Diagnostic imaging modeling of thyroid nodules based on contrast-enhanced ultrasound and cross-scale attention mechanisms
10.19745/j.1003-8868.2025060
- VernacularTitle:基于超声造影和跨尺度注意力机制的甲状腺结节影像诊断模型研究
- Author:
Xiao-chen GONG
1
;
Zi-hao WANG
1
Author Information
1. 苏州市立医院,江苏 苏州 215000
- Publication Type:Journal Article
- Keywords:
contrast-enhanced ultrasound;
channel attention;
spatial attention;
cross-scale attention mechanism;
thyroid nodule;
image diagnosis
- From:
Chinese Medical Equipment Journal
2025;46(4):9-14
- CountryChina
- Language:Chinese
-
Abstract:
Objective To develop an image diagnosis model based on contrast-enhanced ultrasound and cross-scale attention mechanisms to improve the accuracy for diagnosing thyroid nodules.Methods Firstly,relevant data of thyroid nodules were obtained with contrast-enhanced ultrasound and then preprocessed.Secondly,ResNet-101 was selected as the backbone network for feature extraction of images,and feature enhancement was carried out with the modules of attention interaction across scale channels(AIAC)and spatial attention interaction across scales(SAIAC);finally,the generated high-dimensional feature vector was sent to K nearest neighbor(KNN)classifier for classification.To validate the performance of the proposed model,ablation and comparison experiments were performed using diffusion-weighted imaging,diffusion tensor imaging,and ultrasonography images as datasets.Results Ablation experiments showed the proposed model outperformed the baseline model with the average precision(AP),AP in case the intersection over union was 0.5(AP50)or 0.75(AP75)being 0.490,0.938 and 0.485,respectively.Comparison experiments indicated the proposed model behaved better than the models of bi-directional feature pyramid network(Bi-FPN)and feature pyramid network(FPN)with the prediction accuracy,recall rate and F1 score being 0.944,0.939 and 0.924,respectively.Conclusion The proposed model gains advantages in feature capture and image information enhancement,which can improve the accuracy and robustness of thyroid nodule diagnosis.[Chinese Medical Equipment Journal,2025,46(4):9-14]
