| Chest tomosynthesis(CTS)is a new three-dimensional imaging technique.The system employs the limited angle motion of X-ray tube to obtain the projection,and then uses the projection to reconstruct the sectional image.Compared with chest radiography(CR),CTS partly reduces superimposed tissue and thus improves the visibility of anatomy,and the sensitivity of detecting pulmonary nodules is at least 3 times higher than CR.Compared with computed tomography(CT),CTS has higher coronal resolution,lower radiation dose to patients.CTS is a traditional imaging tool for detecting pulmonary nodules that helps in identifying lung parenchyma and describing abnormalities such as pulmonary nodules in greater detail.For CTS imaging,the projection images are acquired in a time period of 10-12 seconds,during which patients are required to be still and hold their breath.However,it is difficult for many patients to hold their breath fully during CTS scan,especially those with chronic obstructive pulmonary disease.The respiratory motion would reduce the detectability of CTS to the level of CR.Based on the above,it is of great clinical significance to automatically identify the patient’s respiratory state during CTS scan to assist the technician in judging whether it is necessary to scan again.On the other hand,the scan angular range of CTS is small,and the reconstructed image will be polluted by the structure before and after the focal plane.When the ribs overlap with part of the lung area,the ribs may block the lung tissue or lesions.Therefore,how to suppress structure occlusion and improve the efficiency of lesion detection and recognition is a problem worthy of study.In view of the above problems,the main research contents of thesis are as follows:1.An automatic detection model of respiratory signal based on CTS projection data is proposed,the model generates respiratory signal in real time by extracting diaphragm motion from the projection.Specifically,the basis motion of the diaphragm induced by the CTS imaging geometry is calibrated by a linear function,and the respiratory signal is obtained by subtracting the fitted linear function from the overall motion of the diaphragm.According to the estimated respiratory signals,the CTS projection was divided into 4 to 8 phases,and the projection of each phase was reconstructed by the iterative reconstruction algorithm with total variation regularization.Simulated multipurpose chest phantom,digital XCAT phantom data and three sets of patient data were adopted for the experiments,the results show that the proposed model can accurately extract respiratory signals from CTS projection.The results of four-dimensional CTS reconstruction show clear reduction of motion-induced blur.2.A CTS image enhancement model based on projection image bone suppression is proposed.Specifically,a model of CTS image enhancement is proposed,which uses CT to generate pseudo-CTS projections,and uses U-Net network combined with multiple loss functions to achieve bone suppression of projection.During the simulation,the lung CT with a thickness of less than 0.3 mm was used to generate pseudo-CTS with a resolution similar to that of the actual CTS images.Meanwhile,super-resolution technology was used to reconstruct the lung CT to improve the in-plane resolution of the simulation image.Data simulation generates data pairs containing projections and corresponding bone-free projections as the input of U-Net network.16 simulation data were used as the training set of the network,and 2 simulation data were used as the verification set.In order to evaluate the bone suppression ability of the model on actual CTS images,patient data obtained from Shimadzu system were used for testing.The results of both simulation data and the patient data show that the proposed method can effectively suppress bone structure while preserving the details of soft tissue in projection and tomography images,thus improving the imaging ability of lesion structure. |
PDF Full Text Request