Weakly supervised inference of personalized heart meshes based on echocardiography videos

Laumer, Fabian; Amrani, Mounir; Manduchi, Laura; Beuret, Ami; Rubi, Lena; Dubatovka, Alina; Matter, Christian M.; Buhmann, Joachim M.

doi:10.1016/j.media.2022.102653

Cited by 8 publications

(6 citation statements)

References 51 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…They recommended further research of 3D reconstruction for X-ray images. Laumer et al [26] proposed an automated approach which can infer a highresolution, individualized 4D (3D plus time) surface mesh of the heart structures from 2D echocardiography video data.…”

Section: Related Workmentioning

confidence: 99%

Improving the Automated Diagnosis of Breast Cancer with Mesh Reconstruction of Ultrasound Images Incorporating 3D Mesh Features and a Graph Attention Network

Chowa,

Azam,

Montaha

et al. 2024

J Digit Imaging. Inform. med.

View full text Add to dashboard Cite

This study proposes a novel approach for breast tumor classification from ultrasound images into benign and malignant by converting the region of interest (ROI) of a 2D ultrasound image into a 3D representation using the point-e system, allowing for in-depth analysis of underlying characteristics. Instead of relying solely on 2D imaging features, this method extracts 3D mesh features that describe tumor patterns more precisely. Ten informative and medically relevant mesh features are extracted and assessed with two feature selection techniques. Additionally, a feature pattern analysis has been conducted to determine the feature’s significance. A feature table with dimensions of 445 × 12 is generated and a graph is constructed, considering the rows as nodes and the relationships among the nodes as edges. The Spearman correlation coefficient method is employed to identify edges between the strongly connected nodes (with a correlation score greater than or equal to 0.7), resulting in a graph containing 56,054 edges and 445 nodes. A graph attention network (GAT) is proposed for the classification task and the model is optimized with an ablation study, resulting in the highest accuracy of 99.34%. The performance of the proposed model is compared with ten machine learning (ML) models and one-dimensional convolutional neural network where the test accuracy of these models ranges from 73 to 91%. Our novel 3D mesh-based approach, coupled with the GAT, yields promising performance for breast tumor classification, outperforming traditional models, and has the potential to reduce time and effort of radiologists providing a reliable diagnostic system.

show abstract

Section: Related Workmentioning

confidence: 99%

Improving the Automated Diagnosis of Breast Cancer with Mesh Reconstruction of Ultrasound Images Incorporating 3D Mesh Features and a Graph Attention Network

Chowa,

Azam,

Montaha

et al. 2024

J Digit Imaging. Inform. med.

View full text Add to dashboard Cite

show abstract

“…GNLs only change the vertex feature values and the feature dimensionality, but not the number of vertices nor their topology. We followed Laumer et al (2023) and introduced learnable upsampling operations into the decoder to gradually increase the number of vertices from the initial 16 in the first GNL, towards the final mesh size consisting of 4038 vertices. The vertex feature dimension of the output of each layer was 32, 16, 8 and 3, respectively.…”

Section: D To 4d Reconstructionmentioning

confidence: 99%

“…Stojanovski et al (2022) used Pix2Vox++ (Xie et al, 2020) to perform reconstruction of the 3D whole-heart anatomy from synthetic 2D standard TTE views. Previously, Laumer et al (2023) presented a generative approach using a weak supervisory signal to infer the 3D heart geometry from a single echo view. In contrast to above efforts, the present study proposes a framework that is able to reconstruct 4D heart geometry over the full cardiac cycle from real human TTE examinations using multiple views.…”

Section: Introductionmentioning

confidence: 99%

Fully automated 4D cardiac modelling from 2D echocardiography for clinical applications

Laumer,

Rubi,

Matter

et al. 2023

Preprint

Self Cite

View full text Add to dashboard Cite

Transthoracic Echocardiography (TTE) is a crucial tool for assessing cardiac morphology and function quickly and non-invasively without ionizing radiation. However, the examination is subject to intra- and inter-user variability and recordings are often limited to 2D imaging and assessments of end-diastolic and end-systolic volumes. Current 3D TTE is time-consuming and provides lower image quality than 2D TTE. We have developed a novel, fully automated machine learning-based framework to generate a 4D (3D plus time) personalized model of the left ventricular (LV) blood pool. This digital twin was generated from specific 2D echocardiographic views employing deep neural networks, pre-trained on a synthetic dataset, and fine-tuned in a self-supervised manner using a novel optimization method for cross-sectional imaging data. This 4D model enabled detailed three dimensional analyses of LV systolic and diastolic function with high temporal resolution. Validation was performed on a multi-center dataset encompassing TTE exams of 144 patients with normal TTEs (controls) and 314 patients with acute myocardial infarction (AMI). LV volume measurements in controls derived from the digital twin showed a high correlation with the values from the clinical TTE reports (Pearson correlation coefficient of 0.7). In AMI patients, these novel biomarkers showed a high predictive value for survival (area under the curve (AUC) of 0.82 for 1-year all-cause mortality) and and excellent diagnostic value for infarct localisation and quantification, bearing the potential to improve diagnostic accuracy and risk assessment. Thus, our digital twin allows for fully automated robust LV shape analyses over time including contractility, relaxation, and regional wall motion abnormalities aiming at replacing expensive 3D imaging modalities with cheaper and more versatile 2D cardiac ultrasound.

show abstract

“…Echocardiography is the most common image modality used to analyze the size and function of the heart but it has limitations in defining normal heart structure and function. The New York Heart Association, European Association of Cardiovascular Imaging, and American Society of Echocardiography have defined normality in Echocardiography by providing standardized values for normal conditions in left ventricular, right ventricular, right atrial, and left atrial [1]. The standardization process in previous research which was done by designing an echocardiogram image standardization mechanism, matching the intensity distribution histogram regarding the distribution, is considered to be the best to characterize the echocardiogram image.…”

Section: Introductionmentioning

confidence: 99%

“…The idea of histogram specification is used to build a model for standardization, namely, a prototype of the shape and scale parameter set of the considered distribution of a representative image [2]. Echocardiography is a cardiac diagnostic tool that assists physicians in analyzing the heart based on recording the heart in 2D images and incorporates low-cost portable instrumentation and rapid image acquisition without the risk of ionizing radiation [1]. The results of echocardiography, namely echocardiogram images, can provide information on the anatomy of the heart without performing surgery, including shape, pump capacity, and size of the heart to the location of damage to the heart [2].…”

Section: Introductionmentioning

confidence: 99%

Echocardiogram Image Quality Enhancement using Upsampling and Histogram Matching Methods

Permana,

Puspasari

2023

Jurnal Teknik Elektro

View full text Add to dashboard Cite

The prevalence of heart disease has been increasing in the last ten years. One of the cardiac diagnostic tools is echocardiography. Echocardiogram medical images provide essential information, including shape, size, pumping capacity, heart function abnormalities, and location of heart damage, but echocardiogram images have high noise content and poor contrast, as well as limitations due to differences in anatomy or body mass. This will affect the reading results of patient diagnosis. Therefore, image quality improvement is needed by removing noise and increasing image contrast. This research has improved image quality using a method with low mathematical complexity and a fast computational process. The method used is the Upsampling method to generate a reference image. The quality of the image produced was the Nearest Neighbor upsampling method: 2.8 dB, Bi-linear Interpolation: 2.78 dB, and Bi-cubic Interpolation: 2.73 dB. Furthermore, the image with the highest SNR value is processed with Histogram Matching to accelerate improving image quality. The Histogram Matching image increases quality by more than 50% with a SSIM value of 0.54. The required computational process to apply this method to each medical image has an average duration of 0.4 s. This result provides a higher value than several methods using linear scaling and speckle reducing.

show abstract

Weakly supervised inference of personalized heart meshes based on echocardiography videos

Cited by 8 publications

References 51 publications

Improving the Automated Diagnosis of Breast Cancer with Mesh Reconstruction of Ultrasound Images Incorporating 3D Mesh Features and a Graph Attention Network

Improving the Automated Diagnosis of Breast Cancer with Mesh Reconstruction of Ultrasound Images Incorporating 3D Mesh Features and a Graph Attention Network

Fully automated 4D cardiac modelling from 2D echocardiography for clinical applications

Echocardiogram Image Quality Enhancement using Upsampling and Histogram Matching Methods

Contact Info

Product

Resources

About