Anatomy‐guided joint tissue segmentation and topological correction for 6‐month infant brain MRI with risk of autism

Abstract

Tissue segmentation of infant brain MRIs with risk of autism is critically important for characterizing early brain development and identifying biomarkers. However, it is challenging due to low tissue contrast caused by inherent ongoing myelination and maturation. In particular, at around 6 months of age, the voxel intensities in both gray matter and white matter are within similar ranges, thus leading to the lowest image contrast in the first postnatal year. Previous studies typically employed intensity images and tentatively estimated tissue probabilities to train a sequence of classifiers for tissue segmentation. However, the important prior knowledge of brain anatomy is largely ignored during the segmentation. Consequently, the segmentation accuracy is still limited and topological errors frequently exist, which will significantly degrade the performance of subsequent analyses. Although topological errors could be partially handled by retrospective topological correction methods, their results may still be anatomically incorrect. To address these challenges, in this article, we propose an anatomy-guided joint tissue segmentation and topological correction framework for isointense infant MRI. Particularly, we adopt a signed distance map with respect to the outer cortical surface as anatomical prior knowledge, and incorporate such prior information into the proposed framework to guide segmentation in ambiguous regions. Experimental results on the subjects acquired from National Database for Autism Research demonstrate the effectiveness to topological errors and also some levels of robustness to motion. Comparisons with the state-of-the-art methods further demonstrate the advantages of the proposed method in terms of both segmentation accuracy and topological correctness.

Hierarchical Fully Convolutional Network for Joint Atrophy Localization and Alzheimer's Disease Diagnosis Using Structural MRI

“…In recent years, deep convolutional neural networks (CNNs) are showing increasingly successful applications in various medical image computing tasks [24], [25], [26], [27], [28], [29], [30]. Capitalizing on task-oriented, high-nonlinear feature extraction for classifier construction, CNNs have also been applied to developing advanced CAD methods for brain disease diagnosis [31], [32], [33], [34].…”

A Survey on Computer-Aided Diagnosis of Brain Disorders through MRI Based on Machine Learning and Data Mining Methodologies with an Emphasis on Alzheimer Disease Diagnosis and the Contribution of the Multimodal Fusion

“…However it often does not provide perfect segmentation. Therefore, to properly estimate the threshold of homogeneity or heterogeneity of the brain regions, it is beneficial to combine all or most of the measures for better segmentation results [138][139][140][141]. For example, the interest of integrating prior knowledge and contextual information has opened up good tracks of research in the segmentation of medical images [47,142].…”

Semi-Supervised Learning With Deep Embedded Clustering for Image Classification and Segmentation

“…To evaluate the performance of our proposed method (SSLDEC), we performed several experiments on various benchmark datasets under standard experimental conditions where we were able to compare our results with the results reported by other semi-supervised learning methods. In particular, we applied our method to the following datasets: Two half moons drawn using the Sklearn package [36]The MNIST dataset, consisting of 70,000 hand written digits [5],The SVHN, consisting of around 125,000 pictures of digits from street house numbers [37],The iSeg, consisting of 3D T1-weighted (T1w) and T2-weighted (T2w) brain MRI scans of 23 six-month old infants as part of the iSeg2017 challenge [1]. Brain tissue segmentation in iSeg is challenging as gray matter and white matter appear with isointensity values on both T1w and T2w MRI scans around six months of age.…”

“…For instance, reliably labeling or segmenting large medical imaging data requires excessive amount of work by a group of expert radiologists or well-trained technologists. For example, manual segmentation of each brain MRI scan in the isointense infant brain MRI segmentation challenge (iSeg2017) took, on average, one week of a neuroradiologist’s time [1]. On the other hand, in many domains including medical imaging, getting access to large unlabeled data is relatively easy and inexpensive.…”

“…We apply this method, called Semi-Supervised Learning with Deep Embedded Clustering (SSLDEC) to benchmark image classification datasets that have been routinely used in the evaluation and comparison of semi-supervised learning algorithms, as well as a challenging medical image segmentation task, i.e. the isointense infant brain MRI segmentation based on the iSeg2017 challenge [1]. Experimental results show that our proposed method achieved competitive results for semi-supervised learning on MNIST, SVHN and iSeg2017 when only a small portion of data is labeled.…”