Automated multiclass tissue segmentation of clinical brain MRIs with lesions

David A Weiss,Rachit Saluja,Long Xie,James C Gee,Leo P Sugrue,Abhijeet Pradhan,R Nick Bryan,Andreas M Rauschecker,Jeffrey D Rudie

doi:10.1016/j.nicl.2021.102769

Abstract

Delineation and quantification of normal and abnormal brain tissues on Magnetic Resonance Images is fundamental to the diagnosis and longitudinal assessment of neurological diseases. Here we sought to develop a convolutional neural network for automated multiclass tissue segmentation of brain MRIs that was robust at typical clinical resolutions and in the presence of a variety of lesions. We trained a 3D U-Net for full brain multiclass tissue segmentation from a prior atlas-based segmentation method on an internal dataset that consisted of 558 clinical T1-weighted brain MRIs (453/52/53; training/validation/test) of patients with one of 50 different diagnostic entities (n = 362) or with a normal brain MRI (n = 196). We then used transfer learning to refine our model on an external dataset that consisted of 7 patients with hand-labeled tissue types. We evaluated the tissue-wise and intra-lesion performance with different loss functions and spatial prior information in the validation set and applied the best performing model to the internal and external test sets. The network achieved an average overall Dice score of 0.87 and volume similarity of 0.97 in the internal test set. Further, the network achieved a median intra-lesion tissue segmentation accuracy of 0.85 inside lesions within white matter and 0.61 inside lesions within gray matter. After transfer learning, the network achieved an average overall Dice score of 0.77 and volume similarity of 0.96 in the external dataset compared to human raters. The network had equivalent or better performance than the original atlas-based method on which it was trained across all metrics and produced segmentations in a hundredth of the time. We anticipate that this pipeline will be a useful tool for clinical decision support and quantitative analysis of clinical brain MRIs in the presence of lesions.

Highlights

Magnetic resonance imaging (MRI) plays a fundamental role in brain structure assessment in health and disease
To assess the tissue segmentation accuracy of our method within lesions, for all validation and test set cases with lesions (n = 62), lesions were manually segmented on T2-FLAIR sequences into “gray matter lesion” and “white matter lesion” classes by JR, a neuroradiology fellow with 2 years of post-residency experience, using ITK-SNAP (Blitstein and Tung, 2007)
Our aim was to assess our model’s ability to identify un derlying tissue types inside of lesions, cortical gray matter and deep gray matter lesions were condensed into a single gray matter lesion label

Summary

Introduction

Magnetic resonance imaging (MRI) plays a fundamental role in brain structure assessment in health and disease. We developed and evaluated a rapid, automated, deep learning pipeline for segmentation of brain tissues on clinical T1weighted MRIs in the presence of lesions from a variety of pathologies. EM techniques are stochastic, making them susceptible to local optima (Avants et al, 2011) with variable results across runs (Tustison et al, 2014). In addition such models must be fitted for each new patient and convergence can be slow (Jamshidian and Jennrich, 1997), limiting their utility in a clinical workflow where exams are read on the order of minutes to tens of minutes (McDonald et al, 2015). Deep learning offers a solution to these problems, as prediction is fast and deterministic once the model weights are learned

Objectives

Methods

Results

Discussion

Conclusion