MRI lung lobe segmentation in pediatric cystic fibrosis patients using a recurrent neural network trained with publicly accessible CT datasets.

PURPOSE: To introduce a widely applicable workflow for pulmonary lobe segmentation of MR images using a recurrent neural network (RNN) trained with chest CT datasets. The feasibility is demonstrated for 2D coronal ultrafast balanced SSFP (ufSSFP) MRI. METHODS: Lung lobes of 250 publicly accessible CT datasets of adults were segmented with an open-source CT-specific algorithm. To match 2D ufSSFP MRI data of pediatric patients, both CT data and segmentations were translated into pseudo-MR images that were masked to suppress anatomy outside the lung. Network-1 was trained with pseudo-MR images and lobe segmentations and then applied to 1000 masked ufSSFP images to predict lobe segmentations. These outputs were directly used as targets to train Network-2 and Network-3 with non-masked ufSSFP data as inputs, as well as an additional whole-lung mask as input for Network-2. Network predictions were compared to reference manual lobe segmentations of ufSSFP data in 20 pediatric cystic fibrosis patients. Manual lobe segmentations were performed by splitting available whole-lung segmentations into lobes. RESULTS: Network-1 was able to segment the lobes of ufSSFP images, and Network-2 and Network-3 further increased segmentation accuracy and robustness. The average all-lobe Dice similarity coefficients were 95.0 ± 2.8 (mean ± pooled SD [%]) and 96.4 ± 2.5, 93.0 ± 2.0; and the average median Hausdorff distances were 6.1 ± 0.9 (mean ± SD [mm]), 5.3 ± 1.1, 7.1 ± 1.3 for Network-1, Network-2, and Network-3, respectively. CONCLUSION: Recurrent neural network lung lobe segmentation of 2D ufSSFP imaging is feasible, in good agreement with manual segmentations. The proposed workflow might provide access to automated lobe segmentations for various lung MRI examinations and quantitative analyses.

The impact of segmentation on whole-lung functional MRI quantification: Repeatability and reproducibility from multiple human observers and an artificial neural network.

PURPOSE: To investigate the repeatability and reproducibility of lung segmentation and their impact on the quantitative outcomes from functional pulmonary MRI. Additionally, to validate an artificial neural network (ANN) to accelerate whole-lung quantification. METHOD: Ten healthy children and 25 children with cystic fibrosis underwent matrix pencil decomposition MRI (MP-MRI). Impaired relative fractional ventilation (RFV ) and relative perfusion (RQ ) from MP-MRI were compared using whole-lung segmentation performed by a physician at two time-points (At1 and At2 ), by an MRI technician (B), and by an ANN (C). Repeatability and reproducibility were assess with Dice similarity coefficient (DSC), paired t-test and Intraclass-correlation coefficient (ICC). RESULTS: The repeatability within an observer (At1 vs At2 ) resulted in a DSC of 0.94 ± 0.01 (mean ± SD) and an unsystematic difference of -0.01% for RFV (P = .92) and +0.1% for RQ (P = .21). The reproducibility between human observers (At1 vs B) resulted in a DSC of 0.88 ± 0.02, and a systematic absolute difference of -0.81% (P < .001) for RFV and -0.38% (P = .037) for RQ . The reproducibility between human and the ANN (At1 vs C) resulted in a DSC of 0.89 ± 0.03 and a systematic absolute difference of -0.36% for RFV (P = .017) and -0.35% for RQ (P = .002). The ICC was >0.98 for all variables and comparisons. CONCLUSIONS: Despite high overall agreement, there were systematic differences in lung segmentation between observers. This needs to be considered for longitudinal studies and could be overcome by using an ANN, which performs as good as human observers and fully automatizes MP-MRI post-processing.

Accurate, rapid and reliable, fully automated MRI brainstem segmentation for application in multiple sclerosis and neurodegenerative diseases.

Neurodegenerative disorders, such as Alzheimer's disease (AD) and progressive forms of multiple sclerosis (MS), can affect the brainstem and are associated with atrophy that can be visualized by MRI. Anatomically accurate, large-scale assessments of brainstem atrophy are challenging due to lack of automated, accurate segmentation methods. We present a novel method for brainstem volumetry using a fully-automated segmentation approach based on multi-dimensional gated recurrent units (MD-GRU), a deep learning based semantic segmentation approach employing a convolutional adaptation of gated recurrent units. The neural network was trained on 67 3D-high resolution T1-weighted MRI scans from MS patients and healthy controls (HC) and refined using segmentations of 20 independent MS patients' scans. Reproducibility was assessed in MR test-retest experiments in 33 HC. Accuracy and robustness were examined by Dice scores comparing MD-GRU to FreeSurfer and manual brainstem segmentations in independent MS and AD datasets. The mean %-change/SD between test-retest brainstem volumes were 0.45%/0.005 (MD-GRU), 0.95%/0.009 (FreeSurfer), 0.86%/0.007 (manually edited segmentations). Comparing MD-GRU to manually edited segmentations the mean Dice scores/SD were: 0.97/0.005 (brainstem), 0.95/0.013 (mesencephalon), 0.98/0.006 (pons), 0.95/0.015 (medulla oblongata). Compared to the manual gold standard, MD-GRU brainstem segmentations were more accurate than FreeSurfer segmentations (p < .001). In the multi-centric acquired AD data, the mean Dice score/SD for the MD-GRU-manual segmentation comparison was 0.97/0.006. The fully automated brainstem segmentation method MD-GRU provides accurate, highly reproducible, and robust segmentations in HC and patients with MS and AD in 200 s/scan on an Nvidia GeForce GTX 1080 GPU and shows potential for application in large and longitudinal datasets.

Improving Accuracy of Brainstem MRI Volumetry: Effects of Age and Sex, and Normalization Strategies.

Background: Brainstem-mediated functions are impaired in neurodegenerative diseases and aging. Atrophy can be visualized by MRI. This study investigates extrinsic sources of brainstem volume variability, intrinsic sources of anatomical variability, and the influence of age and sex on the brainstem volumes in healthy subjects. We aimed to develop efficient normalization strategies to reduce the effects of intrinsic anatomic variability on brainstem volumetry. Methods: Brainstem segmentation was performed from MPRAGE data using our deep-learning-based brainstem segmentation algorithm MD-GRU. The extrinsic variability of brainstem volume assessments across scanners and protocols was investigated in two groups comprising 11 (median age 33.3 years, 7 women) and 22 healthy subjects (median age 27.6 years, 50% women) scanned twice and compared using Dice scores. Intrinsic anatomical inter-individual variability and age and sex effects on brainstem volumes were assessed in segmentations of 110 healthy subjects (median age 30.9 years, range 18-72 years, 53.6% women) acquired on 1.5T (45%) and 3T (55%) scanners. The association between brainstem volumes and predefined anatomical covariates was studied using Pearson correlations. Anatomical variables with associations of |r| > 0.30 as well as the variables age and sex were used to construct normalization models using backward selection. The effect of the resulting normalization models was assessed by % relative standard deviation reduction and by comparing the inter-individual variability of the normalized brainstem volumes to the non-normalized values using paired t- tests with Bonferroni correction. Results: The extrinsic variability of brainstem volumetry across different field strengths and imaging protocols was low (Dice scores > 0.94). Mean inter-individual variability/SD of total brainstem volumes was 9.8%/7.36. A normalization based on either total intracranial volume (TICV), TICV and age, or v-scale significantly reduced the inter-individual variability of total brainstem volumes compared to non-normalized volumes and similarly reduced the relative standard deviation by about 35%. Conclusion: The extrinsic variability of the novel brainstem segmentation method MD-GRU across different scanners and imaging protocols is very low. Anatomic inter-individual variability of brainstem volumes is substantial. This study presents efficient normalization models for variability reduction in brainstem volumetry in healthy subjects.

Standardized Assessment of Automatic Segmentation of White Matter Hyperintensities and Results of the WMH Segmentation Challenge.

Quantification of cerebral white matter hyperintensities (WMH) of presumed vascular origin is of key importance in many neurological research studies. Currently, measurements are often still obtained from manual segmentations on brain MR images, which is a laborious procedure. The automatic WMH segmentation methods exist, but a standardized comparison of the performance of such methods is lacking. We organized a scientific challenge, in which developers could evaluate their methods on a standardized multi-center/-scanner image dataset, giving an objective comparison: the WMH Segmentation Challenge. Sixty T1 + FLAIR images from three MR scanners were released with the manual WMH segmentations for training. A test set of 110 images from five MR scanners was used for evaluation. The segmentation methods had to be containerized and submitted to the challenge organizers. Five evaluation metrics were used to rank the methods: 1) Dice similarity coefficient; 2) modified Hausdorff distance (95th percentile); 3) absolute log-transformed volume difference; 4) sensitivity for detecting individual lesions; and 5) F1-score for individual lesions. In addition, the methods were ranked on their inter-scanner robustness; 20 participants submitted their methods for evaluation. This paper provides a detailed analysis of the results. In brief, there is a cluster of four methods that rank significantly better than the other methods, with one clear winner. The inter-scanner robustness ranking shows that not all the methods generalize to unseen scanners. The challenge remains open for future submissions and provides a public platform for method evaluation.

Tracking the Evolution of Cerebral Gadolinium-Enhancing Lesions to Persistent T1 Black Holes in Multiple Sclerosis: Validation of a Semiautomated Pipeline.

BACKGROUND: Some gadolinium-enhancing multiple sclerosis (MS) lesions remain T1-hypointense over months ("persistent black holes, BHs") and represent areas of pronounced tissue loss. A reduced conversion of enhancing lesions to persistent BHs could suggest a favorable effect of a medication on tissue repair. However, the individual tracking of enhancing lesions can be very time-consuming in large clinical trials. PURPOSE: We created a semiautomated workflow for tracking the evolution of individual MS lesions, to calculate the proportion of enhancing lesions becoming persistent BHs at follow-up. METHODS: Our workflow automatically coregisters, compares, and detects overlaps between lesion masks at different time points. We tested the algorithm in a data set of Magnetic Resonance images (1.5 and 3T; spin-echo T1-sequences) from a phase 3 clinical trial (n = 1,272), in which all enhancing lesions and all BHs had been previously segmented at baseline and year 2. The algorithm analyzed the segmentation masks in a longitudinal fashion to determine which enhancing lesions at baseline turned into BHs at year 2. Images of 50 patients (192 enhancing lesions) were also reviewed by an experienced MRI rater, blinded to the algorithm results. RESULTS: In this MRI data set, there were no cases that could not be processed by the algorithm. At year 2, 417 lesions were classified as persistent BHs (417/1,613 = 25.9%). The agreement between the rater and the algorithm was > 98%. CONCLUSIONS: Due to the semiautomated procedure, this algorithm can be of great value in the analysis of large clinical trials, when a rater-based analysis would be time-consuming.