Quantitative Analysis On Three-dimensional Images With COPD Lung Lesion In CT

In recent years, the incidence of critical diseases(such as cancer, cardiovascular, chronicdiseases and so on) increases gradually year by year with changes of environment and livinghabits. The diagnosis and treatment of critical diseases which not only affect the normal life andwork of patients but also their families become the focus in clinial work. With the increasingatmospheric pollution, there is a tendency of increasing number together with lower age of theincidence of chronic lung diseases year by year. chronic obstructive pulmonary disease(COPD)is the fourth leading cause of death among a variety of worldwide diseases, at the same time, itbecomes a priority for the prevention and treatment of critical diseases. Among COPD,idiopathic pulmonary fibrosis(IPF) with high prevalence, long sick time and premature mortalityis listed as one of difficult diseases by the World Health Organization. With the development ofmedical imaging technology, multi-slice CT and64-slice spiral CT which allow radiologists canobverse the organization and structure of the human body in the sub-millimeter thickness areused in clinical practice widely and become the first choice for the diagnosis of lung by imaging.IPF mainly show as ground-glass, honeycombing and so on in imaging, because of their diffusedistribution, low specificity, using conventional medical image segmentation algorithm can notobtain reliable result. Although the radiologists can rely on their extensive experience in readingfilms to aim the lesion area, but the quantitative analysis and evaluation of the lesion still is adifficult issue. For dealing with the above challenges, the paper mainly work around the3Dsegmentation and quantification of ground-glass and honeycombing which are common imagingmanifestation in CT images. The research can provide data to evaluate the progression of thedisease and the effect of treatment, meanwhile, the further research on imaging biomarker ofground-glass and honeycombing provides a solution for pushing diagnosis to enter uponquantitative diagnosis phase from qualitative diagnosis phase.The main contents and innovations of this research are as follows:1. For quantitative diagnosis of COPD,3D segmentation of ground-glass and honeycombingwhich are common imaging manifestation in CT images of lung was researched. On the basisof pathology of ground-glass and honeycombing, according to their different imagingmanifestation characteristics, we used a Markov random field model and support vectormachine classifier respectively to implement2D segmentation of lesion, and using the areaoverlap relationship between slices removed false area to achieve3D segmentation. Thesensitivity of3D segmentation algorithm of ground-glass was86.34%, the specificity was 97.14%, the accuracy was94.93%.The sensitivity of3D segmentation algorithm ofhoneycombing was85.17%, the specificity was95.41%, the accuracy was94.07%.2. Imaging biomarker of ground-glass and honeycombing in IPF was explored. Based on3Dsegmentation and quantification, we applied innovatively imaging biomarker on the basis of2D and3D imaging information to describe the two lesions. In view of the clinicalassessment principles for the severity of disease by the area ratio between lesion and the lungby visual inspection, there was variation between different observation by the sameradiologist and observation by different radiologists. For quantitative purpose, imagingbiomarker in2D and3D was calculated on the basis of segmentation of the two lesions.Contrasting the2D area and the3D volume imaging biomarker to corresponding diagnosisreports showed the volume radio between lesion and lung not only show a tendency whichwas consistent with the diagnosis report but also was more sensitive to changes in lesion size.The imaging biomarker can provide intuitive and reliable data representation for medicaltreatment and effect of drug.3. A sample database was built aimed at broad kinds of critical diseases. Based on the databasedifference of features in imaging manifestation were analyzed. In order to expand the abovestudy result to widely critical diseases, the paper did preliminary research on themulti-dimensional characteristic in imaging which can provide candidate feature forsubsequent quantitative analysis and imaging biomarkers.4. Designed and implemented a quantitative pulmonary multi-lesion component which can beintegrated into PACS system for clinical applications. The component was implementedbased on hybrid programming of VTK, ITK, MATLAB according their respective advantageto achieve the3D segmentation and quantification of ground-glass and honeycombing.Except that, the component can save the result of the3D segmentation and quantification byintroducing AIM which can facilitate clinicians to reproduce result and further interpret theimage content.