Abstract
Diagnosing diseases is conventionally done manually. However, humans are prone to errors that might result in misdiagnosis and misclassifications which could cause the patient to undergo wrong treatments. Artificial Intelligence is one of the most prosperous branch of computer science. Harnessing the power of AI in health care is one of the most brilliant techniques to reduce such dependency on human beings for detection, and it is an excellent way to increase detection accuracy. We propose models to detect bone fractures from medical images. The primary goal is to detect these fractures utilizing various deep learning models. The dataset has X-Ray images of different patients organized into Elbow, Finger, Forearm, Hand, Humerus, and Shoulder. It is a superset of images provided by a hospital in India combined with MURA V1.1 Stanford dataset and comprises approximately more than 36000 train and 3000 test images.
This task is a binary classification problem, with images as the input and the presence of fracture as an output (0 - Negative and 1 - Positive). As the first step we aim to perform data preprocessing and design, implement, and analyze a model with high accuracy. Kappa score is used to measure the performance of the model. We could achieve higher Kappa score that other recent state-of-the-art model, respectively equal to 0.769,0.769, 0.49, 0.736, 0.841, and 0.687 for different models experimented on the elbow, shoulder, finger, forearm, hand, and humerus datasets.
The secondary goal is implementing a front end as a web application that integrates the AI models developed in the previous step. This application lets users know about their fracture status by asking them some basic details and instructing them to upload the X-ray image. The application will then process the image using the models embedded at the backend and present all the model's outputs (whether a fracture has been found or not) to the front end. The application integrates python scripts into a Dotnet framework with C# as its base language.