EchoONE: Segmenting Multiple echocardiography Planes in One Model

Jiongtong Hu,Wei Zhuo,Jun Cheng,Yingying Liu,Wufeng Xue,Dong Ni

2024-12-04

Abstract:In clinical practice of echocardiography examinations, multiple planes containing the heart structures of different view are usually required in screening, diagnosis and treatment of cardiac disease. AI models for echocardiography have to be tailored for each specific plane due to the dramatic structure differences, thus resulting in repetition development and extra complexity. Effective solution for such a multi-plane segmentation (MPS) problem is highly demanded for medical images, yet has not been well investigated. In this paper, we propose a novel solution, EchoONE, for this problem with a SAM-based segmentation architecture, a prior-composable mask learning (PC-Mask) module for semantic-aware dense prompt generation, and a learnable CNN-branch with a simple yet effective local feature fusion and adaption (LFFA) module for SAM adapting. We extensively evaluated our method on multiple internal and external echocardiography datasets, and achieved consistently state-of-the-art performance for multi-source datasets with different heart planes. This is the first time that the MPS problem is solved in one model for echocardiography data. The code will be available at <a class="link-external link-https" href="https://github.com/a2502503/EchoONE" rel="external noopener nofollow">this https URL</a>.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: In clinical echocardiography examinations, in order to screen, diagnose, and treat heart diseases, multiple planar images of the heart structure from different perspectives are usually required. However, due to the huge differences in the heart structure between different planes, existing AI models must be custom - developed for each specific plane, which leads to repetitive development and additional complexity. Therefore, an effective solution to the multi - plane segmentation (MPS) problem is urgently needed to simplify the development and deployment processes.

Specifically, the paper proposes a new method named EchoONE, aiming to achieve accurate segmentation of multiple echocardiography planes with a single model. This is the first attempt to solve the multi - plane segmentation problem of echocardiography data in one model. The following are the main contributions of the paper:

1. **Proposing the EchoONE model**: Based on the SAM (Segment Anything Model) architecture, it can accurately segment the heart structure from different echocardiography planes. This is the first unified model for multi - plane segmentation of medical images.
2. **Introducing the PC - Mask module**: It is used to generate semantically - aware dense cues without knowing the specific plane information of the input image, thus making EchoONE a unified model applicable to all planes.
3. **Designing the LFFA module**: It is used for feature interaction and fusion, helping to improve the segmentation performance and accelerate convergence.
4. **Unified mask representation**: A unified mask representation is designed for multi - source echocardiography datasets using different annotation protocols, enabling EchoONE to be trained and validated uniformly on these datasets.

Through extensive experiments on multiple internal and external echocardiography datasets, EchoONE has demonstrated its superior performance in different ventricle, left atrium, and myocardial segmentation tasks, proving its effectiveness and robustness in the multi - plane segmentation problem.