SmartMRI Framework for Segmentation of MR Images Using Multiple Deep Learning Methods
Cristina-Petruta Manoila, Alexe Ciurea, Felix Albu
Abstract
MRI analysis frameworks are practical tools for accelerating the analysis by clinical researchers. However, over time, the focus has shifted to creating rigid frameworks. This paper presents a new framework that allows researchers to run different deep learning models based on predefined parameters suitable for automatically delineating the region of interest from magnetic resonance (MR) images of the knee joint. In addition, we present different deep learning methods for the automated segmentation of knee bones trained using data from the SKI10 challenge, concentrating on a convolutional neural network (CNN), which has proven promising potential in musculoskeletal imaging applications. We also propose a novel method that weighs the average of the surrounding pixels when the image is downsampled within a CNN.