Segmentation of left ventricular (LV) endocardium IntroductionCardiac left ventricular plays a crucial role in the cardiac functions and diagnosis of cardiac disease. Hence, LV volume estimation has attracted much research attention, and segmentation of left ventricular endocardium from 3D echocardiography (3DE) has become a hot topic. Although this is a challenging task, which has to handle many problems inherent in ultrasound imaging, such as low signal-noise ratio, edge dropout and artifacts, researchers have proposed many methods to segment the left ventricular endocardium from 3DE. The typical proposed method can be classified as: deformable models, statistical models and classification methods.Deformable models are the most common methods of the segmentation of LV endocardium, which translate the problem to an optimization of energy function which is defined by the geometrical constraints.[1] proposed a coupled deformable model which used the speckle statistics information and volume information to form and evolve two surfaces to segment the myocardium. This method can make an automatic segmentation of the full myocardium. Due to intensity distribution of ultrasound tend to Gaussian approximately, [2] combined regionand edge-based level to acquire coarse shape, then adopted this coarse shape as the initial boundary and additional constraint to deform to segment echocardiography.[3] made a motion prior energy as the new constraint of level-set to track the whole myocardium, and used anatomical and image information to adjust hyperparameters. Although many researches have shown deformable models are successful, the models depend much on the initialization and image conditions.Statistical models, such as active shape model and active appearance model, are based on large labeled information from experts.[4] used a 3D active shape model whose parameters were updated by an extended Kalman filter to segment 3D cardiac ultrasound and the results were promising.[5] proposed a model driven method combining 3D Active Shape Model with local appearance models to segment 3D left ventricular and quantify the left ventricular function. The experiments results suggested the proposed method can achieve acceptable accuracy for the segmentation of fast rotating ultrasound.[6] developed Multiview, multiframe and landmark, and dynamic programming constrained active appearance models. Local edge detector was incorporated into these models. Comparing the experiments results, dynamic programming constrained model is better than the landmark constrained model and active appearance model. Due to depending on a large annotated dataset, initialization and assumption of shape and appearance [7], statistical models are limited for the segmentation of LV.Classification methods train classifier which is based on different features to segment LV. [8] proposed an automated method and system which used knowledgebased probabilistic model and marginal space learning and can be used to detect standard multiplanar reformatted planes.[9]...
Estimation of left ventricular (LV) volumes from 3D echocardiography (3DE) is a popular clinical approach
State-of-the-art video object detection methods maintain a memory structure, either a sliding window or a memory queue, to enhance the current frame using attention mechanisms. However, we argue that these memory structures are not efficient or sufficient because of two implied operations: (1) concatenating all features in memory for enhancement, leading to a heavy computational cost; (2) frame-wise memory updating, preventing the memory from capturing more temporal information. In this paper, we propose a multi-level aggregation architecture via memory bank called MAMBA. Specifically, our memory bank employs two novel operations to eliminate disadvantages of existing methods: (1) light-weight key-set construction which can significantly reduce the computational cost; (2) fine-grained feature-wise updating strategy which enables our method to utilize knowledge from the whole video. To better enhance features from complementary levels, i.e., feature maps and proposals, we further propose a generalized enhancement operation (GEO) to aggregate multi-level features in a unified manner. We conduct extensive evaluations on the challenging ImageNetVID dataset. Compared with existing state-of-the-art methods, our method achieves superior performance in terms of both speed and accuracy. More remarkably, MAMBA achieves mAP of 83.7%/84.6% at 12.6/9.1 FPS with ResNet-101.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.