Search

BACKGROUND AND PURPOSE: Accurate and reliable detection of white matter hyperintensities and their volume quantification can provide valuable clinical information to assess neurologic disease progression. In this work, a stacked generalization ensemble of orthogonal 3D convolutional neural networks, StackGen-Net, is explored for improving automated detection of white matter hyperintensities in 3D T2-FLAIR images. MATERIALS AND METHODS: Individual convolutional neural networks in StackGen-Net were trained on 2.5D patches from orthogonal reformatting of 3D-FLAIR (n = 21) to yield white matter hyperintensity posteriors. A meta convolutional neural network was trained to learn the functional mapping from orthogonal white matter hyperintensity posteriors to the final white matter hyperintensity prediction. The impact of training data and architecture choices on white matter hyperintensity segmentation performance was systematically evaluated on a test cohort (n = 9). The segmentation performance of StackGen-Net was compared with state-of-the-art convolutional neural network techniques on an independent test cohort from the Alzheimer’s Disease Neuroimaging Initiative-3 (n = 20). RESULTS: StackGen-Net outperformed individual convolutional neural networks in the ensemble and their combination using averaging or majority voting. In a comparison with state-of-the-art white matter hyperintensity segmentation techniques, StackGen-Net achieved a significantly higher Dice score (0.76 [SD, 0.08], F1-lesion (0.74 [SD, 0.13]), and area under precision-recall curve (0.84 [SD, 0.09]), and the lowest absolute volume difference (13.3% [SD, 9.1%]). StackGen-Net performance in Dice scores (median = 0.74) did not significantly differ (P = .22) from interobserver (median = 0.73) variability between 2 experienced neuroradiologists. We found no significant difference (P = .15) in white matter hyperintensity lesion volumes from StackGen-Net predictions and ground truth annotations. CONCLUSIONS: A stacked generalization of convolutional neural networks, utilizing multiplanar lesion information using 2.5D spatial context, greatly improved the segmentation performance of StackGen-Net compared with traditional ensemble techniques and some state-of-the-art deep learning models for 3D-FLAIR.

BACKGROUND AND PURPOSE: TSE-based T2-weighted imaging of the spine has long scan times. This work proposes a fast imaging protocol using variable refocusing flip angles, optimized for blurring and specific absorption rate. MATERIALS AND METHODS: A variable refocusing flip angle echo-train was optimized for the spine to improve the point spread function and minimize the specific absorption rate, yielding images with improved spatial resolution and SNR compared with the constant flip angle sequence. Data were acquired from 51 patients (35 lumbar, 16 whole-spine) using conventional TSE and the proposed sequence, with a single-shot variant for whole-spine. Noninferiority analysis was performed to evaluate the efficiency of the proposed technique. RESULTS: The proposed multishot sequence resulted in a 2× shorter scan time with a >1.5× lower specific absorption rate. The variable flip angle sequence was noninferior to the conventional TSE (P < .025) for all image-quality and clinical criteria except signal-to-noise ratio for the lumbar spine protocol. However, mean image scores for the TSE-variable refocusing flip angle were ≥4.3 for all criteria, and concordance analysis showed high agreement (>90%) with the TSE, indicating clinical equivalence. The single-shot sequence resulted in 4× shorter whole-spine scans, and image scores were ≥4.4 for all criteria, attesting to its clinical utility. CONCLUSIONS: We present a fast T2-weighted spine protocol using variable refocusing flip angles, including a single-shot variant. The sequences have better point spread function behavior than their constant flip angle counterparts and, being faster, should be less sensitive to patient motion, often seen in the longer TSE scans.