Intelligent Detection Technology And Application Of Coronary Artery Lesions Based On Intravascular Ultrasound

With the continuous development of medical technology,medical image has gradually become the key to providing reliable evidence for doctors.In the new era,artificial intelligence technology is also constantly improving,and a large number of excellent algorithms have emerged.How to use complex machine learning(ML)or deep learning(DL)models to detect and segment the lesion area contained in medical images has become the focus of many researchers.In recent years,due to the rapid development of deep learning,the field of image segmentation,including medical images,has achieved unprecedented development.How to study image segmentation algorithms while making them applicable to the practice of auxiliary diagnosis has become a hot topic.Since there are currently few public datasets on Intravascular Ultrasound(IVUS)images,how to construct intravascular ultrasound image datasets,and the datasets need to have professional evaluation and labeling,which has become a major challenge in the process of image segmentation algorithm research.At the same time,because the intravascular ultrasound image is a grayscale image,and there is no obvious boundary feature and uniform pixel area like the natural image,which makes the feature extraction of the algorithm face great difficulties.How to construct an effective internal and external membrane segmentation model for feature extraction becomes particularly critical.In the process of doctor’s clinical diagnosis and treatment,after obtaining the patienr’s intravascular ultrasound image,the doctor not only needs to observe the image to judge the patient’s condition,but also needs to calculate some key clinical parameters implicit in the image based on the intravascular ultrasound image.For example,plaque burden,remodeling index,eccentricity index,etc.These indicators can assess the severity and fragility of coronary atherosclerotic plaques,so that doctors have a clearer understanding of the patienr’s condition.It is helpful for doctors to make accurate diagnosis and targeted surgical treatment of coronary atherosclerotic heart disease.Most of the information hidden in intravascular ultrasound images needs to be manually annotated by doctors and obtained through professional parameter calculation tools.This is time-consuming and laborious for the doctor.How to use image segmentation algorithms to simplify the work of doctors has become another challenge currently facing.In response to the above challenges,this thesis studies the intelligent detection technology of coronary artery lesions based on intravascular ultrasound,and develops a computer aided diagnosis system(CADS)for the intelligent detection of coronary artery lesions on this basis,which is of great importance to assisting doctors in clinical diagnosis.The work of this thesis mainly includes the following three:(1)In view of the scarcity of public datasets of intravascular ultrasound images,this article uses IVUS image data of real patients collected from Qilu Hospital of Shandong University,which is clinically significant.The dataset includes 5242 intravascular ultrasound images of more than ten patients and their corresponding gold standard for vascular intima and adventitia segmentation provided by cardiologists or medical students,ensuring the validity and standardization of the dataset.This article also constructed the IVUS image parameter calculation dataset,which has a small sample size and contains 383 images manually annotated by cardiologists.Among them,there are 211 non-continuous frame images and 172 continuous frame images related parameters,which are used as the gold standard for system parameter calculation.(2)In view of the complexity of medical images,this thesis compares the excellent model effects in natural image segmentation and finds that U-net performs better.Therefore,this thesis uses active contour models,edge operators based on intensity changes,and the K-means clustering algorithm performs feature extraction of prior knowledge,using U-net as the network framework,using the attention mechanism to replace the skip structure in the U-net network,and constructing a segmentation model for intravascular ultrasound images.After the a priori segmentation algorithm outputs the segmentation result,the relevant parameter calculation method is used to calculate the 10 clinical parameter indicators according to the segmentation result,which provides a basis for assisting doctors in clinical diagnosis.It was verified on the constructed dataset that the mIoU value of intravascular ultrasound image segmentation model reached the highest 0.8616,which verified the effectiveness of the model.(3)Since the ultimate goal is to provide doctors with actual auxiliary diagnosis,a computer aided diagnosis system for the intelligent detection of coronary artery lesions was developed.The seven functional modules of the system mainly include DICOM file analysis,IVUS image segmentation,segmentation result modification,parameter calculation,result storage,system status display,and display type conversion can simplify the doctor’s diagnosis and treatment process to a large extent and improve the doctor’s diagnosis and treatment efficiency.The segmentation result modification module can realize feedback to the image segmentation algorithm and provide data support for algorithm optimization.