BMC Medical Imaging, Год журнала: 2024, Номер 24(1)

Опубликована: Июль 25, 2024

Abstract Objective There are two major issues in the MRI image diagnosis task for Parkinson's disease. Firstly, there slight differences images between healthy individuals and patients, medical field has not yet established precise lesion localization standards, which poses a huge challenge effective prediction of disease through images. Secondly, early traditionally relies on subjective judgment doctors, leads to insufficient accuracy consistency. This article proposes an improved YOLOv5 detection algorithm based deep learning predicting classifying Methods improves YOLOv5s network as basic framework. CA attention mechanism was introduced enable model dynamically adjust local features image, significantly enhancing sensitivity PD related small pathological features; replace dynamic full dimensional convolution module optimize multi-level extraction Finally, coupling head strategy is adopted improve execution efficiency classification tasks separately. Results We validated effectiveness proposed method using dataset 582 from 108 patients. The results show that achieves 0.961, 0.974, 0.986 Precision, Recall, mAP, respectively, experimental superior other algorithms. Conslusion achieved high accuracy, can accurately detect recognize complex Significance shown good performance provide clinical assistance doctors diagnosis. It compensates limitations traditional methods.

Язык: Английский