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

Abstract

Funding information Swiss Cystic Fibrosis Society (CFCH)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.

Artificially-generated consolidations and balanced augmentation increase performance of U-net for lung parenchyma segmentation on MR images

“…Recently, a reformation o of 2D images was used to for 3D lung lobe segmentation [ 34 ]. Since 2D CNNs do not consider adjacent consecutive slices for the segmentation, more accurate lung parenchyma or lung lobe segmentation of 3D data is probably achieved with 3D CNNs.…”

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

“…An ANN (observer C) was trained to segment the lung parenchyma of baseline‐images automatically 37 . The recurrent neural network's main layers consist of multi‐dimensional gated recurrent units (MD‐GRU) for voxel‐wise binary classification.…”

“…Furthermore, on‐the‐fly data augmentation is applied during training to increase the network robustness, that is, images and masks are both randomly and slightly scaled, rotated, skewed, distorted, noise is added, and image signal intensity is marginally variated. The MD‐GRU neural network has already shown competitive accuracy for brain segmentation tasks, and specifically for lung segmentation, it previously reached a Dice similarity coefficient of 0.93 37‐39 . The ANN can be found under https://github.com/zubata88/mdgru.…”

“…The MD-GRU neural network has already shown competitive accuracy for brain segmentation tasks, and specifically for lung segmentation, it previously reached a Dice similarity coefficient of 0.93. [37][38][39] The ANN can be found under https://github.com/zubat a88/ mdgru.…”

“…An ANN (observer C) was trained to segment the lung parenchyma of baseline-images automatically. 37 The recurrent neural network's main layers consist of multi-dimensional gated recurrent units (MD-GRU) for voxel-wise binary classification. Furthermore, on-the-fly data augmentation is applied during training to increase the network robustness, that is, images and masks are both randomly and slightly scaled, rotated, skewed, distorted, noise is added, and image signal intensity is marginally variated.…”