An automatic multi-tissue human fetal brain segmentation benchmark using the Fetal Tissue Annotation Dataset

Abstract

It is critical to quantitatively analyse the developing human fetal brain in order to fully understand neurodevelopment in both normal fetuses and those with congenital disorders. To facilitate this analysis, automatic multi-tissue fetal brain segmentation algorithms are needed, which in turn requires open datasets of segmented fetal brains. Here we introduce a publicly available dataset of 50 manually segmented pathological and non-pathological fetal magnetic resonance brain volume reconstructions across a range of gestational ages (20 to 33 weeks) into 7 different tissue categories (external cerebrospinal fluid, grey matter, white matter, ventricles, cerebellum, deep grey matter, brainstem/spinal cord). In addition, we quantitatively evaluate the accuracy of several automatic multi-tissue segmentation algorithms of the developing human fetal brain. Four research groups participated, submitting a total of 10 algorithms, demonstrating the benefits the dataset for the development of automatic algorithms.

Synthetic magnetic resonance images for domain adaptation: Application to fetal brain tissue segmentation

“…Among the 80 subjects in the publicly available Fetal Tissue Annotation Dataset (FeTA 2.0) [11], 40 were reconstructed with an isotropic resolution of 0.8594mm (FeTA-irtk) and 40 with an isotropic resolution of 0.5mm (FeTA-mial). The SR images of this latter set was resampled to an isotropic resolution of 0.8mm and annotations were refined [16].…”

“…Superresolution (SR) reconstruction techniques take advantage of the redundancy between orthogonal low-resolution (LR) series to reconstruct an isotropic high-resolution volume of the fetal brain with reduced intensity artifacts and motion sensitivity [2,3,4,5,6]. Subsequent multi-tissue segmentation is key for advanced quantitative analysis of the fetal brain [7,8,9,10,11]. Although manual annotation is a cumbersome and tedious task prone to human error, it is a prerequisite for the training of supervised deep learning approaches that in turn enable accurate automated delineation of fetal brain tissues [8,9,10,12].…”

“…Although manual annotation is a cumbersome and tedious task prone to human error, it is a prerequisite for the training of supervised deep learning approaches that in turn enable accurate automated delineation of fetal brain tissues [8,9,10,12]. Despite two recent public releases of annotated datasets [11,13] that support the development of accurate automatic deep learning segmentation methods, their generalization across different domains remains barely explored [14]. Specifically, the training of a deep learning network relies on an optimization problem to fit a data distribution, and tends to perform poorly on a new dataset with a different distribution due to variations of MR vendors, acquisition parameters or image reconstruction techniques.…”

Automated 3D Fetal Brain Segmentation Using an Optimized Deep Learning Approach

“…This method, however, was trained on a very small cohort (n ¼ 12). Recently, Payette et al 30 evaluated several 2D segmentation methods using the Fetal Tissue Annotation and Segmentation Dataset. Of the deep learning models assessed, the combined IBBM model 30 that included information from 3 separate 2D U-Net architectures (ie, axial, coronal, and sagittal) performed the best, suggesting the superiority of using information from 3 planes.…”

“…Recently, Payette et al 30 evaluated several 2D segmentation methods using the Fetal Tissue Annotation and Segmentation Dataset. Of the deep learning models assessed, the combined IBBM model 30 that included information from 3 separate 2D U-Net architectures (ie, axial, coronal, and sagittal) performed the best, suggesting the superiority of using information from 3 planes. 3D U-Net leverages the anatomic information in 3 directions and avoids segmentation failure due to section discontinuity that may arise with 2D models.…”