Using computer vision for X-Rays
I came up with the idea of using Computer Vision to detect X-Rays when I broke my arm, and the fracture went undetected by several doctors. I wanted to research if we could incorporate AI to assist doctors make less mistakes.
As a Research Lead for ACM Research at UTD, I lead a team of 4 to develop AI-based pediatric wrist trauma diagnostics using Python.
YOLOv7-8 📋
YOLO(You Only Look Once) is an efficient and accurate object detection algorithm that uses deep convolutional neural networks to recognize and localize objects in an image. Unlike traditional object detection algorithms that require multiple passes through an image, YOLO divides the image into a grid and predicts the bounding boxes and class probabilities for each grid cell in a single pass. We used two models for comparison.
- YOLOv7 is one of the latest releases of YOLO, around 120% faster than previous iterations held at the same accuracy.
- YOLOv8 (the latest model) was released within a month of starting this research project. We use the most lightweight model, “YOLOv8nano”, for it’s shorter training times.
Results 🚀
We achieved remarkable precision where 90% of fractures were identified, the rest unclassified. The model also found metal objects with great accuracy. This model often outperformed Doctors in a night ER setting where human error is increased.
After presenting our results in the ACM Research Symposium our research project was awarded 1st place by a panel of machine learning experts amongst 10 research teams.
Here is the link to the repository
Screenshots 📸
