CineMyoPS: Segmenting Myocardial Pathologies from Cine Cardiac MR

Wangbin Ding, Lei Li, Junyi Qiu, Bogen Lin, Mingjing Yang, Liqin Huang, Lianming Wu, Sihan Wang, Xiahai Zhuang · 2025

Read CineMyoPS: Segmenting Myocardial Pathologies from Cine Cardiac MR by Wangbin Ding, Lei Li, Junyi Qiu, Bogen Lin, Mingjing Yang, Liqin Huang, Lianming Wu, Sihan Wang, Xiahai Zhuang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-02289.txt.

Introduction

Wangbin Ding and eight colleagues present CineMyoPS, a network that segments heart muscle damage using only cine cardiac magnetic resonance images. Heart attacks leave scars and swelling that normally require contrast-enhanced or T2-weighted scans to identify. Those extra sequences take time and may be impossible when contrast agents cannot be given.

Cine imaging is fast and contrast-free, and it shows both motion and structural abnormalities. CineMyoPS extracts motion and anatomy features, trains them jointly through a consistency loss, and aggregates information across the cardiac cycle. On a multi-center dataset it performs promisingly at pathology segmentation, motion estimation, and anatomy segmentation alike.

Catalog record

CREATORWangbin Ding, Lei Li, Junyi Qiu, Bogen Lin, Mingjing Yang, Liqin Huang, Lianming Wu, Sihan Wang, Xiahai Zhuang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE42 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-02289

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

CineMyoPS: Segmenting Myocardial Pathologies from Cine Cardiac MR Wangbin Ding, Lei Li, Junyi Qiu, Bogen Lin, Mingjing Yang, Liqin Huang, Lianming Wu, Sihan Wang, Xiahai Zhuang Abstract Myocardial infarction (MI) is a leading cause of death worldwide. Late gadolinium enhancement (LGE) and T2-weighted cardiac magnetic resonance (CMR) imaging can respectively identify scarring and edema areas, both of which are essential for MI risk stratification and prognosis assessment. Although combining complementary information from multi-sequence CMR is useful, acquiring these sequences can be time-consuming and prohibitive, e.g., due to the administration of contrast agents. Cine CMR is a rapid and contrast-free imaging technique that can visualize both motion and structural abnormalities of the myocardium induced by acute MI. Therefore, we present a new end-to-end deep neural network, referred to as CineMyoPS, to segment myocardial pathologies, i.e. , scars and edema, solely from cine CMR images. Specifically, CineMyoPS extracts both motion and anatomy features associated with MI. Given the interdependence between these features, we design a consistency loss (resembling the co-training strategy) to facilitate their joint learning. Furthermore, we propose a time-series aggregation strategy to integrate MI-related features across the cardiac cycle, thereby enhancing segmentation accuracy for myocardial pathologies. Experimental results on a multi-center dataset demonstrate that CineMyoPS achieves promising performance in myocardial pathology segmentation, motion estimation, and anatomy segmentation. I Introduction The incidence of myocardial infarction (MI) is rising worldwide [ 1 ] . MI is typically caused by blockages in the coronary arteries, leading to the formation of the area at risk (AAR) in the myocardium. In acute MI cases, the AAR typically consists of scarring and edema regions [ 2 ] . Clinicians can calculate the myocardial salvage index and assess the prognosis of MI patients based on these two regions [ 3 ] . Thus, AAR quantification is critical for clinical applications. Cardiac magnetic resonance (CMR) imaging techniques can visualize myocardial structure, motion, and pathological changes [ 4 ] , making them invaluable for quantifying the AAR in MI patients [ 5 ] . As shown in the left panel of Fig. 1 , late gadolinium enhancement (LGE) CMR visualizes scarring areas, while T2-weighted (T2w) CMR identifies edema [ 5 ] . However, these techniques face clinical limitations. For instance, the average acquisition time of multi-sequence CMR (MS-CMR), comprising T2w, first-pass perfusion, cine, and LGE, is approximately 32 \pm 8 minutes [ 6 ] , making it impractical for severe acute MI cases. In particular, acquiring the LGE sequence is time-consuming, accounting for approximately 50% of the total examination duration [ 7 ] . Moreover, the potential harm of gadolinium deposition remains under debate [ 8 ] , and there is a risk of cross-reactivity with other macrocyclic contrast agents [ 9 ] . These limitations highlight the need for alternative methods that require fewer CMR sequences and avoid the use of contrast agents [ 10 ] . MI can induce abnormal motion patterns in the myocardium, and the analysis of these patterns has been widely investigated for infarction localization. Medical imaging modalities, such as cine CMR images, are capable of capturing these motion abnormalities as well as structural malformations, as shown in the right panel of Fig. 1 . Various methods have been proposed to localize infarcts based on myocardial deformation patterns, with the aim of reducing the risks associated with contrast agents. One of the most straightforward approaches is to threshold the deformation patterns [ 11 ] . To achieve better localization accuracy, more advanced machine learning techniques have been employed, such as kernel regression [ 12 ] , dictionary learning [ 13 ] , and random forests [ 14 ] . In addition, due to the complexity of deformation patterns, strain parameters [ 15 , 16 ] and embedding techniques [ 17 , 18 ] have been incorporated to improve scar identification performance. Fig. 1: Representative LGE, T2w, and cine CMR images from an MI patient. The infarcted myocardium remains relatively stationary (indicated by red arrows), whereas the viable myocardium exhibits noticeable deformation (indicated by yellow arrows). This observation suggests that myocardial pathologies can induce abnormal motion patterns in the myocardium. ED: end-diastolic frame, ES: end-systolic frame. Recently, deep neural network (DNN) techniques have improved non-contrast quantification of myocardial pathology by leveraging cine CMR images [ 19 , 20 , 21 ] . These methods can be categorized into three groups: • Motion estimation-based methods [ 20 ] : These methods establish associations between motion fields and scars using DNN. However, they rely on an external optical flow algorithm [ 22 ] to extract wall motion from cine CMR images. • Spatio-temporal convolution network-based methods [ 19 ] : Since cine CMR images capture both spatial and temporal information of the myocardium, these methods employ convolution layers, such as ConvLSTM [ 23 ] and 3D Unet [ 24 ] , to extract underlying spatio-temporal features for scar prediction. • Synthesis-contrast-based methods [ 21 ] : These methods utilize conditional generative adversarial networks (GANs) [ 25 ] to synthesize LGE-equivalent images, from which scars can be directly extracted using existing segmentation methods, such as Unet [ 26 ] . Even though cine CMR provides a fast and contrast-free way for myocardial pathology quantification, existing cine CMR-based methods primarily focus on scar segmentation. Limited attention has been paid to edema pathology , which is critical in identifying the potential rescued heart tissue after an intervention [ 27 ] . Nevertheless, several studies have demonstrated that cine CMR sequences exhibit internal consistency in the assessment of edema in the myocardium, providing a new opportunity to jointly assess myocardial scars and edema from the cine CMR [ 28 , 29 ] . In this work, we propose an end-to-end myocardial pathology segmentation (MyoPS) network, referred to as CineMyoPS, based on DNN techniques. To the best of our knowledge, CineMyoPS is the first fully automatic network for joint scars and edema segmentation from cine CMR images. Specifically, it introduces a motion estimation module to track myocardial movements and a segmentation module to extract anatomical structures from cine CMR images. Given that both modules rely on structural features, we introduce a consistency loss to improve the feature extraction of the two modules. To further enhance MyoPS performance, we propose a time-series aggregation strategy that flexibly fuses myocardial motion, anatomy, and texture features across cardiac cycles. structure Overall, CineMyoPS is an extension of motion estimation-based methods, as it captures various features, including motion, to enhance MyoPS. Additionally, it is inspired by spatio-temporal methods, specifically leveraging time-series information to enhance MyoPS across cardiac cycles. II Method Fig. 2: The architecture of the myocardial pathology segmentation (MyoPS) network for cine CMR images, i.e. , CineMyoPS. It consists of three modules: a motion estimation module (consisting of E and D_{motion} ), an anatomy segmentation module (consisting of E and D_{anatomy} ), and a MyoPS module. The motion estimation module extracts motion from the cine CMR images, while the anatomy segmentation module predicts myocardial structures from the same images. The MyoPS module aggregates motion, anatomy, and texture features (represented by the original cine CMR images) for pathology segmentation. For details of the MyoPS module, please refer to Fig. 4 . Fig. 2 shows the network architecture of

Read the complete text