.png){kind=icon}
.png){kind=icon}
Return
Uncertainty-aware point interaction for echocardiographic segmentation refinement
.png){kind=icon}
PRE
AI
C
X
W
Y
A
X
DOI：10.1016/j.displa.2026.103449.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Accurate segmentation of cardiac structures in echocardiographic imaging is crucial for reliable cardiac function assessment and clinical decision-making. However, ambiguous anatomical boundaries, imaging artifacts, and the coexistence of structures with varying sizes pose significant challenges for segmentation, often exacerbating instability at the boundaries. To address this issue, this study proposes a plug-and-play module termed Uncertainty-aware Point Interaction for Refinement (UPI). The proposed UPI module first identifies uncertain points and enables information interaction among them by constructing learnable information centers inspired by clustering principles, thereby enhancing the discriminative feature representation of these ambiguous regions. A subsequent loss-guided refinement process is applied to reclassify these uncertain points, leading to improved accuracy and enhanced stability in segmentation. Comprehensive experiments were conducted by integrating UPI into multiple state-of-the-art segmentation networks. The results demonstrate that incorporating UPI consistently improves the performance of these networks on the private echocardiographic dataset, CAMUS, and EchoNet-Dynamic. Across the three datasets, the average Dice improvements were 1.83%, 0.99%, and 0.51%, respectively, while the HD95 values decreased by 0.76, 0.68, and 0.60. Moreover, the UPI module achieves these performance gains with minor increases in parameters and computational cost, which remain substantially lower than enlarging baseline models. These results highlight the effectiveness and efficiency of UPI as a generalizable refinement strategy for echocardiographic segmentation. The code will be available at https://github.com/ Lv0620/UPImodel.
Keywords:
Uncertainty
Information interaction
Echocardiographic
Segmentation
Journal
IF:
3.4
Papers:
2.1K
Citations:
3.2K
Organization
Cited Papers