Spatiotemporal Distillation via Recurrent Bottlenecks for Aortic Tracking

Dexter Wen Jie Teo; Nairouz Shehata; Herve Lombaert

eess.IVAug 24, 2026

Spatiotemporal Distillation via Recurrent Bottlenecks for Aortic Tracking

Authors: Dexter Wen Jie Teo, Nairouz Shehata, Herve Lombaert

Organizations: College of Computing and Data Science, Nanyang Technological University, Singapore · Department of Computer and Software Engineering, Polytechnique Montreal, Canada

Abstract

Cardiac cine-MRI serves as a direct visual indicator of cardiovascular hemodynamics by capturing the continuous wall motion of the aorta. Quantifying these dynamic structural changes across the cardiac cycle is essential for measuring aortic distensibility, a primary marker of arterial stiffness. However, standard 2D segmentation networks focus on each frame independently. Consequently, when rapid systolic flow temporarily obscures the aorta's boundaries, this lack of continuous context results in frame-to-frame tracking dropouts and boundary inconsistencies. Spatiotemporal ( $2\text{D}+t$ ) networks can enforce temporal consistency across the sequence but suffer from a scarcity of expert annotations. To address this, we present a semi-supervised spatiotemporal ( $2\text{D}$ to $2\text{D}+t$ ) knowledge distillation framework exploiting the cardiac cycle. The framework distills a spatial teacher's expertise into a spatiotemporal student network by executing a dynamic latent interception, pairing a recurrent spatiotemporal bottleneck with a residual spatial bypass. Our model selection strategy applies a baseline validation threshold ( $\text{DSC} \ge 0.50$ ) prior to selecting the epoch that maximizes anatomical consistency. This strategy enables the spatiotemporal student model to achieve superior surface tracking accuracy ( $\text{NSD@1mm} = 92.3\% \pm 0.2\%$ ) and high structural reliability ( $\text{Frac}_{2\text{CC}} = 99.2\% \pm 0.6\%$ ), reducing population-wide structural anomalies by over 56% compared to a 2D nnU-Net baseline.

Figures & tables

Figure 1: Overview of the spatiotemporal student architecture. A dynamic hook routes deep features through a bidirectional ConvLSTM to capture sequence context. This temporal stream merges with a residual spatial bypass ( $\oplus$ ) to preserve tracking continuity and enforce anatomical plausibility ( $N=2$ ) across the cardiac cycle.

Parameter	Symbol / Variable	Configured Value
Teacher Validation Floor	$\tau_{\text{dice}}$	$0.50$ (DSC gatekeeper)
Pseudo-Label Binarization Threshold	$\tau_{\text{bin}}$	$0.50$
Teacher Focal Loss Weight	$\lambda_{\text{focal}}$	$0.50$
Temporal Residual Scaling Factor	$\alpha$	$0.10$
Compressed Bottleneck Width	$C_{\text{temp}}$	$128$ channels
Temporal Smoothness Regularizer	$\beta$	$0.10$

Table 1: Summary of framework hyperparameters and operational thresholds.

Architectural Configuration	$\mathbf{\text{Frac}_{2\text{CC}}}$ (%) $\uparrow$	NSD@1mm (%) $\uparrow$	HD95 (mm) $\downarrow$	ASSD (mm) $\downarrow$	DSC (%) $\uparrow$
Static 2D Teacher (Fully-Supervised Base)	$93.0\pm 4.4$	$86.6\pm 5.7$	$2.88\pm 0.78$	$0.58\pm 0.06$	$88.8\pm 2.4$
Static Student	$98.7\pm 0.7$	$90.7\pm 0.4$	$2.18\pm 0.09$	$0.45\pm 0.03$	$90.9\pm 0.3$
Spatiotemporal Student (Ours)	$\mathbf{99.2\pm 0.6}$	$\mathbf{92.3\pm 0.2}$	$2.05\pm 0.07$	$0.41\pm 0.02$	$90.9\pm 0.2$
2D nnU-Net Baseline	$97.0\pm 1.6$	$89.9\pm 1.4$	$\mathbf{1.23\pm 0.36}$	$\mathbf{0.18\pm 0.08}$	$\mathbf{92.9\pm 1.1}$

Table 2: Quantitative evaluation and distillation ablation summary across the independent 25-patient test set (948 total frames). Values represent the mean

\pm

standard deviation across the 5 cross-validation folds. Our model enforces topology.

Figure 2: Qualitative tracking comparison across consecutive frames ( $t-1$ , $t$ , $t+1$ ). 2D nnU-Net baseline mistakes a nearby bright vessel for the aorta during a phase of fading visibility ( $N=3$ ). Our spatiotemporal student leverages sequence context to look past this fading effect, maintaining correct anatomical structure ( $N=2$ ).

Metric / Configuration	Fold 1	Fold 2	Fold 3	Fold 4	Fold 5	Mean Summary
Total Evaluated Frames	17,539	17,539	17,539	17,539	17,539	17,539 Total Frames
Static 2D A-UNet Teacher [ 8 ]
– Structural Anomalies ( $N\neq 2$ )	1,333	2,843	1,243	1,626	1,362	$1,681.4\pm 663.0$
– Low-Confidence Frames	0	837	0	0	0	$167.4\pm 374.3$
2D nnU-Net Baseline
– Structural Anomalies ( $N\neq 2$ )	1,169	1,115	1,089	1,109	702	$1,036.8\pm 190.2$

Table 3: Large-scale inference consistency and structural anomaly summary across all 5 cross-validation folds. Each entry evaluates fold-specific model performance over the entire unannotated cohort (

17,539

total frames per run).

$\beta$	Anomalies ( $N\neq 2$ ) $\downarrow$	Low-Conf. Frames $\downarrow$	$\text{Frac}_{2\text{CC}}$ (%) $\uparrow$
$0.00$ (Static Base)	$492.2\pm 49.9$	$13.2\pm 8.6$	$98.7\%\pm 0.7\%$
$0.05$	$461.2\pm 60.1$	$0.0\pm 0.0$	$98.9\%\pm 0.6\%$
$0.08$	$451.8\pm 64.2$	$0.0\pm 0.0$	$99.0\%\pm 0.7\%$
$\mathbf{0.10}$	$\mathbf{451.2\pm 67.8}$	$\mathbf{0.0\pm 0.0}$	$\mathbf{99.2\%\pm 0.6\%}$
$0.15$	$476.0\pm 59.2$	$0.0\pm 0.0$	$99.1\%\pm 0.8\%$
$0.50$	$478.6\pm 14.7$	$0.0\pm 0.0$	$98.6\%\pm 0.4\%$

Table 4: Population-wide hyperparameter sensitivity analysis varying the temporal smoothness weight

\beta

across all

17,539

unannotated frames.

Medical Image Segmentation

Medical image segmentation delineates organs, lesions, and other structures in scans to support measurement, diagnosis, treatment planning, and research.

See all 786 papers

Sep 3, 2026

Segmentation of the aorta in 4D flow MRI using 4D convolutional kernels and learning from sparse annotations

Automated aortic segmentation in 4D flow MRI is essential for reproducible hemodynamic assessment but is limited by scarce dense annotations and high computational demands. We developed a fully automated 4D (3D+time) U-Net for segmenting the ascending aorta, arch, and proximal descending aorta, using a parameter-efficient hybrid 4D kernel to capture temporal context and sparse 4D labels derived from existing 2D expert contours and centerlines, thereby avoiding the need for dense 4D annotations. Training comprised 268 scans from 8 centers and 2 vendors, with evaluation on an internal test set (32 scans) and an external post-contrast set (30 scans; different site, protocol, and annotator), compared against frame-wise 3D networks and two semi-automatic references. Against time-resolved annotations, the 4D U-Net achieved Dice scores of 0.927 (internal) and 0.911 (external), versus 0.919/0.847 for the 3D U-Net, 0.893 for static PC-MRA, and 0.808 for registration-based propagation; differences were small in systole but pronounced in diastole. Agreement with expert contours for peak velocity, net flow, axial and circumferential wall shear stress, and diameters was excellent (ICC >=0.954 internal, >=0.980 external), while semi-automatic references performed worse. The method thus provides reproducible, time-resolved aortic segmentation for automated hemodynamic analysis and generalizes across multicenter, multivendor, and independent post-contrast data. The model is publicly available.

Hinrich Rahlfs, Julio Garcia, Chiara Manini +13

Medical Image Segmentation Magnetic Resonance Imaging U-Net

Ask PDF

Jun 30, 2026

Self-Supervised Temporal Regularization for Landmark-Based Cardiac Segmentation with Automatic AHA Regional Mapping

Graph-based cardiac segmentation with implicit anatomical correspondences provides topological guarantees and population-level analysis capabilities, but models trained on independent frames of image sequences exhibit temporal discontinuities that affect reliable clinical measurements, particularly in cardiac ultrasound. In this work, we introduce self-supervised temporal regularization as a post-training refinement stage that exploits the temporal coherence in image sequences to enforce consistent cardiac segmentation and motion estimation over time, without requiring per-frame annotations. By penalizing velocity and acceleration discontinuities across consecutive frames, our method achieves temporally consistent segmentations while maintaining the learned anatomical correspondences. We further leverage these correspondences to automatically map landmarks to the AHA 17-segment clinical standard, enabling standardized regional assessment and detection of pathological myocardial motion patterns. Validation on CAMUS dataset demonstrates the clinical utility of combining temporal consistency with automatic regional mapping. The code is publicly available at https://github.com/david-montalvoo/MaskHybridGNet-TempReg

David Montalvo-García, Nicolás Gaggion, María J. Ledesma-Carbayo +1

Biomedical Image Technologies, ETSI Telecomunicación, Universidad Politécnica de Madrid, Madrid, Spain · Centro de Investigación Biomédica en Red de Bioingeniería, Biomateriales y Nanomedicina (CIBER-BBN), Instituto de Salud Carlos III, Madrid, Spain · APOLO Biotech, Ciudad Autónoma de Buenos Aires, Argentina +2

Ultrasound Image Segmentation Self-Supervised Learning Medical Image Analysis Medical Image Segmentation

Ask PDF

Mar 27, 2026

OSA: Echocardiography Video Segmentation via Orthogonalized State Update and Anatomical Prior-aware Feature Enhancement

Accurate and temporally consistent segmentation of the left ventricle from echocardiography videos is essential for estimating the ejection fraction and assessing cardiac function. However, modeling spatiotemporal dynamics remains difficult due to severe speckle noise and rapid non-rigid deformations. Existing linear recurrent models offer efficient in-context associative recall for temporal tracking, but rely on unconstrained state updates, which cause progressive singular value decay in the state matrix, a phenomenon known as rank collapse, resulting in anatomical details being overwhelmed by noise. To address this, we propose OSA, a framework that constrains the state evolution on the Stiefel manifold. We introduce the Orthogonalized State Update (OSU) mechanism, which formulates the memory evolution as Euclidean projected gradient descent on the Stiefel manifold to prevent rank collapse and maintain stable temporal transitions. Furthermore, an Anatomical Prior-aware Feature Enhancement module explicitly separates anatomical structures from speckle noise through a physics-driven process, providing the temporal tracker with noise-resilient structural cues. Comprehensive experiments on the CAMUS and EchoNet-Dynamic datasets show that OSA achieves state-of-the-art segmentation accuracy and temporal stability, while maintaining real-time inference efficiency for clinical deployment. Codes are available at https://github.com/wangrui2025/OSA.

Rui Wang, Huisi Wu, Jing Qin

College of Computer Science and Software Engineering, Shenzhen University · Centre for Smart Health, School of Nursing, The Hong Kong Polytechnic University

Ultrasound Image Segmentation Echocardiography

Ask PDF

Abstract

Medical Image Segmentation

Explore similar work

Segmentation of the aorta in 4D flow MRI using 4D convolutional kernels and learning from sparse annotations

Self-Supervised Temporal Regularization for Landmark-Based Cardiac Segmentation with Automatic AHA Regional Mapping

OSA: Echocardiography Video Segmentation via Orthogonalized State Update and Anatomical Prior-aware Feature Enhancement