MVSegNet: A Lightweight Boundary-Aware Network for Fetal Lateral Ventricle Segmentation and Atrial Width Estimation in Prenatal Ultrasound

2026-06-08 · Source: cs.CV updates on arXiv.org · Field: Health & Wellbeing — Medical Devices & Health Technology, Clinical Care & Medical Practice, Health & Medical Research · Depth: Expert, long

Summary

MVSegNet is a lightweight encoder-decoder network designed for fetal lateral ventricle segmentation and atrial width estimation in prenatal ultrasound images, a critical task for assessing fetal ventriculomegaly. The model addresses challenges like acoustic shadowing and low contrast by integrating multi-scale feature extraction and boundary-aware refinement. Evaluated on 584 expert-annotated transventricular ultrasound frames, MVSegNet achieved a Dice score of 80.79%, IoU of 68.47%, a 95th-percentile Hausdorff distance of 4.07 mm, and an atrial width mean absolute error of 3.40 mm. With 2.31 million parameters, it operates at 165.6 frames per second on an NVIDIA T4 GPU, demonstrating superior performance over six baseline models in boundary and measurement accuracy while maintaining high computational efficiency.

Key takeaway

For Machine Learning Engineers developing automated prenatal ultrasound analysis tools, MVSegNet demonstrates that a lightweight, boundary-aware network can achieve superior segmentation and measurement accuracy with high computational efficiency. You should consider task-specific architectures that integrate multi-scale feature extraction and boundary refinement, especially when hardware resources are limited. However, ensure validation on diverse external datasets and align width estimation methods with clinical protocols before deployment.

Key insights

A lightweight, boundary-aware network with multi-scale features and gestational age conditioning excels in fetal ventricle segmentation.

Principles

• Boundary-aware refinement improves contour quality.
• Gestational age conditioning stabilizes boundary predictions.
• Compact, task-specific models can outperform larger general architectures.

Method

MVSegNet combines a MobileNetV3-Small encoder, FiLM for gestational age conditioning, Multi-Scale Ventricle Attention Modules, attention gates, and an Adaptive Boundary Refinement Block for precise, efficient segmentation.

In practice

• Employ MobileNetV3-Small for efficient backbones.
• Condition models with gestational age for stability.
• Use attention gates to suppress irrelevant features.

Topics

• Fetal Ventricle Segmentation
• Prenatal Ultrasound
• MVSegNet Architecture
• Lightweight Deep Learning
• Boundary-aware Segmentation
• Atrial Width Estimation

Best for: Computer Vision Engineer, AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• MVSegNet: A Lightweight Boundary-Aware Network for Fetal Lateral Ventricle Segmentation and Atrial Width Estimation in Prenatal Ultrasound — MVSegNet accurately segments fetal lateral ventricles for atrial width estimation despite ultrasound challenges.
• ResNet-34 with Lightweight Decoder for Accurate and Efficient Segmentation of Fetal Brain MRI — A ResNet-34 encoder with a lightweight MLP decoder achieves accurate and efficient fetal brain MRI segmentation, outperforming baselines.
• GloResNet: A lightweight 3D CNN with global topological features for preterm brain injury prediction — Lightweight 3D CNN with global topological features predicts preterm brain injury effectively.
• Structure-Augmented Standard Plane Detection with Temporal Aggregation in Blind-Sweep Fetal Ultrasound — Structure-augmented temporal aggregation improves fetal abdomen plane detection in blind-sweep ultrasound for reliable biometry.
• Multitasking Embedding for Embryo Blastocyst Grading Prediction (MEmEBG) — Automated multitask embedding improves blastocyst grading consistency and reduces subjectivity in IVF.
• Vision-Language Model-Guided Deep Unrolling Enables Personalized, Fast MRI — PASS uses a VLM to guide personalized MRI sampling and reconstruction for faster, anomaly-aware imaging.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.