Medical Image Deep Parameter Measurement And Disease Prediction Based On Deep Learning

In recent years,the integration of deep learning and medical imaging has gradually increased,and the use of computers to extract depth parameters of clinical medical images and assist disease prediction has become a general trend Computer-assisted disease diagnosis and treatment can provide doctors with more details of the patient’s condition,and can even avoid the deterioration of the patient’s condition caused by the doctor’s misdiagnosis.This article mainly deals with the field of deep parameter extraction of intravascular ultrasound(IVUS)and prediction of hemorrhagic transformation based on non-perfusion CT.In terms of deep parameter extraction,IVUS can provide high-resolution cross-sectional images of coronary arteries,as well as display the deviation of the intima,adventita,and multiple types of plaque.Through IVUS observation,early coronary atherosclerotic plaque can be detected and treated in time.The multiparameter synchronous measurement of IVUS can assist cardiologists in analyzing,diagnosing,post-operative treatment and even disease prevention of intravascular lesions.In terms of disease prediction,hemorrhagic transformation(HT)has been regarded as a patient safety checkpoint for the treatment and secondary prevention of arterial ischemic stroke in recent years Accurate HT prediction can significantly reduce the number of patients who are misdiagnosed by doctors.In recent years,all HT predictions have relied on contrast perfusion images using contrast agents.The use of contrast agents will greatly increase the workload of doctors,as well as cause high medical costs for patients and even cause secondary brain damage.And due to the great challenge of non-perfusion images,almost all machine learning algorithms have never used non-perfusion CT for HT prediction.Aiming at the task of IVUS multi-parameter extraction,this research proposes a fully automatic IVUS multi-parameter synchronous measurement framework.The whole framework is based on DeepLabV3+’s intima and adventitia segmentation results,and an IVUS adaptive target noise reduction post-processing method is proposed.By proposing two novel graph analysis algorithms,the multi-parameter automatic synchronous measurement framework realizes the basic parameter extraction of IVUS.This framework uses standard medical parameter formulas to post-process the basic parameters,and achieves the extraction of up to 10 common IVUS medical indicators.In summary,the framework proposed in this paper realizes the automatic acquisition of IVUS clinical medical parameters,and realizes the three advantages of large number,fast speed,and high accuracy.The framework greatly reduces the manual labeling workload of doctors and provides extremely important clinical diagnosis.The clinical parameters of the patient’s condition can be used to analyze the patient’s deterioration more accurately.For the difficult prediction task of hemorrhagic transformation disease based on non-perfusion CT,a dual branch separation and enhancement network(DBSE-Net)is proposed for weak feature extraction and high-safety hemorrhagic transformation prediction without contrast agent.DBSE-Net innovatively uses a dual-branch separation and fusion mechanism to achieve adaptive extraction of weak features.In the self-encoding sub-module of DBSE-Net,the brain compression assessment branch(BCAB)and the infarct area assessment branch(IAB)are proposed to implement a lightweight encoding structure of the adaptive receptive field.In the feature selection stage,based on DBSE-Net’s keyframe selection algorithm and region guidance knowledge,DBSE-Net removes redundant information and clearly shows the severity of the lesion area.In general,DBSE-Net integrates global features and local features to obtain multi-scale and multi-category brain state information,thereby enhancing weak features in non-perfusion CT images and ultimately achieving accurate HT prediction.At the same time,this article has undergone a number of horizontal comparison experiments and ablation experiments to fully test the effectiveness of each network layer structure.The experimental results fully demonstrate that DBSE-Net can help doctors perform HT risk assessment for patients with intracranial stroke and may become an effective tool to prevent doctors from misdiagnosing high-risk HT patients in the future.