare facilities. Future versions of AfriScan could support additional lung conditions such as pneumonia, lung cancer indicators, and other respiratory diseases. We also plan to improve the collaboration features so that doctors can review scans together, share case notes, and receive AI-assisted recommendations. Another exciting direction is integrating multimodal AI models that can combine patient symptoms, medical history, and imaging data to provide more comprehensive clinical decision support. Ultimately, the vision is to turn AfriScan into a scalable AI-powered medical imaging platform that helps healthcare professionals detect diseases earlier and deliver better patient outcomes.

Built With

• fastapi
• grad-cam
• python
• pytorch
• react
• resnet18
• tailwindcss
• typescript
• vite