Evidential Calibrated Uncertainty-Guided Interactive Segmentation paradigm for Ultrasound Images

TL;DR

This work tackles robust ultrasound image segmentation with minimal user interaction by introducing EUGIS, an evidential uncertainty-guided interactive framework. It combines Dempster-Shafer theory and Subjective Logic to produce calibrated per-pixel uncertainty maps via a two-stage process: Stage I learns evidential uncertainty, while Stage II uses those maps to simulate targeted point prompts and iteratively refine segmentation. The approach employs a Dirichlet-based evidential learning objective (including CEU loss and KL regularization) and a hybrid encoder with multi-head outputs to capture both local and global information, achieving state-of-the-art results on BUSI, DDTI, and EchoNet with as few as one prompt. The findings demonstrate that uncertainty-guided sampling significantly reduces user prompts and iterations, enabling efficient and reliable interactive segmentation suitable for clinical deployment.

Abstract

Accurate and robust ultrasound image segmentation is critical for computer-aided diagnostic systems. Nevertheless, the inherent challenges of ultrasound imaging, such as blurry boundaries and speckle noise, often cause traditional segmentation methods to struggle with performance. Despite recent advancements in universal image segmentation, such as the Segment Anything Model, existing interactive segmentation methods still suffer from inefficiency and lack of specialization. These methods rely heavily on extensive accurate manual or random sampling prompts for interaction, necessitating numerous prompts and iterations to reach satisfactory performance. In response to this challenge, we propose the Evidential Uncertainty-Guided Interactive Segmentation (EUGIS), an end-to-end, efficient tiered interactive segmentation paradigm based on evidential uncertainty estimation for ultrasound image segmentation. Specifically, EUGIS harnesses evidence-based uncertainty estimation, grounded in Dempster-Shafer theory and Subjective Logic, to gauge the level of uncertainty in the predictions of model for different regions. By prioritizing sampling the high-uncertainty region, our method can effectively simulate the interactive behavior of well-trained radiologists, enhancing the targeted of sampling while reducing the number of prompts and iterations required.Additionally, we propose a trainable calibration mechanism for uncertainty estimation, which can further optimize the boundary between certainty and uncertainty, thereby enhancing the confidence of uncertainty estimation.

Figures (4)

• Figure 1: Different interactive segmentation paradigms, $x^t$ and $p^t$ refer to the input image and prompt in the iteration $t$, respectively. (a) Semi-automatic paradigm with the human-in-loop interaction process; (b) End-to-End paradigm with human-simulated prompt information; (c) Our proposed method with prompts generated by uncertainty map.
• Figure 2: Overview of our two-stage interactive segmentation paradigm through evidence-based uncertainty estimation. In the first stage, an evidential calibrated model is trained to generate the uncertainty map, and it retains uncertainty map from the first-stage model are used as the point prompt generator, retain the interactive segmentation model in the second stage, which can be repeated for $M-1$ iterations in the process of forward propagation.
• Figure 3: Visualization of comparison experimental results on the BUSI dataset. The two columns on the far right represent our proposed method EUGIS and the corresponding uncertainty map. Label, MSA and Med2D refer to the Ground Truth, Medical SAM Adapter and SAM-Med2D, respectively.
• Figure 4: The effect of the number of point prompts and iterations on performance in EUGIS.