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 <a class="link-external link-http" href="http://required.Additionally" rel="external noopener nofollow">this http URL</a>, 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.

What problem does this paper attempt to address?

This paper attempts to solve several key problems in ultrasound image segmentation, specifically including: 1. **Fuzzy Boundaries and Speckle Noise**: The inherent fuzzy boundaries and speckle noise in ultrasound images make it difficult for traditional segmentation methods to obtain accurate segmentation results. These problems lead to a decline in segmentation performance. 2. **Inefficiency of Interactive Segmentation**: Existing interactive segmentation methods rely on a large number of manual or random sampling prompts and require multiple iterations to achieve satisfactory performance. This not only increases the burden on users but also reduces efficiency. 3. **Lack of Targeting**: Traditional random sampling strategies lack targeting and cannot accurately locate the areas where the model needs the most improvement, thus affecting the learning effect of the model. To address these challenges, the authors propose an Evidential Uncertainty - Guided Interactive Segmentation (EUGIS) paradigm. The main goal of EUGIS is to introduce evidence uncertainty estimation, simulate the interactive behavior of expert radiologists, thereby achieving more efficient segmentation and reducing the number of required prompts and iterations. ### Main Contributions 1. **Proposing the EUGIS Paradigm**: Generate point - prompts based on evidence - based uncertainty estimation, simulate the interactive behavior of expert radiologists, achieve high - quality segmentation results and generalization performance, and at the same time reduce the number of interactions. 2. **Introducing Evidence Uncertainty**: Use Subjective Logic and Dempster - Shafer Theory (DST) to parameterize the Dirichlet concentration distribution, and develop a trainable uncertainty calibration mechanism to optimize the boundaries of certainty and uncertainty. 3. **Hybrid Image Encoder**: Adopt a hybrid image encoder to comprehensively capture local and global information, generate multiple segmentation results to optimize prediction, and enhance the robustness of the model. ### Method Overview The EUGIS paradigm is divided into two stages: - **First Stage**: Generate a high - quality uncertainty map. Generate evidence through a hybrid encoder - decoder architecture \( f_\theta \), and estimate uncertainty scores by combining DST and SL. Calibrate the uncertainty estimation to ensure that the model exhibits high uncertainty when making inaccurate predictions and high confidence when making accurate predictions. - **Second Stage**: Perform efficient interactive segmentation based on the uncertainty map. By adding a prompt encoder \( f_{\text{prompt}}^\theta \), use the high - quality uncertainty map generated in the first stage to create more targeted prompts, enabling the prompts to more flexibly reflect the interactive behavior of expert radiologists. ### Loss Function To better calibrate the uncertainty estimation, the authors introduce a new loss function CEU (Calibrated Evidential Uncertainty), which is defined as follows: \[ L_{\text{CEU}} = -\left( \alpha_t \sum_{j,k \in \{\hat{y}_{j,k} = y_{j,k}\}} \sum_{n = 1}^N b_{n,j,k} \log(1 - u_{j,k}) \right) - \left( (1 - \alpha_t) \sum_{j,k \in \{\hat{y}_{j,k} \neq y_{j,k}\}} \sum_{n = 1}^N (1 - b_{n,j,k}) \log(u_{j,k}) \right) \] where \(\alpha_t\) is an annealing factor used to control the penalty strength for uncertainty in different training stages. ### Experimental Results The experimental results show that EUGIS outperforms existing non - interactive and interactive segmentation methods when using only a single point - prompt, demonstrating its superior performance in ultrasound image segmentation. In summary, this paper aims to improve the accuracy and efficiency of ultrasound image segmentation and reduce the number of user interactions by introducing evidence - based uncertainty estimation, providing a new research perspective in the field of medical image segmentation.

Evidential Calibrated Uncertainty-Guided Interactive Segmentation paradigm for Ultrasound Images

A Novel Quality Evaluating Method for Over-Segmentation Approaches Using Real-Time Boundary Information

Towards Reliable Medical Image Segmentation by utilizing Evidential Calibrated Uncertainty

MedUHIP: Towards Human-In-the-Loop Medical Segmentation

Navigating Uncertainty in Medical Image Segmentation

Region-based evidential deep learning to quantify uncertainty and improve robustness of brain tumor segmentation

Evidence-based uncertainty-aware semi-supervised medical image segmentation

Uncertainty-aware Evidential Fusion-based Learning for Semi-supervised Medical Image Segmentation

Evidential Uncertainty-Guided Mitochondria Segmentation for 3D EM Images

Give me (un)certainty -- An exploration of parameters that affect segmentation uncertainty

SPA: Efficient User-Preference Alignment against Uncertainty in Medical Image Segmentation

Uncertainty-aware Cascade Network for Ultrasound Image Segmentation with Ambiguous Boundary

Uncertainty-Error correlations in Evidential Deep Learning models for biomedical segmentation

EDUE: Expert Disagreement-Guided One-Pass Uncertainty Estimation for Medical Image Segmentation

Hierarchical Uncertainty Estimation for Medical Image Segmentation Networks

Uncertainty Modeling in Ultrasound Image Segmentation for Precise Fetal Biometric Measurements

Uncertainty estimation using a 3D probabilistic U-Net for segmentation with small radiotherapy clinical trial datasets

Enhancing the Reliability of Segment Anything Model for Auto-Prompting Medical Image Segmentation with Uncertainty Rectification

Inter-Rater Uncertainty Quantification in Medical Image Segmentation via Rater-Specific Bayesian Neural Networks

Benchmarking Scalable Epistemic Uncertainty Quantification in Organ Segmentation

Calibrating Ensembles for Scalable Uncertainty Quantification in Deep Learning-based Medical Segmentation