Automatic fetal segmentation from clinical magnetic resonance images (MRIs)
Currently, accurate segmentations of the whole fetus and its organs are done manually in order to quantify the weight and/or volume of the fetus and its internal organs, as well as to visualize the potential presence of pathologies. However, this method is highly time-consuming and heavily dependent on the skills and experience of the operator. This approach also carries inherent challenges caused by the image background noise, motion artefacts, and similar tissue intensity of the organ of interest and the adjacent tissues. The purpose of this study is to develop an automatic method to achieve accurate segmentation of the whole fetus and its organs. We will employ machine learning algorithms including convolutional neural networks to achieve this goal. We will implement deep learning methods, in particular U-Net algorithm, which is a highly successful fully convolutional network that has been shown effective for processing of clinical images. Specific Aims 1: To pre-process all fetal MR images to normalize the dataset and introduce a standardized dataset into the neural network. 2: To label, classify, and segment the fetal images using a highly accurate convolutional neural networks (CNN), such as U-Net. 3: To explore the utility of various image augmentation approaches in amplifying the clinical dataset. 4: To train, validate and test the predictive model to automatically segment the whole fetus and its organs from MRIs.
The summer student will: 1. carry out image segmentation of fetal MRIs 2. implement and compare approaches to 3D image augmentation 3. train U-Net with a variety of datasets 4. assist with drafting a manuscript for publication
Third-year students are welcome to apply if they: 1. have completed signals & systems I and II course 2. have experience in coding in Matlab and python 3. are able to work independently and in a team 4. are reliable, responsible, and have proven ability to multitask and meet deadlines
Dafna Sussman : Automatic fetal segmentation from clinical magnetic resonance images (MRIs) | Wednesday April 3rd 2019 02:49 PM