Intravascular ultrasound (IVUS) can provide high-resolution cross-sectional images of coronary arteries, showing detailed information of the vascular lumen, tube wall, and athermanous plaques, which is helpful for the discovery or identification of early coronary atherosclerotic plaques. Multiple parameters extractable from the IVUS image can help the cardiologist analyze the pathology and assist in disease diagnosis and postoperative treatment. Typically, cardiologists manually label the intima and adventitia in the IVUS image, and obtain a limited number of parameters through the IVUS instrument, which is time consuming and labor-intensive. To assist the cardiologist in automatically obtaining more clinically relevant value parameters, a fully automatic IVUS multiparameter extraction framework is proposed. Based on the intima and adventitia obtained by DeepLab V3+, we propose a targeted noise reduction preprocessing framework adapted to IVUS. The framework implements the basic parameter extraction of IVUS through two newly proposed algorithms. And through the standard medical formula, the basic parameters are converted into 10 standard medical indicators. Standardized medical indicators are obtained by clinically relevant basic parameters. In terms of accuracy, this article used a clinical database obtained from Qilu Hospital of Shandong University and compared the results of the framework with the gold standard of cardiologists. The relative error of continuous IVUS main parameters between frames did not exceed 10.10%. The relative error of independent IVUS did not exceed 10.03%. Based on the distribution consistency of the parameters and the gold standard, a Bland-Altman plot of the parameters is proposed. It was verified that this distribution is basically in line with the gold standard of cardiologists. The algorithm in this article obtained a total of 10 parameters, far exceeding the parameters obtained by cardiologists and traditional IVUS machines. Its accuracy and speed can also meet the requirements of cardiologists for clinical diagnosis.

中文翻译：

---

使用 IVUS 图像进行多参数同步测量以智能诊断冠心病

血管内超声（IVUS）可以提供冠状动脉的高分辨率横截面图像，显示血管腔、管壁和动脉粥样硬化斑块的详细信息，有助于早期冠状动脉粥样硬化斑块的发现或识别。可从 IVUS 图像中提取的多个参数可以帮助心脏病专家分析病理并协助疾病诊断和术后治疗。通常，心脏病专家在 IVUS 图像中手动标记内膜和外膜，并通过 IVUS 仪器获取有限数量的参数，这既费时又费力。为了帮助心脏病专家自动获取更多临床相关的值参数，提出了一种全自动 IVUS 多参数提取框架。基于DeepLab V3+获得的内膜和外膜，我们提出了一个适应IVUS的有针对性的降噪预处理框架。该框架通过两种新提出的算法实现了IVUS的基本参数提取。并通过标准医学公式，将基本参数转化为10个标准医学指标。标准化的医学指标是通过临床相关的基本参数获得的。在准确性方面，本文使用了山东大学齐鲁医院的临床数据库，并将该框架的结果与心脏病专家的金标准进行了比较。帧间连续IVUS主要参数的相对误差不超过10.10%。独立IVUS的相对误差不超过10.03%。基于参数和金标准的分布一致性，提出了参数的Bland-Altman图。经验证，这种分布基本符合心脏病专家的金标准。本文的算法一共获得了10个参数，远远超过了心脏病专家和传统IVUS机器获得的参数。其准确性和速度也能满足心脏病专家对临床诊断的要求。