Efficient Claustrum Segmentation in T2-weighted Neonatal Brain MRI Using Transfer Learning from Adult Scans

Abstract

Purpose Intrauterine claustrum and subplate neuron development have been suggested to overlap. As premature birth typically impairs subplate neuron development, neonatal claustrum might indicate a specific prematurity impact; however, claustrum identification usually relies on expert knowledge due to its intricate structure. We established automated claustrum segmentation in newborns. Methods We applied a deep learning-based algorithm for segmenting the claustrum in 558 T2-weighted neonatal brain MRI of the developing Human Connectome Project (dHCP) with transfer learning from claustrum segmentation in T1-weighted scans of adults. The model was trained and evaluated on 30 manual bilateral claustrum annotations in neonates. Results With only 20 annotated scans, the model yielded median volumetric similarity, robust Hausdorff distance and Dice score of 95.9%, 1.12 mm and 80.0%, respectively, representing an excellent agreement between the automatic and manual segmentations. In comparison with interrater reliability, the model achieved significantly superior volumetric similarity (p = 0.047) and Dice score (p < 0.005) indicating stable high-quality performance. Furthermore, the effectiveness of the transfer learning technique was demonstrated in comparison with nontransfer learning. The model can achieve satisfactory segmentation with only 12 annotated scans. Finally, the model’s applicability was verified on 528 scans and revealed reliable segmentations in 97.4%. Conclusion The developed fast and accurate automated segmentation has great potential in large-scale study cohorts and to facilitate MRI-based connectome research of the neonatal claustrum. The easy to use models and codes are made publicly available.

Aberrant claustrum structure in preterm-born neonates: an MRI study

“…Due to the intricate shape, the neonate claustrum is not listed in standard brain atlases or the Draw-EM coregistration (Makropoulos et al, 2018;. As manual segmentation both show substantial inter-rater variability and is very laborious, we used a previously published deep learning algorithm to segment the neonate claustrum (Neubauer et al, 2022) (Fig. 1C, dark blue).…”

“…In more detail, using a deep learning segmentation approach, a convolutional neural network was pretrained with manual claustrum segmentations in adult scans and extensively evaluated; it performed clearly above manual segmentation including intra-rater reliability (Li et al, 2021). Then, using an additional transfer learning approach, the 'adult-claustrumfocused' network was adapted for neonate claustrum segmentation by training and optimizing with 20 scans of preterm-born neonates of the dHCP including adjusted hyperparameters and data augmentation methods (Neubauer et al, 2022). The resulting 'neonate-claustrumfocused' network was also extensively evaluated with 538 scans of the dHCP using different volumetric and spatial scores (Neubauer et al, 2022).…”

“…Then, using an additional transfer learning approach, the 'adult-claustrumfocused' network was adapted for neonate claustrum segmentation by training and optimizing with 20 scans of preterm-born neonates of the dHCP including adjusted hyperparameters and data augmentation methods (Neubauer et al, 2022). The resulting 'neonate-claustrumfocused' network was also extensively evaluated with 538 scans of the dHCP using different volumetric and spatial scores (Neubauer et al, 2022). Critically, automated neonate claustrum segmentation exceeded substantially the inter-rater reliability of manual segmentation (Neubauer et al, 2022).…”

“…The resulting 'neonate-claustrumfocused' network was also extensively evaluated with 538 scans of the dHCP using different volumetric and spatial scores (Neubauer et al, 2022). Critically, automated neonate claustrum segmentation exceeded substantially the inter-rater reliability of manual segmentation (Neubauer et al, 2022). The code was made publicly available on GitHub (https://github.com/hongweilibran/claustrum_multi_view).…”

“…As this is the first analysis of the human neonatal claustrum, the study provides contextual background about the claustrum development around term: first, we evaluated the claustrum structure across the spectrum of 377 term-born neonates of the dHCP around term; second, preterm claustrum development was examined by a longitudinal evaluation of 53 preterm-born neonates of the dHCP who passed a follow-up scan in the range of GA 38 to 45. To perform all analyses of the claustrum, we used a previously published deep learning algorithm to segment the neonate claustrum (Neubauer et al, 2022), which was adapted from the corresponding adult claustrum segmentation tool (Li et al, 2021). Claustrum macrostructure was assessed with volume estimations in T2-weighted MRI scans.…”