Research On Algorithm For Detecting Aortic Anomalies In CT Images

In recent years,coarctation of aorta has gradually become one of the main hazards to the life and health of newborns,and the importance of research related to its abnormality detection has become increasingly prominent.Currently,the clinical diagnosis of this disease mainly relies on medical imaging techniques,such as CT,echocardiography,and so on.It mainly relies on the subjective interpretation of medical images by doctors,and is prone to interference from various factors,such as fatigue level,experience,and image quality.Using artificial intelligence methods which are based on cardiac CT images and echocardiographic reports to develop an objective and accurate algorithm for detecting aortic coarctation anomalies can effectively improve diagnostic efficiency.Most of the existing algorithms for detecting coarctation of aorta are based on artificial intelligence technology,and only use a single feature of medical images,such as two-dimensional spatial features,and do not utilize multiple features such as two-dimensional and three-dimensional spatial features contained in the image,as well as association features between image sequences,resulting in accidental and inaccurate detection.Secondly,from the perspective of clinical practice,anomaly detection often relies on cardiac CT images or echocardiogram reports and other data,while the existing research mostly only rely on single mode data,and the model expression ability is not complete,resulting in poor detection reliability.In response to the above shortcomings,this article has done the following two tasks:(1)A double-flow network model for detecting aortic coarctation anomalies is proposed by fusing multidimensional spatial and sequential features contained in cardiac CT images.The network includes a 3DCoA model and an optical flow model.The 3DCoA model comprehensively considers the two-dimensional and three-dimensional spatial characteristics of cardiac CT images,while the optical flow model mainly considers the sequence information in the CT images.This algorithm makes full use of various feature information in cardiac CT images,effectively improving the accuracy of aortic coarctation anomaly detection.(2)A multimodal model for detecting aortic coarctation anomalies is proposed by fusing physiological index data from cardiac CT images and echocardiography.The model optimizes the traditional multimodal strategy by adding the similarity term of the feature vector to the traditional loss function,adding supplementary supervision terms of other modal data for the training of each modal model,and realizing the complementary advantages of different modal data.Compared with the traditional method,it can alleviate the error transmission between different modalities and effectively improve the reliability of aortic constriction abnormality detection.