| 钱泽程,戴修斌,朱书进,冒添逸,王东苗.基于通道和空间多重注意力协同网络模型自动检测下颌第二磨牙牙根外吸收[J].中国医学影像技术,2026,42(3):419~424 |
| 基于通道和空间多重注意力协同网络模型自动检测下颌第二磨牙牙根外吸收 |
| Automatic detection of mandibular second molar external root resorption based on channel and spatial multi-attention collaborative network model |
| 投稿时间：2025-08-22 修订日期：2025-11-06 |
| DOI：10.13929/j.issn.1003-3289.2026.03.020 |
| 中文关键词: 下颌第二磨牙 牙根外吸收 放射摄影术,全景 通道和空间多重注意力 深度学习 |
| 英文关键词:mandibular second molar external root resorption radiography,panoramic channel and spatial multi-attention deep learning |
| 基金项目:江苏省社会发展——临床前沿技术项目(BE2023833)。 |
|
| 摘要点击次数: 67 |
| 全文下载次数: 53 |
| 中文摘要: |
| 目的 观察基于通道和空间多重注意力协同网络模型自动检测下颌第二磨牙(MM2)牙根外吸收(ERR)的价值。方法 构建融合通道注意力模块与空间-通道转换融合注意力模块的自动检测网络模型,以敏感度(Sen)、特异度(Spe)、准确率(Acc)和精确度(Pre)评估其自动检测MM2 ERR的效能;根据曲线下面积(AUC)和交并比(IoU)为0.5时的平均精度均值(mAP@0.5)等对该模型与人工判读和其他网络模型如AlexNet、GoogLeNet、VGG-16、ResNet-50及YOLOv5x进行对比。结果 所获通道和空间多重注意力协同网络模型在测试集的Sen、Spe、Acc、Pre及AUC均最高,分别为88.64%、81.82%、85.23%、82.98%及0.867,且其mAP@0.5为0.863。结论 基于通道和空间多重注意力协同网络模型用于自动检测MM2-ERR效能良好。 |
| 英文摘要: |
| Objective To explore the value of channel and spatial multi-attention collaborative network model for automatic detection of mandibular second molar (MM2) external root resorption (EER). Methods An automatic detection network model integrated channel attention module with spatial-channel transformation fusion attention module was constructed, and its performance for automatically detecting MM2-EER was evaluated according to sensitivity (Sen), apecificity (Spe), accuracy (Acc) and precision (Pre). The area under the curve (AUC) and the mean average precision at intersection over union (IoU) threshold 0.5 (mAP@0.5) of the model were compared with those of manual interpretation and other network models including AlexNet, GoogLeNet, VGG-16, ResNet-50 and YOLOv5x. Results A channel and spatial multi-attention collaborative network model was successfully constructed, with Sen, Spe, Acc, Pre and AUC in test set of 88.64%, 81.82%, 85.23%, 82.98% and 0.867, respectively, all were the highest ones, and its mAP@0.5 was 0.863. Conclusion Spatial multi-attention collaborative network model had good performance for automatically detecting MM2-ERR. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
