Synthetic Data for Blood Vessel Network Extraction

TL;DR

This work tackles the data bottleneck in extracting brain vessel topology from volumetric microscopy by introducing a parameterized synthetic data pipeline that generates ground-truth vessel graphs, corresponding masks, and realistic imaging artifacts. A two-stage 3D U-Net framework then performs node detection (coordinates and confidence) and edge prediction (adjacency), with a loss that combines coordinate matching, confidence calibration, and penalties for excess predictions, using Hungarian matching to align predictions with ground truth. Pretraining on large synthetic datasets, followed by fine-tuning on a small set of real patches, yields significant improvements in edge-graph reconstruction, notably increasing edge F1 from 0.496 to 0.626 when five labeled real samples are used. The findings demonstrate that synthetic data can provide crucial prior knowledge for robust vessel-network extraction in data-scarce biomedical imaging, enabling scalable analyses in stroke research.

Abstract

Blood vessel networks in the brain play a crucial role in stroke research, where understanding their topology is essential for analyzing blood flow dynamics. However, extracting detailed topological vessel network information from microscopy data remains a significant challenge, mainly due to the scarcity of labeled training data and the need for high topological accuracy. This work combines synthetic data generation with deep learning to automatically extract vessel networks as graphs from volumetric microscopy data. To combat data scarcity, we introduce a comprehensive pipeline for generating large-scale synthetic datasets that mirror the characteristics of real vessel networks. Our three-stage approach progresses from abstract graph generation through vessel mask creation to realistic medical image synthesis, incorporating biological constraints and imaging artifacts at each stage. Using this synthetic data, we develop a two-stage deep learning pipeline of 3D U-Net-based models for node detection and edge prediction. Fine-tuning on real microscopy data shows promising adaptation, improving edge prediction F1 scores from 0.496 to 0.626 by training on merely 5 manually labeled samples. These results suggest that automated vessel network extraction is becoming practically feasible, opening new possibilities for large-scale vascular analysis in stroke research.

Figures (17)

• Figure 1: The synthetic data generation is split into three phases. First, the ground truth graph structure, including all node and edge placements, is generated while considering biological constraints. Then, Vessel features, such as the outline and curves, are generated. Finally, we add noise, artifacts, and distortions during the simulated imaging to derive a synthetic image of a blood vessel network.
• Figure 2: Real Blood Vessels
• Figure 3: Synthetic Homogenous images
• Figure 6: Architecture of the node prediction model, which can detect up to a maximum of 32 nodes. The model follows a U-Net structure with an encoder path and decoder path with additional skip connections between corresponding layers. The final processing block outputs features used by the coordinate prediction head (3D node coordinates) and confidence head (confidence scores).
• Figure 7: Node Prediction