Research On Anomaly Detection And Quantitative Evaluation Of Pulmonary Perfusion Images

Pulmonary embolism is a respiratory disease caused by various emboli blocking the pulmonary artery.Due to its unclear clinical manifestations,patients have a high rate of missed diagnosis and misdiagnosis.Pulmonary embolism disease has a high mortality rate and seriously endangers people’s physical health.Therefore,early diagnosis of this disease is of great significance.SPECT pulmonary perfusion imaging is currently a commonly used non-invasive examination method for diagnosing pulmonary embolism and other lung diseases.However,due to the problems of low spatial resolution,large individual differences,and blurred imaging region boundaries in pulmonary perfusion imaging,it poses challenges for the diagnosis and quantitative analysis of pulmonary embolism diseases.For the above issues,proposing a three-stage quantitative evaluation method for obstructed areas in SPECT lung perfusion images to identify the occurrence of obstruction and evaluate its severity in a fine-grained manner.Overall,the mainly conducts the following research work:(1)Proposing a construction method for standard lung templates.Proposing two methods to establish a standard lung template based on the hot spot and background statistical values of lung perfusion images.The image fusion method is based on the distribution characteristics of imaging hotspots and background grayscale values.The pixel values are accumulated for image additive fusion to establish a normal lung contour,and combined with grayscale histograms and contour map thresholds to segment the lung area to obtain a lung template.The projection counting method is based on image data that can be represented as a matrix.By counting the elements of the imaging hot spot and background matrix,the normal lung contour is established using projection counting,and the matrix is automatically binarized to obtain the lung template.Evaluate the standard lung templates constructed using two methods in terms of shape,size,and position,and ultimately obtain high consistency results.(2)Proposing an adaptive threshold segmentation and an energy function based edge free active contour unsupervised hot spot segmentation model.Adaptive threshold segmentation is based on the characteristics of image imaging,with normal lung blood flow imaging and no blood flow defect imaging.By maximizing the inter class variance,a functional lung perfusion imaging hotspot is segmented.The unsupervised depth segmentation model optimizes the network parameters by minimizing the self supervised learning mechanism of the borderless active contour based on the energy function to segment the hot spot.The proposed method was evaluated using clinical data from SPECT pulmonary perfusion images.The experimental results showed that the unsupervised segmentation model can accurately and automatically segment the hot spots of pulmonary perfusion images and obtain reliable quantitative evaluation results of obstructed areas.The average DSC values for dual lung hot spot segmentation are 0.9590.(3)Proposing a method for detecting and evaluating the area of pulmonary obstruction.Based on the segmentation results of lung template and hot spot,image registration is performed by defining the relative position and size of the two to obtain the lung obstructed area.Specifically,by using the bounding rectangle of the target area to limit the scaling size,the registration of the hot spot and lung template is achieved based on image centroid translation,and the position of the obstructed area is obtained and its proportion is calculated.The quantitative evaluation results of the obstructed area obtained ICC consistency index values of 0.9155 and 0.8982 in the left and right lungs,respectively.Through the above research,the three-stage quantitative evaluation method for obstructed areas in lung perfusion images proposed can accurately delineate the obstructed areas in low resolution SPECT lung perfusion images and perform reliable quantitative evaluation.The proposed method has certain feasibility and reliability in the diagnosis of pulmonary embolism diseases.