A feasibility study based on a 2.5D U-Net architecture to detect GME in space-time connected data, resulting in improved detection of moving GMEs against the background with respect to classical spot detection algorithms and 2D U-Net, yet retaining real-time execution speed with respect to more complex deep-learning architectures is proposed.
Abstract
Gaseous microemboli (GME) represent a common complication of cardiac structural interventions across both surgical and transcatheter approaches. Intraoperative transesophageal echocardiography (TEE) represents a convenient methodology to monitor and visualize the presence of circulating GME. However, their detection and quantification are far from trivial due to operator-dependent view, high velocity, and objects with similar structure in the background. Here, we propose a feasibility study based on a 2.5D U-Net architecture to detect GME in space-time connected data. We applied and tested such an architecture on a pilot dataset of eight TEE recordings ($60$ fps, $600\times 800$ pixels) from eight different patients undergoing cardiac surgery, resulting in improved detection of moving GMEs against the background with respect to classical spot detection algorithms and 2D U-Net, yet retaining real-time execution speed with respect to more complex deep-learning architectures. Under leave-one-patient-out cross-validation, the selected model achieved strong detection performance under a three-pixel radius-tolerant grace-zone evaluation, with a precision of 92.55\% and recall of 80.54\%, corresponding to radius-tolerant Intersection over Union (IoU) and Dice coefficients of 73.95\% and 84.13\%, respectively. Complementarily, strict pixel-based segmentation metrics were also computed, yielding an IoU of 41.74\% and a Dice coefficient of 57.98\%. The selected model achieved an average inference time of $0.12 s$ per batch on the tested hardware. To assess specificity on unseen data, we additionally evaluated the model on an external GME-negative TEE dataset, where it produced predominantly empty or near-empty masks, indicating a low rate of spurious detections. These results support the technical feasibility of real-time GME segmentation.
BACKGROUND
Interventional cardiovascular MRI (iCMR) uses rapid high-contrast images to guide invasive procedures, such as right-heart catheterization (RHC), an important diagnostic tool for characterizing pulmonary hypertension and heart failure. Current iCMR uses interleaved multi-planar 2D acquisitions (1-3 planes) with balanced steady-state free precession (bSSFP) contrast. A common challenge is tracking devices as they move out of plane and the operational burden of updating views in real-time. We propose a volumetric method that images the heart and great vessels in their entirety, and offers flexible real-time re-slicing, volumetric 3D visualization, and the potential for improved device monitoring.
METHODS
Volumetric real-time MRI (3DRT) was implemented at 0.55T using a stack-of-spirals bSSFP acquisition paired with a low-latency online spatiotemporally constrained image reconstruction, to achieve 3.5 mm3 isotropic spatial resolution and ~500 ms/volume temporal resolution. Numerical simulations were performed to evaluate tracking capability of gadolinium balloons as a function of device speed, and to verify image quality. Six healthy volunteers were imaged without catheterization to test performance on human subjects. Six naïve pigs were imaged during RHC, with gadolinium-filled polymer balloon wedge end-hole catheters.
RESULTS
Simulation experiments demonstrated that 3DRT imaging with low-latency spatiotemporally constrained reconstruction can resolve moving catheters at ~2.0cm/s. 3DRT with bSSFP contrast was successfully demonstrated for right-heart catheterization procedures in porcine models, providing adequate blood-myocardium contrast, and balloon-blood contrast to track the catheter tip. 3DRT images provided adequate image quality and resolved respiratory motion in healthy volunteers. Volumes were re-sliced live and re-sliced planes could be flexibly manipulated during the scan in real-time before the next volume was acquired.
CONCLUSION
Volumetric real-time MRI is feasible for the guidance of invasive procedures such as right-heart catheterization at 0.55T. 3DRT imaging has been demonstrated in model pig experiments with ~500 ms temporal resolution and 3.5 mm3 isotropic spatial resolution, providing improved balloon tracking.
Prakash Kumar, R. Ramasawmy, A. Javed et al.· Journal of Cardiovascular Ma...· 0 citations
Peripheral Artery Disease (PAD) is a serious and potentially limb and life-threatening condition that demands a timely and accurate diagnosis to prevent severe complications. Traditional diagnostic methods are often slow and error prone. Although PAD localization remains relatively less explored using machine learning, the development of efficient and deployable models for real-time arterial blockage identification demands more investigation. In this paper, we propose PAD-Net, a novel end-to-end object detection framework specifically optimized for the identification and localization of arterial blockage. By integrating Light-Weight Feature Adapter (LFA) and Transformer block into You Only Look Once (YOLO) architecture, our model enhances the possibility of capturing multi-scale vascular structures often obscured in complex medical backgrounds/conditions. The proposed model was evaluated on a novel dataset of lower-extremity CT angiograms collected from the National Institute of Cardiovascular Disease, Bangladesh. Experimental results demonstrate that the model achieves a Mean Average Precision of 0.87 (mAP@0.5) and 0.423 (mAP@0.5:0.95), achieving competitive performance compared with state-of-theart YOLO and RT-DETR (Real Time Detection Transformer) models. At the same time, it maintains real-time speed to make predictions. The proposed PAD-Net gives a strong and scalable solution for AI-based PAD diagnosis. Furthermore, to validate real-world applicability, the model is deployed on a web-based Streamlit Community Cloud. The deployment of the proposed model highlights the effectiveness of reliable PAD detection and its integration into clinical workflows.
Md Raisul Islam, M. Zaman, S. Ahmmed et al.· International Conference on...· 0 citations
Background/Objectives: Cardiovascular diseases remain a leading cause of death worldwide, and early diagnosis is essential. Phonocardiogram (PCG) signals provide a non-invasive and inexpensive means of cardiac assessment. However, accurate interpretation of PCG signals requires clinical expertise. In this paper, we propose TriFusion-PCGNet, a hybrid multi-branch deep learning framework for automated heart sound classification. Methods: The proposed method was evaluated on the PhysioNet/Computing in Cardiology Challenge 2016 dataset. Three complementary PCG representations—the raw signal, band-pass filtered signal, and Hilbert envelope signal—were processed through dedicated feature extraction branches using residual convolutional networks, dilated residual convolutions, and a BiGRU with a temporal attention mechanism. The extracted features were fused and classified using fully connected layers. Performance was evaluated using the original train–validation split and stratified 10-fold cross-validation. Results: Under a strictly separated record-level 10-fold cross-validation protocol, TriFusion-PCGNet achieved 90.87% mean accuracy and 95.34% AUC, with outer test folds isolated from training and model selection. Subject-level analysis showed substantially lower performance, emphasizing the importance of subject-disjoint evaluation for patient-level generalization. Conclusions: TriFusion-PCGNet integrates complementary PCG representations for automated normal/abnormal heart sound classification. However, the substantially lower subject-disjoint performance indicates that the record-level results should not be interpreted as evidence of patient-independent generalization, and further validation on independent subject-annotated datasets is required.
Tolga Çakmak, Ç. Mirzaoğlu, Melike Demir et al.· Life· 0 citations
OBJECTIVE
Deep learning-based automated analysis of transgastric short-axis view (TSV) transesophageal echocardiography (TEE) remains under-explored. In this study, we propose a deep learning-based method for fully automated left ventricular segmentation and ejection fraction (EF) prediction in TSV TEE videos.
METHODS
We built upon the U-Net network and proposed an Echo Efficient U-Net (Echo-EU-Net) segmentation model by replacing the original standard convolutions with depth-wise separable convolutions and by introducing the Multi-Efficient Channel Attention (MECA) and Enhanced Atrous Spatial Pyramid Pooling (EASPP) modules. We also incorporated automatic cardiac phase tracking and EF calculation. Experiments were performed on a TSV TEE dataset containing 694 videos from 451 patients, with expert manual segmentations and manual EF measurements as the reference standard.
RESULTS
The proposed Echo-EU-Net, with an average Dice similarity coefficient of 92.91% and a Jaccard similarity coefficient of 87.23%, outperformed U-Net and its variants for left ventricular segmentation in TSV TEE, particularly in challenging cases. The model parameter size of Echo-EU-Net was 1.30 million, compared with 7.79 million for U-Net. The proposed EF prediction method had a satisfying agreement with the manual EF measurements (Pearson's r=0.84), with a mean absolute error of 6.44%. An ablation study demonstrated the effectiveness of the MECA and EASPP modules.
CONCLUSION
The feasibility of the proposed Echo-EU-Net-based method in automatically segmenting the left ventricle and measuring EF in TSV TEE has been demonstrated. The findings of this study may shed light on lightweight deep learning-based fully automated left ventricular segmentation and EF quantification in TSV TEE.
Yijun Ling, Yuan Tian, W. Qin et al.· Ultrasound in Medicine and B...· 0 citations
BACKGROUND
Accurate assessment of left ventricular outflow tract (LVOT) gradients is critical for hypertrophic cardiomyopathy management, yet Doppler-based measurements are technically demanding and require expertise. The objective of this work was to develop a multi-view deep learning model capable of classifying LVOT obstruction (>20 mm Hg) using routine 2-dimensional echocardiographic windows without reliance on Doppler imaging.
METHODS
We trained and externally validated a cross-attention-based video-to-video fusion framework that integrated EchoPrime-derived video representations from 3 standard transthoracic echocardiographic views to classify LVOT gradients.
RESULTS
Training was performed on a derivation cohort (N=1833) from a tertiary care system in the United States, with model performance evaluated on an internally held-out test set (N=275) and a Korean external validation cohort (N=46). Single-view baselines showed limited discrimination (external area under the receiver operating curves, 0.47-0.70). Conversely, the domain-specific foundational model (EchoPrime) achieved superior single-view performance (area under the receiver operating curves, 0.75-0.80 internal; 0.79-0.83 external), highlighting the importance of echo-specific pretraining and temporal modeling. The proposed multi-view fusion further enhanced predictive performance, with the late fusion model reaching an area under the receiver operating curve of 0.84 on the external cohort with significant population-shift.
CONCLUSIONS
These results suggest LVOT physiology is encoded in routine 2-dimensional imaging and can be leveraged for clinically relevant gradient classification without Doppler input. The proposed artificial intelligence-guided strategy demonstrates substantial cost savings compared with the screen-all approach. By integrating complementary spatial-temporal information across multiple views, our approach generalizes robustly across populations and may enable real-time decision support, extend LVOT assessment to portable or resource-limited settings, and complement Doppler-based evaluation for longitudinal hypertrophic cardiomyopathy management.
O. Crystal, J. Farina, I. Scalia et al.· Circulation Cardiovascular I...· 0 citations
The large majority of patients with congenital heart disease (CHD) survive into adulthood in the modern era of surgical advancements. However, patients require life-long follow-up and non-invasive imaging is crucial for diagnostic and therapeutic purposes. Cardiovascular magnetic resonance (CMR) plays a major role in patients with CHD as it is a non-invasive and ionizing radiation free imaging tool which overcomes the technical limitations of echocardiography. 4D flow is an evolving technique for flow quantification allowing improved CMR exam quality and reduced scan time. However, comprehensive postprocessing of hemodynamics in CHD may be demanding. This document addresses the needs for an "ideal" post-processing software for 4D flow in patients with CHD. It provides information on general requirements, velocity encoding, offset and aliasing correction, visualization, segmentation and reconstruction as well as output data. This document aims to support stakeholders in the field to develop solutions for improved 4D flow post-processing and visualization in CHD patients and thus improve the accuracy and efficiency of patient care.
Julio Garcia, Julia Geiger, Adam B. Christopher et al.· Journal of Cardiovascular Ma...· 0 citations
