Using a Convolutional Neural Network to Segment the Knee
College of Engineering
Department of Mechanical & Biomedical Engineering
Dr. Clare Fitzpatrick
For this project, I will develop an algorithm to segment the knee which consists of the femur, tibia, and patella bones and their respective connected cartilage. The algorithm is based on a convolutional neural network (CNN) which will be fed two different data sets. The first data set consists of 26 magnetic resonance (MR) images of patients who had reoccurring patellar dislocation. The second data set is from the Osteoarthritis Initiative (OAI) dataset which consists of 20 more patients. This combined total of 46 patient’s MRI images will be segmented into sagittal slices with each MRI consisting of around 130 sagittal slices giving a total of 5980 sagittal slices. The number of sagittal slices will fluctuate depending on the resolution of the MR images and the type of scanner used. Once a working model is developed, I will validate and test it on more of the OAI dataset which consists of 4,796 patients. The overall objective of this project was to efficiently segment subject-specific knee structures, and then to later develop a way of transforming these CNN generated segmentations into a 3D model. These models will ultimately be used for subject-specific finite element simulations to evaluate joint mechanics.
Wright, Donovan and Alvarez, Oliver, "Using a Convolutional Neural Network to Segment the Knee" (2019). 2019 Undergraduate Research and Scholarship Conference. 156.