Clinical Application Of Dual-branch Deep Network CT Computer-Assisted Diagnosis In The Quantitative Evaluation Of Honeycombing Area In Idiopathic Pulmonary Fibrosis

Objective: To evaluating the clinical value of double-branch deep network CT computer assisted diagnosis in the quantitative measurement of pulmonary honeycombing volume in patients with idiopathic pulmonary fibrosis.Materials and methods:1.Case date: 33 cases of interstitial pulmonary fibrosis diagnosed by chest HRCT in the radiology department of our hospital from May 2016 to May 2018 were collected,and their pulmonary respiratory function was examined within 30 days of the imaging examination.Screening was conducted in strict accordance with the inclusion criteria diagnosed as IPF;Patients with known etiology and complications such as acute episodes,infections,fluid overload,or pulmonary embolism were excluded.There were 27 male patients and 6 female patients.Average age: 75.00±7.07years;Average smoking index(sticks/year)284±32;Body mass index(BMI):22.30±5.20kg/m2.2.Equipment and technology2.1 CT Scanning equipment and technology: More than 80% of IPF patients’ images were collected from dual-source 64-row CT scanner for chest high-resolution CT(HRCT)plain scanning,with a high consistency of scanning conditions.All patients were placed in supine position with hands raised above the head.At the time of scanning,the range of examination was from the tip of the lung to the posterior costal diaphragm Angle.Scanning parameters: the tube voltage is 120KV;The tube current is 150 m AS;Layer thickness 1 mm;Layer spacing 1mm,matrix 512 x 512.Lung window of the image: the window width was 1200 HU,and the window position was-600HU;Mediastinal window of image: window width 400 HU,window position 40 HU.2.2 The radiologist manually measured the honeycombing area: Without knowing the basic clinical information and pulmonary respiratory function results of the patients,two radiologists manually measured the honeycombing area of 6 Lung fields at the selected level,tracheal carina level,right inferior pulmonary vein level and the middle plane,using PACS system post-processing software of the radiology workstation.The measurement indexes were Honeycombing area(HA)and corresponding Lung area(LA),and the percentage of total HA area in total LA area(Honeycombing area%,HA%)and Mean Honeycombing area%(MHA%)in each Lung field were calculated.2.3 The honeycombing volume was quantitatively measured by the computer aided diagnosis technique of dual-branch deep network CT: Dual-branch deep network CT computer aided diagnosis technology using manual interactive semi-automatic segmentation algorithm to complete the segmentation of different types of lung shadows,and the software has the function of correcting the marked area.The quantitative measurement of honeycombing volume can be divided into three steps: extraction of double lungs,resection of intrapulmonary bronchi,and segmentation of different types of shadows.Then,Honeycombing volume(HV)and its percentage in total lung volume(Honeycombing volume%,HV%)were calculated using the segmentation results.2.4 Pulmonary function test: Before the examination,the patient was prepared to comply with the requirements of the pulmonary function examination.Records in a history information collection,relatively perfect forced vital capacity percent predicted(FVC%pred),1 seconds forced expiratory volume percent predicted(FEV1%pred),Forced expiratory volume as a percentage of forced vital capacity at 1s(FEV1/FVC%),carbon monoxide diffusion capacity percent predicted(DLco%pred),Carbon monoxide dispersion as a percentage of alveolar ventilation(DLco/VA%pred)CPI=91.0-[0.65×DLco%pred]-[0.53×FVC%pred]+[0.34×FEV1%pred]3.Statistical Analysis: SPSS 24.0 statistical software was used to analyze and process the data,and mean standard deviation(s)was used to record the measurement data.The Weighted Kappa coefficient was adopted to check the consistency of the data manually measured by two radiologists.Spearman correlation coefficient was used to quantitatively measure the honeycombing volume,manually measure the honeycombing area by radiologists,and analyze the correlation between PFTs and CPI using the dual-branch deep network CT computer-aided diagnosis technology.The absolute value of Spearman correlation coefficient(rs)was 1,p < 0.05 was considered statistically significant.Results: There was a strong consistency between radiologists A and B in measuring the range of honeycomb shadow at a particular level,and the weighted Kappa coefficient was 0.717.There was a significant correlation between the volume percentage measured by the dualbranch deep network CT computer aided diagnosis and the area range manually measured by the radiologist,and the difference was statistically significant(p < 0.001).Compared with the results of quantitative measurement of the percentage of honeycombing volume and lung function by dual-branch deep network CT computer aided diagnosis technology,spearman correlation coefficient was negative,and HV% was negatively correlated with FVCpred,FEV1%pred and DLco%pred,DLco/VA%pred showing statistical significance(p<0.05).HV% has no correlation with FEV1/FVC%,CPI,showing no statistical significance(P>0.05).Conclusion:1.There is a significant correlation between the volume percentage of quantitative measurement of honeycombing by dual-branch deep network CT computer-aided diagnostic technology and the manual measurement of honeycombing area by radiologists,with high accuracy.2.Dual-branch deep network CT computer aided diagnosis technology quantitatively measuring the volume percentage of honeycombing is correlated with FVCpred,FEV1%pred and DLco%pred,DLco/VA%pred showing statistical significance,which is helpful for clinical evaluation of the extend of desease,the therapeutic effect and prognosis of patients.