摘要
鉴于良性阵发性位置性眩晕的复杂致病因素和诊断困难等问题,提出一种新的基于非局部卷积和卷积注意力模块(CBAM)的眩晕眼震诊断方法。首先,通过构建目标检测模型定位瞳孔,从而捕捉眼球运动并提取水平和垂直运动轨迹时序数据。其次,采用分类模型进行分类检测,该分类模型采用非局部卷积来捕获眼震数据中的远程依赖关系特征,并引入CBAM来提取特征层中的高级和低级语义信息,从而提高了分类模型的检测性能。在眼耳鼻喉科医院提供的视频眼震数据集上进行了实验,结果表明,与主流方法相比,本文所提出的诊断方法在精确率、召回率、准确率、平均F1值等评估指标上比主流方法分别提高了1.82%、2.09%、1.62%和1.96%,表明了本文方法的显著性优势。
In view of complex pathogenic factors and diagnostic difficulties of benign paroxysmal positional vertigo,a novel method for diagnosing nystagmus in vertigo based on non-local convolutional and convolutional block attention module(CBAM)is proposed.An object detection model is constructed to locate the pupil,thereby tracking eye movement and extract temporal data of horizontal and vertical motion trajectories.Subsequently,a classification model is employed for detection and classification,utilizing a non-local convolutional module to capture remote dependency relationships in nystagmus data,and introducing CBAM to extract high-and low-level semantic information in the feature layer for enhancing the detection performance.Experiments were conducted on a video nystagmus dataset provided by the Eye,Ear,Nose and Throat Hospital.The results show that compared with the best mainstream method,the proposed method improves precision,recall rate,accuracy,and average F1 score by 1.82%,2.09%,1.62%,and 1.96%,respectively,demonstrating its superiority.
作者
贺斌
高永彬
HE Bin;GAO Yongbin(School of Electronic and Electrical Engineering,Shanghai University of Engineering Science,Shanghai 201620,China)
出处
《中国医学物理学杂志》
CSCD
2024年第5期571-578,共8页
Chinese Journal of Medical Physics
基金
上海市科委“科技创新行动计划”社会发展科技攻关项目(21DZ1204900)。
关键词
良性阵发性位置性眩晕
医学图像处理
时序数据分类
目标检测
视频眼震数据分类
benign paroxysmal positional vertigo
medical image processing
temporal data classification
object detection
video nystagmus data classification
