BMC Medical Imaging最新文献

Chronic wound analysis is crucial for effective wound care, necessitating precise and efficient segmentation techniques. This study explores the use of YOLOv8 and YOLO11, renowned deep learning algorithms for object detection, in medical image segmentation. A comprehensive evaluation was conducted comparing different versions of YOLOv8 and YOLO11 with the benchmark U-Net model using the FUSeg and Wound Data databases. Both databases were used simultaneously for training and testing to achieve a cross-dataset validation, ensuring the robustness of our models, and highlighting the importance of having data from different sources to represent the problem in various clinical contexts, thereby allowing AI models to be more adaptable and generalizable. Results show that both YOLO models significantly outperform U-Net in terms of segmentation accuracy, generalization, and inference speed. YOLOv8n achieved the highest performance with an IoU of 71.7%, a precision of 83.8%, a recall of 77.9%, and a DSC of 79.3%, and, while YOLO11s, the best performing of the YOLO11 models (IoU of 70.1%, Precision of 81.9%, Recall of 79.9%, and DSC of 78.5%), distinguished itself for its stability between thresholds and robustness across datasets. This research confirms that modern YOLO architectures offer fast, accurate, and robust solutions for automated wound segmentation, laying the groundwork for further development in AI-driven wound analysis and diagnosis.

This work introduces DyABD, a novel and complex benchmark dataset of dynamic abdominal MRIs from patients with abdominal hernias and associated high quality abdominal muscle annotations. DyABD is the first-of-its-kind in four key ways; (1) it proposes the first abdominal muscle segmentation task, (2) the dynamic MRIs are acquired whilst the patients perform various exercises, introducing extreme anatomical variability, making it one of the most challenging segmentation datasets to date, (3) it includes both pre and post corrective MRIs and (4) DyABD promotes clinical research into the high recurrence rates of abdominal hernias. Beyond dataset introduction, this work provides a comprehensive evaluation of the generalisation capabilities of existing segmentation models across Supervised, Few Shot and Zero Shot paradigms on the unseen DyABD dataset. This work reveals that there is still room for substantial improvement in the field of medical image segmentation, with the majority of techniques achieving a Dice Coefficient of 0.82. This work therefore sheds light on the true progress of the field and redefines the benchmark for progress in medical image segmentation.

Background: Placenta accreta spectrum(PAS)is a major cause of maternal mortality during the perinatal period. This study aims to develop and validate a novel nomogram prediction model integrating clinical characteristics and ultrasound radiomics features for predicting PAS in pregnant women with placenta previa.

Methods: This retrospective two-center study included 271 pregnant women with placenta previa from two medical centers in China. Center 1 (n = 190) served as the training cohort, and Center 2 (n = 81) as the external validation cohort. Radiomic features were extracted from two-dimensional gray-scale ultrasound images of the placenta. Least absolute shrinkage and selection operators (LASSO)were used to select the radiomic features, and multivariate logistic regression identified independent clinical risk factors. A nomogram was constructed by incorporating selected radiomic features and significant clinical risk factors. Model performance was assessed using receiver operating characteristic (ROC) curve analysis (area under the curve, AUC) and decision curve analysis (DCA).

Results: The radiomics-only model achieved an AUC of 0.853 in the training cohort and 0.778 in the external validation cohort. The number of prior cesarean deliveries (CD) was identified as an independent risk factor and was therefore included in the clinical model, which was further integrated into a nomogram. The clinical-radiomics model, which combines radiomic features and clinical independent risk factors, achieved AUCs of 0.859 (training cohort) and 0.822 (external validation cohort). The integrated nomogram demonstrated slightly higher diagnostic performance than radiomics-only models (AUC 0.859 vs. 0.853 in the training cohort; 0.822 vs. 0.778 in the external validation cohort). DCA showed that the nomogram based on the clinical-radiomics model provided the highest clinical net benefit.

Conclusions: The clinical-ultrasound radiomics nomogram has favorable predictive value for PAS in patients with placenta previa. This tool may assist clinicians in optimizing prenatal management and potentially improving maternal prognosis.