BMC Medical Imaging最新文献

Continuous and accurate segmentation of airways in chest CT images is essential for preoperative planning and real-time bronchoscopy navigation. Despite advances in deep learning for medical image segmentation, maintaining airway continuity remains a challenge, particularly due to intra-class imbalance between large and small branches and blurred CT details. To address these challenges, we propose a progressive curriculum learning pipeline and a Scale-Enhanced U-Net (SE-UNet) to improve detail extraction, thereby enhancing segmentation continuity. Compared with previous connectivity-aware methods, our framework directly tackles the imbalance between large and small branches through end-to-end progressive learning, while balancing airway tree completeness and accuracy. Specifically, our curriculum learning pipeline comprises three stages. Stage 1 performs coarse learning to extract main airways. Stage 2 introduces a General Union Loss (GUL) to improve the identification of smaller airways. In Stage 3, we propose an Adaptive Topology-Responsive Loss (ATRL), which encourages the network to focus on preserving airway continuity. Throughout all stages, a crop sampling strategy is employed to reduce feature interference between airways of varying scales, effectively addressing the intra-class imbalance. The progressive training pipeline shares the same SE-UNet, integrating multi-scale inputs and Detail Information Enhancers (DIEs) to enhance information flow and effectively capture the intricate details of small airways. Additionally, we propose a robust airway tree parsing method and hierarchical evaluation metrics to provide more clinically relevant and precise analysis. Extensive experiments demonstrate that our method outperforms existing approaches on the ATM'22 challenge dataset and achieves a 9.631% improvement in the Tree length Detection rate (TD) of small airways and a 4.622% improvement in the Branch Detection rate (BD) of the overall airway tree on our in-house dataset, thereby significantly enhancing small-airway accuracy and overall airway tree completeness.

Background: The cervix undergoes morphological and structural changes during pregnancy in preparation for delivery. Assessing the progression of these changes using transvaginal ultrasound (TVUS) is crucial for preterm birth prediction. However, existing methods such as cervical length have limitations in capturing subtle tissue changes. Although tissue analysis using TVUS has been explored to address these limitations, achieving consistent and reproducible results in quantitative analysis remains challenging due to high inter-observer variability and a lack of standardized region of interest (ROI) definitions. This study proposes an oriented cervical canal region (O-CCR) framework that identifies regions containing ultrasound features while preserving anatomical spatial information.

Methods: We utilized 1436 TVUS images for training, validation, and testing, with 189 additional images from a different hospital for external validation. CCR was defined to include the cervical canal and its surrounding region after aligning the IO and EO parallel to ensure anatomical consistency in the cervix. To validate the effectiveness of O-CCR in handling various orientations, we applied five oriented object detection models (Oriented R-CNN, ReDet, S²A-Net, R³Det, and Oriented RepPoints) and evaluated their CCR localization performance.

Results: We compared the performance of five models implemented within O-CCR framework. Among them, Oriented RepPoints achieved the highest average precision (AP) of 0.981 at the intersection over union (IoU) threshold of 0.5, compared to Oriented R-CNN (0.968), S²A-Net (0.962), ReDet (0.964), and R³Det (0.980) on the test dataset. Notably, Oriented RepPoints demonstrated superior performance even at higher thresholds of 0.6 (0.931) and 0.7 (0.743) and the lowest average orientation error (AOE) of 9.1980 in CCR localization.

Conclusion: The proposed O-CCR framework exhibited reliable performance in CCR localization regardless of varying orientations and morphological configurations, while providing standardized regions that preserve the anatomical spatial information of the cervix. The consistent CCR could be applied to quantitative analysis of cervical tissue properties in future research. Ultimately, this approach could support the development of automated cervical change assessment for prenatal care.