Key result
The proposed spatiotemporally-aware deep learning model (VesNetSCT++) achieved a Dice score of 0.927 for femoral artery segmentation, outperforming the baseline UNet model score of 0.775.
Why the study?
Accurate, real-time segmentation of vessel structures in ultrasound sequences aids clinical assessment, but remains challenging especially for small vessels that are difficult to visualize.
Does a deep learning model with spatiotemporal context learning improve automatic ultrasound vessel segmentation of small-scale lower-extremity arteries compared to baseline UNet models?
Does a deep learning model with spatiotemporal context learning improve automatic ultrasound vessel segmentation of small-scale lower-extremity arteries compared to baseline UNet models?
A deep learning model incorporating spatiotemporal context and dual-input B-mode/Color Doppler signals significantly improves the real-time segmentation accuracy of small lower-extremity arteries in ultrasound images.
Spatiotemporal DL models may boost ultrasound artery segmentation accuracy; leaves open prospective clinical validation before adoption.
Accurate, real-time segmentation of vessel structures in ultrasound image can aid in the measurement of lumen diameters and assessment of diseases. This, however, remains a challenging task, particularly for small vessels that are difficult to visualize. We propose to leverage rich spatiotemporal context available in ultrasound to improve segmentation small-scale lower-extremity arterial vasculature. We describe efficient deep methods that incorporate temporal, spatial, and feature-aware embeddings at multiple resolution scales while jointly utilizing from B-mode and Color Doppler signals. Evaluating on femoral and artery scans performed on healthy subjects by an expert, and comparing to consensus expert ground-truth annotations of lumen boundaries, we demonstrate real-time segmentation using the-aware models and show that they significantly outperform comparable approaches.
No takes yet. Share an insight, caveat, or question.
Jiang et al. (2021) studied Healthy subjects (lower-extremity arterial vasculature) (n=7). VesNetSCT++ (deep spatiotemporal context learning model) vs. Baseline UNet model was evaluated on Dice score for segmentation of femoral arteries. The proposed spatiotemporally-aware deep learning model (VesNetSCT++) achieved a Dice score of 0.927 for femoral artery segmentation, outperforming the baseline UNet model score of 0.775.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: