TrueScan is a computer vision system designed to verify the authenticity of medical scans (specifically knee X-rays as of now) and detect anomalies. It utilizes an ensemble of state-of-the-art Deep Learning models to differentiate between valid medical imaging and images that have been potentially manipulated using AI methods, serving the results via a user-friendly web dashboard.
The ensemble consists of ResNet50, VGG19_BN (Batch Normalization) and YOLOv8m-cls.
- Ensemble Architecture: Aggregates predictions from ResNet50, VGG19_BN, and YOLOv8m-cls using a voting mechanism to achieve high-confidence classification.
- Automated Reporting: Generates downloadable PDF reports with prediction confidence and patient details.
- Web Dashboard: A full-stack Flask application with user (doctor) authentication, user access management, scan image analysis and report generation.
If you want this project on your local machine
git clone https://github.com/prajwalnayaka/TrueScan.git
pip install -r requirements.txt
cd Python_Scripts
python app.py OR flask runImportant Note: This repo does NOT include the trained models because of the size limit. You can download the models from this Google Drive. Replace the path to the models accordingly in test.py.
-
Prajwal Nayaka T (GitHub)
- Trained the Core ML models (ResNet50, VGG19_BN, YOLOv8m-cls).
- Engineered the Ensemble Voting Mechanism and Inference Pipeline.
- Developed the Report Generation module.
- Integrated above mentioned features into the Flask API. Note: Training scripts located in /Training.
-
Pragya MV (GitHub)
- Designed and developed frontend files along with styling.
- Initialized the database structure.
- Built the baseline Flask API.