A benchmark for 2D foetal brain ultrasound analysis

TL;DR

A set of 104 2D foetal brain ultrasound images acquired during the 20th week of gestation that have been co-registered to a common space from a rough skull segmentation and annotated to highlight landmark points from structures of interest to analyse brain development are presented.

Abstract

Brain development involves a sequence of structural changes from early stages of the embryo until several months after birth. Currently, ultrasound is the established technique for screening due to its ability to acquire dynamic images in real-time without radiation and to its cost-efficiency. However, identifying abnormalities remains challenging due to the difficulty in interpreting foetal brain images. In this work we present a set of 104 2D foetal brain ultrasound images acquired during the 20th week of gestation that have been co-registered to a common space from a rough skull segmentation. The images are provided both on the original space and template space centred on the ellipses of all the subjects. Furthermore, the images have been annotated to highlight landmark points from structures of interest to analyse brain development. Both the final atlas template with probabilistic maps and the original images can be used to develop new segmentation techniques, test registration approaches for foetal brain ultrasound, extend our work to longitudinal datasets and to detect anomalies in new images.

Figures (7)

• Figure 1: Methodology: 1) The skull is automatically segmented using a UNet network (top-left). 2) An ellipse is fitted to the skull segmentation (top-right) 3) to estimate an affine transformation to a reference image (bottom-right). The axes of the fitted ellipse are warped to the image coordinate axes and are re-scaled to fit the ellipse in the reference image (bottom-left).
• Figure 2: Qualitative comparison between a low quality image where structures are not clearly visible and and image from the dataset.
• Figure 3: Quantitative results for all the methods on each structure (Hausdorff distance, $d_H$, lower values indicate better registration). The upper part of each boxplot figure indicates the results of pairwise statistical Wilcoxon tests: (ns: 5.00e-02 $<$ p $\leq$ 1.00e+00, *: 1.00e-02 $<$ p $\leq$ 5.00e-02, **: 1.00e-03 $<$ p $\leq$ 1.00e-02, ***: 1.00e-04 $<$ p $\leq$ 1.00e-03, ****: p $\leq$ 1.00e-04).
• Figure 4: Quantitative results for all the methods on each structure with multiple points (polygon DSC, higher values indicate better registration). The upper part of each boxplot figure indicates the results of pairwise statistical Wilcoxon tests: (ns: 5.00e-02 $<$ p $\leq$ 1.00e+00, *: 1.00e-02 $<$ p $\leq$ 5.00e-02, **: 1.00e-03 $<$ p $\leq$ 1.00e-02, ***: 1.00e-04 $<$ p $\leq$ 1.00e-03, ****: p $\leq$ 1.00e-04).
• Figure 5: Quantitative results for all the methods according to the SSIM metric (higher values indicate better registration). The upper part of each boxplot figure indicates the results of pairwise statistical Wilcoxon tests: (ns: 5.00e-02 $<$ p $\leq$ 1.00e+00, *: 1.00e-02 $<$ p $\leq$ 5.00e-02, **: 1.00e-03 $<$ p $\leq$ 1.00e-02, ***: 1.00e-04 $<$ p $\leq$ 1.00e-03, ****: p $\leq$ 1.00e-04).