Open Collections

Efficient deep learning of 3D structural brain MRIs for manifold learning and lesion segmentation with application to multiple sclerosis

University of British Columbia

Deep learning methods have shown great success in many research areas such as object recognition, speech recognition, and natural language understanding, due to their ability to automatically learn a hierarchical set of features that is tuned to a given domain and robust to large variability. This motivates the use of deep learning for neurological applications, because the large variability in brain morphology and varying contrasts produced by different MRI scanners makes the automatic analysis of brain images challenging. However, 3D brain images pose unique challenges due to their complex content and high dimensionality relative to the typical number of images available, making optimization of deep networks and evaluation of extracted features difficult. In order to facilitate the training on large 3D volumes, we have developed a novel training method for deep networks that is optimized for speed and memory. Our method performs training of convolutional deep belief networks and convolutional neural networks in the frequency domain, which replaces the time-consuming calculation of convolutions with element-wise multiplications, while adding only a small number of Fourier transforms. We demonstrate the potential of deep learning for neurological image analysis using two applications. One is the development of a fully automatic multiple sclerosis (MS) lesion segmentation method based on a new type of convolutional neural network that consists of two interconnected pathways for feature extraction and lesion prediction. This allows for the automatic learning of features at different scales that are optimized for accuracy for any given combination of image types and segmentation task. Our network also uses a novel objective function that works well for segmenting underrepresented classes, such as MS lesions. The other application is the development of a statistical model of brain images that can automatically discover patterns of variability in brain morphology and lesion distribution. We propose building such a model using a deep belief network, a layered network whose parameters can be learned from training images. Our results show that this model can automatically discover the classic patterns of MS pathology, as well as more subtle ones, and that the parameters computed have strong relationships to MS clinical scores.

Text

2016-07-13

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/58427

Biomedical Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/