Measuring proximity to standard planes during fetal brain ultrasound scanning

TL;DR

This work tackles the challenge of guiding ultrasound scans to standard planes in fetal brain imaging by fusing semi-supervised brain segmentation with a 6D plane pose regression that yields proximity to a target SP. The SS-Seg+Class model uses labeled SP slices and unlabeled 3D US slices to robustly detect and segment the fetal brain, while a ResNet-18-based regressor estimates the plane pose $\theta_{Pred}$ and computes proximity to the target plane $\theta_{GT}$ via translation distance and the angle between plane normals; the method excludes in-plane rotation to focus on salient plane changes. Key contributions include (i) a semi-supervised segmentation/classification framework with inter-volume consistency losses, (ii) a pose regression module that benefits from brain masks, and (iii) a proximity metric that can guide sonographers during scanning. Validation on real fetal scans across operators of varying expertise demonstrates that the predicted distances to the tv SP align with expert quality assessments, suggesting practical utility for reducing operator workload and improving SP navigation. The study identifies future work on handling zoom and off-center anatomies, expanding training data, incorporating temporal dynamics, and achieving real-time deployment to facilitate clinical adoption, with potential applicability to other SPs and fetal regions.

Abstract

This paper introduces a novel pipeline designed to bring ultrasound (US) plane pose estimation closer to clinical use for more effective navigation to the standard planes (SPs) in the fetal brain. We propose a semi-supervised segmentation model utilizing both labeled SPs and unlabeled 3D US volume slices. Our model enables reliable segmentation across a diverse set of fetal brain images. Furthermore, the model incorporates a classification mechanism to identify the fetal brain precisely. Our model not only filters out frames lacking the brain but also generates masks for those containing it, enhancing the relevance of plane pose regression in clinical settings. We focus on fetal brain navigation from 2D ultrasound (US) video analysis and combine this model with a US plane pose regression network to provide sensorless proximity detection to SPs and non-SPs planes; we emphasize the importance of proximity detection to SPs for guiding sonographers, offering a substantial advantage over traditional methods by allowing earlier and more precise adjustments during scanning. We demonstrate the practical applicability of our approach through validation on real fetal scan videos obtained from sonographers of varying expertise levels. Our findings demonstrate the potential of our approach to complement existing fetal US technologies and advance prenatal diagnostic practices.

Figures (4)

• Figure 1: Comparison of the proposed approach with SPs classification approaches.
• Figure 2: Components of our semi-supervised learning model. Labeled (brain and non-brain) and unlabeled brain us images are fed into a UNet. The encoder extracts the features from the images; the classification branch takes the encoder output and classifies the images as containing brain or not, whereas the decoder predicts the masks for both labeled and unlabeled brain images.
• Figure 3: Inference pipeline on frames from us videos. The frames are classified as containing or not the brain; if the brain is detected, the mask is generated, and the frame with the mask applied is fed into a pose regression network to regress the 6d pose of the frame and compute the distance to the sp.
• Figure 4: (a) Distances to TV SP and SonoNet's predictions. (b) Obstetrician assessment of the tv sp per operator, with SP re-acquisitions for operators 4 and 5 due to required structures not being visible and the plane being too oblique.