Third Ventricle Width Measurements Based on YOLO and Localized Intensity Features

Transcranial ultrasound (TCS) imaging offers a quantitative assessment of the third ventricle width (TVW) in patients with Parkinson’s Disease (PD), helping physicians identify cognitive dysfunctions linked to PD as early as possible. Conventional TVW measurement techniques, however, suffer from inefficiencies and instabilities. To address these shortcomings, this study presents an automated TVW measurement for TCS images. Firstly, to address the challenge of detecting the third ventricle against the complex background of TCS images, this study introduces the YOLO-TV enhancement model to improve detection accuracy. Integrated into the feature extraction network is a multi-head attention mechanism, which strengthens the network’s capability to extract features at different scales. Meanwhile, a depth-separable path aggregation module is employed to improve the channel sensitivity during feature fusion. Secondly, the paper tackles the segmentation and measurement of the third ventricle. It proposes a segmentation algorithm based on regional local intensity feature calculation and multi angle projection positioning. The YOLO-TV model demonstrates a detection precision of 98.50% on the validation set. Moreover, the mean measurement deviation between the automated results produced by the proposed method and the manual measurements by physicians stands at 0.089 mm. The correlation coefficient between the two measurement methods is 0.9643, indicating the potential of the proposed method for accurate third ventricle width determination.