Hip prosthesis segmentation

This repository contains the code for the project "X-ray image segmentation"

Team members:

• Iryna Voitsitska
• Taras Lysun
• Sviatoslav Stehnii

Project description

The goal of this project is to create a model and additional functionality to segment X-ray images of hip and calculate the size of potential prosthesis using methods of computer vision and deep learning.

Results of different models and keypoints

Segmentation	Keypoints

Installation

To install the required packages, run the following command:

pip install -r requirements.txt

Currently the inference is not available due to free plan limitations of Roboflow :) But you can open the demo.ipynb to see the results of the model.

Data

The dataset used in this project consists of over 50 different X-ray scans which we labeled manually, after consulting with a therapist. Dataset was created using Roboflow platform, which made the training of the model much easier. For the sake of confidentiality, we cannot share the dataset publicly. But in the test_data folder there are some examples of the images we used for training.

Model

The model we used for segmentation is a very popular YOLO. We fine-tuned the model, and here are some metrics we got: