Application of snake model imaging selection and analysis to the ultrasound diagnosis of white matter damage in preterm infants.

Abstract

Background: Approximately 20% to 30% of preterm infants may develop white matter injury. Early detection of brain injury is of great significance for clinical practice. The snake model is a boundary detection technology used in image processing and computer vision. It automatically locates the closed boundaries of objects in the image by simulating the behavior of elastic snakes. The quantitative ultrasound segmentation technique using the snake model can improve the accuracy of the diagnosis of early white matter injury.

Methods: In this retrospective single-center study, cranial ultrasound scans from 60 preterm infants with clinically confirmed WMD and 40 healthy controls were analyzed. Five regions of interest (ROIs) around key brain structures were automatically delineated by the snake model, and mean grayscale values were measured. Reproducibility was assessed via inter- and intra-observer analyses, and receiver operating characteristic (ROC) curves determined optimal thresholds for WMD prediction.

Results: The snake model achieved a mean area under the ROC curve (AUC) of 0.87 for ROI₁, 0.85 for ROI₂, and 0.82 for ROI₃, with corresponding sensitivities of 0.82, 0.70, and 0.63 and specificities of 0.88, 0.85, and 0.88. The average Dice coefficient across ROIs was 0.75, and The total coefficient of variation was 6.7%, which was less than 10%, indicating high reproducibility.

Conclusions: Quantitative analysis of ultrasound images with the snake model demonstrates promising accuracy and reproducibility for early WMD detection in preterm infants. Future multicenter studies with larger cohorts are warranted to validate these findings.

Clinical trial number: Not applicable.