Mir.AI – AI Powered MRI Enhancement
01. Overview
Developed an AI model leveraging Generative Adversarial Networks to enhance 1.5T MRI scans to 3T scans and up, improving accessibility while reducing costs. Engineered a fast and efficient image processing pipeline with FastAPI, achieving up to 50 epochs in an hour for training models. Applied insights from Cycle-Free CycleGAN using invertible generators to refine unsupervised image enhancements. Leveraged advanced AI techniques to improve diagnostic accuracy and accessibility in healthcare.
02. Deep Dive
This project addresses the critical need for high-quality medical imaging in under-resourced areas. By utilizing advanced Generative Adversarial Networks (GANs), particularly Cycle-Free CycleGAN architectures with invertible generators, we successfully upscaled and enhanced 1.5T MRI scans to resemble the quality of 3T MRI machines. The system effectively reduces the need for expensive hardware upgrades while providing doctors with clearer, more precise diagnostic images. Our robust pipeline built on FastAPI ensures that the processing is not only accurate but also incredibly efficient, processing extensive datasets and reaching 50 epochs in roughly an hour. This efficiency enables rapid iteration and deployment, ultimately aiming to democratize access to top-tier healthcare diagnostics across the globe.
Project Info
Achievement
Hackalytics 2025 Healthcare Track Winner
Timeline
Jan 2025 - Present