At the University of Kentucky Health Center, muscle injuries are commomplace, occurring for people who experience an accident or an athlete who gets injured in the intensity of their sport. While MRI imaging can provide a deep understanding of the nature of the injury, it can be more difficult to differentiate thigh muscles. These partitions are often made by hand, which is time consuming and subject to human error. The solution would be to have an automated way of recognizing the shapes of thigh muscles and highlight them in a way that can be useful for diagnoses and treatments.

Our project's goal was to provide a basis through which the UK Health Department can automatically identify muscles of the thigh. In our two-week long work, we were tasked with creating a model to isolate the vastus lateralis, a muscle located on the lateral side of the thigh.

The Freeman-Wilson-Delgado-Ellis (FWDE) model was trained with images from real subjects of varying thigh sizes and time frames. Our model enacts a process of classifying images pixel-by-pixel, referred to commonly as image segmentation. Using the U-Net architecture, our model downscales the images after each layer, then upscales them into a matrix of classification values, or binary masks. These binary masks, when overlayed on the original image, will highlight the area of interest where the vastus lateralis is located.

Provided that we can identify the vastus lateralis reliably enough, the next steps would be for the UK Health Center to train the FWDE Model to identify more muscles of the thigh.

Libraries used include Tensorflow (through Keras and Tensorflow Datasets), Numpy, Scipy, and Matplotlib.

The GitHub repository for this project, at this time, is private, as confidential patient data is privatized appropriately.