Application Of Texture Feature In Lung Cancer Of¹⁸F-FDG PET/CT Imaging

Purpose:Lung cancer has replaced liver cancer to become the leading causes of cancer death in China. It’s important for lung cancer patients to get early diagnosis and treatment. Imaging commonly used in lung cancer diagnosis includes chest X-ray, CT, ultrasound, MRI, PET, PET/CT, etc. PET/CT imaging, as a noninvasive method of tumor, combines the metabolic function and anatomical features in lung cancer diagnosis, differential diagnosis, staging, effective monitoring, radiotherapy planning, etc. Image texture features, as a response indicator of statistical measurement of changes in gray level within the region of interest, can quantify the characteristics. Texture features are extracted and applied to texture description, classification, etc. Texture features can be used for X ray, CT, MRI, ultrasound images of the lesion recognition, segmentation, diagnosis, and can aid clinical diagnosis and treatment. Based on18F-FDG PET/CT imaging, the first section describes the software development, and the second part is to analyze the texture in the normal population and to explore the diagnostic value of texture feature, and then make the comparison with SUV method and physicians method.Methods:Siemens biograph64and Siemens biograph sensation16were used in the18F-FDG PET/CT imaging. This study selected190healthy people, including99male and91female, and190healthy people were grouped into three (≤44,45～59,≥60).301patients were selected, of which180cases were confirmed lung cancer via surgery, biopsy and bronchoscopy, the other121patients were diagnosed with benign lung lesions. Lung cancer was grouped in accordance with pathological type, tumor size, FDG uptake (SUV). PET/CT DICOM image data of190healthy people and301patients were transfered to the software system, PET images and CT images got automatic registration. In the190cases of healthy people, right upper lobe, middle, lower lobe and left upper lobe, lower lobe were outlined3cm*3cm region of interest (ROI) respectively, and each lobe outline two interesting area. In301patients, the contours of lung lesions in CT images were outlined t at each slices, the contours of PET imaging were mapped in the same position.13texture parameters (mean, standard deviation, skewness, kurtosis, ASM, entropy, GTSDM_contrast, correlation, coarseness, contrast, busyness, complexity and strength) and SUV values were calculated. FDG PET/CT imaging performance of texture features in the normal population were observed, and effective texture parameters were extracted to determine its value in the diagnosis of lung cancer, and the texture feature method were compared with the SUV method, the physicians method in the sensitivity, specificity and accuracy of differential diagnosis. All results were analyzed with SPSS13.0.Results:In the normal population18F-FDG PET/CT imaging, there was no statistical difference in kurtosis between groups of5lung lobes (P>0.05). All texture features except for contrast, frequency, complexity, intensity were no different in gender groups (P>0.05). Kurtosis in the age groups was no difference in age groups (P>0.05).In301patients,180patients with lung cancer had1154slices in PET image and121patients with benign lesions had517. There were statistically significant differences in13texture parameters between malignant and benign groups (P<0.05), Mean, standard deviation, skewness, entropy, correlation, contrast, complexity and strength in malignant group were higher than benign group, while kurtosis, ASM, GTSDM_contrast, coarseness and busyness were lower. Mean, kurtosis, roughness and complexity were extracted via Logistic regression analysis. Area under the ROC curve was0.806(0.784,0.828)(P<0.001), which showed that the texture features had an moderate diagnosis in lung cancer of18F-FDG PET/CT imaging.In the group of Lung cancer, patients were divided into squamous cell carcinoma, adenocarcinoma, and other groups. Mean, kurtosis, coarseness and complexity were significant differences between3groups (P<0.0001). Patients were divided into>3. Ocm and≤3.0cm group, mean, coarseness and complexity were statistically significant between2groups (P<0.0001), kurtosis had no significant difference (P>0.05). Patients were also divided into three groups by SUVmax2.5and4.0, mean, coarseness and complexity were statistically significant among3groups (P<0.0001), and kurtosis had no significant difference.The sensitivity, specificity, and accuracy of texture features, SUV, nuclear medicine physician to diagnose benign and malignant pulmonary lesions in FDG PET/CT imaging were87.87%(1014/1154),47.78%(247/517),75.46%(1261/1671),80.0%(144/180),58.7%(71/121),71.4%(215/301),92.8%(167/180),85.1%(103/121),89.7%(270/301). Texture was more sensitive than SUV (P<0.01), while specificity is lower (P<0.05). Texture sensitivity had no difference with physician (P>0.05), the specificity and accuracy of texture is lower than physician(P<0.01).Conclusion:There were different texture parameters of FDG PET/CT imaging in lung lobes of normal people. Texture feature can be used for FDG PET/CT imaging in the classification of benign and malignant lung lesions. Texture parameters were different in histologic type, tumor size, FDG uptake groups of lung cancer, and the sensitivity of texture feature had no significant difference with that of physicians, and texture feature can be further used for PET/CT images of the computer-aided diagnosis.