| Part I Establishment of A diagnostic model for charactering solitary pulmonary nodule based on 18F-FDG PET/CT and clinical dataObjectiveIntegrate 18F-FDG PET/CT and clinical information to create a new differential diagnostic model of solitary pulmonary nodule(SPN) and then explore its diagnostic efficacy.Materials and Method1. Research objects. Between November 2004 and May 2014,186 patients with an indeterminate SPN detected during normal clinical work or for other reason referred for PET/CT scanning were retrospectively identified from the database of our PET center.164 cases have been included with pathological conclusion or clinical follow-up time was more than 2 years, excluding the 22 cases without pathological result or follow-up time was less than 2 years. All lesions were in line with SPN disease diagnostic criteria, and were solid nodules of diameters between 8mm and 30mm. All medical records were reviewed to obtain the following data:age, gender, smoking history (current or former cigarette smoker), disease history of malignancy, pathological conclusion, and last date of clinical and radiologic follow-up (including disappearance of the SPN or decreased size, no growth, or growth).2.18F-FDG PET/CT scan. PET/CT scanning was performed with the Discovery LS PET/CT scanner (GE,USA) or Biograph mCTx scanner (Siemens, Germany).18F produced by PETtrace cyclotron (GE, USA), and 18F-FDG synthesized by chemical synthesis module automatically, radiochemical purity> 95%. Fasting more than 6h, before injection of 18F-FDG, patients should be measured venous blood glucose levels to ensure that blood glucose levels (<7mmol/L). Venous blood glucose levels should be controlled to lower than 7mmol/L in patients with diabetes. After intravenous injection of 18F-FDG 259～444MBq in calm state, lying in the dark for 50～60min, patients should urinate before PET/CT scanning. Whole body scan ranged from mid-thigh to the top of the head. Image reconstruction used the OSEM. The PET and CT images transferred to the workstation for image fusion. After the whole body PET /CT scanning, thin-section CT scanning on the same machine were performed in the nodule.3. Imaging analysis. PET/CT fusing images, PET images and CT images of same patient were analyzed frame by frame by the specific software of workstation. Two experienced nuclear medicine physicians reviewed PET scans and scored the intensity of 18F-FDG uptake using a two point scoring scale(TPSS) (0, no higher than mediastinal blood pool; 1, higher than mediastinal blood pool). Two radiologists visually confirmed lesions received from thin-section CT and record the location of the lesion, nodule diameter and the presence of calcification, spiculated sign, lobulated sign, vessel convergence sign, pleural indentation sign and air bronchogram. After drawing the ROI, SUVmax was calculated automatically by the workstation The criteria of malignancy was SUVmax≥2.5.4. Diagnostic basis. Solitary pulmonary nodule diagnosed based on histopathologic findings or clinical and radiologic follow-up. The time of follow-up was defined as the time between PET imaging and histologic diagnosis or the date of the last radiologic follow-up. Radiologic follow-up typically consisted of repeat chest CT scans. Lesions were classified as benign in case of benign pathologic findings, the disappearance of the lesion at radiologic follow-up, or decreased or no change in size within an observation period of at least 2 years. All malignant lesions were identified by pathology.5. Model establishment. All data were conducted binary logistic regression analysis by using statistical software SPSS 13.0. The selected variables were statistically significant on the establishment of a logistic regression model. This model expresses the probability of malignancy as follows:P=1/(1+e-z), where z=α+β1X1+ β2X2+...+pnXn; e is the base of natural logarithms; a is a constant; X1、X2...Xn are independent variables including age (years), gender (0:Male,1:female), smoking history (1:now or ever smoked; 0:others), history of malignancy (1:has a history of extrathoracic cancer that had been diagnosed>5 years ago; 0:others), nodule location (1:located on the upper lobe; 0:others), nodule diameter (mm), nodule spiculated sign(1:exist; 0:others), lobulated sign(1:exist; 0:others), vessel convergence sign(1:exist; 0:others), pleural indentation sign(1:exist; 0:others), air bronchogram(1:exist; 0:others), calcification(1:exist; 0:others), TPSS (1:higher than mediastinum; 0:others); β is the regression coefficient; P represents a conditional probability, ranging between 0 and 1.6. Statistical analysis. All data were conducted statistical analysis by using statistical software SPSS 13.0. The comparison of two groups of relevant clinical indicators used two sample t test or Mann-Whitney test. The diagnostic accuracy of two method received the paired ROC curve analysis by using statistical software Medcalc. P<0.05 was considered statistically significant.Results1. Clinical information of patients with SPN. In total,164 eligible patients with SPN were identified, of whom 104 proven to have malignant nodules and 60 have benign lesions. All patients with malignancy were diagnosed by histopathology. The diagnosis of a benign lesion was based on the histopathologic results in 41 patients and on the stabilization or spontaneous decrease in the size of the lesion on a follow-up CT scan in 19 patients.2. Diagnostic potency comparison of different diagnostic criteria (TPSS and SUVmax threshold method) of 18F-FDG PET/CT for differentiating SPN. The AUC of ROC curves of PET using TPSS and SUVmax threshold method for charactering SPN were 0.80 (95% confidence interval [CI]:0.73-0.86) and 0.83 (95%CI:0.76～ 0.88). Discriminant effect of both methods was no statistically significant difference (z= 1.31, P= 0.19). Classifying a SPN with SUVmax> 2.5 as malignancy yielded a sensitivity of 94.2%(98/104) and a specificity of 60.0%(36/60). When a nodule 18F-FDG uptake was higher than mediastinal blood pool considered as malignancy, the diagnostic sensitivity and specificity were 94.2%(98/104) and 65.0%(39/60).3. Establishment of the SPN diagnostic model. Established a diagnostic model with clinical factors, CT information and PET metabolic information by using the binary logistic regression. Among 13 independent variables(including age, gender, smoking history, disease history of malignancy, size of lesion, location of lesion, calcification, spiculated sign, lobulated sign, vessel convergence sign, pleural indentation sign, air bronchogram, TPSS), only 3 variables which were the age, spiculation and metabolic information were demonstrated to have statistically significant difference (x2 were 4.164,12.430,29.981, respectively; all P values lower than 0.05). History of malignancy due to too few cases (only two cases have a history of malignancy) were excluded from the logistic regression analysis. The model was as follows:P= 1/(1+e-x), Where x=-5.124+0.058 x (age)+2.244 x (nodule speculation)+3.881 x (TPSS).4. Diagnostic potency comparison of the model and the TPSS for charactering SPN. The ROC-AUC value of the model and the TPSS were 0.91 (95% CI:0.86-0.95) and 0.80 (95% CI:0.73-0.86). Diagnostic potency of the model outperformed the TPSS (z= 4.271, P< 0.0001). When the model taken optimum operating point P= 0.8475 (ie P> 0.8475 diagnosed with lung cancer), the diagnostic sensitivity of lung cancer of the model and the TPSS was 92.3%(96/104) compared to 94.2%(98/104) respectively (x2= 0.500, P= 0.5), while the specificity was 78.3%(47/60) more than 65.0%(39/60) (X2= 6.125, P= 0.008)ConclusionIn this paper, incorporating the patients’ clinical, CT and PET information to establish a mathematical model for the diagnosis of SPN, the results indicated the diagnostic efficacy of the model was superior to the traditional TPSS method for PET and had the potential for clinical application.Part Ⅱ Validation and comparison of the new SPN diagnostic model based on 18F-FDG PET/CT and clinical information and two existing SPN modelsObjectiveValidated the new SPN diagnostic model based on 18F-FDG PET/ CT and clinical information and two existing SPN models (Mayo model and Li model), and compared their diagnostic potency for better guiding clinical application. Materials and Method1. Research objects. Between June 2014 and December 2015,160 patients with an indeterminate SPN detected during normal clinical work or for other reason referred for PET/CT scanning were retrospectively identified from the database of our PET center.128 cases were included with pathological conclusion or clinical follow-up time was more than 2 years, excluding the 32 cases without pathological result or follow-up time was less than 2 years. All lesions were in line with SPN disease diagnostic criteria, and were solid nodules of diameters between 8mm and 30mm. All medical records were reviewed to obtain the following data:age, gender, smoking history (current or former cigarette smoker), disease history of malignancy, family history of cancer, pathological conclusion, and last date of clinical and radiologic follow-up (including disappearance of the SPN or decreased size, no growth, or growth).2.18F-FDG PET/CT scan. PET/CT scanning was performed with the Biograph mCTx scanner (Siemens, Germany).18F produced by PETtrace cyclotron (GE, USA), and 18F-FDG synthesized by chemical synthesis module automatically, radiochemical purity> 95%. Fasting more than 6 h, before injection of 18F-FDG, patients should be measured venous blood glucose levels to ensure that blood glucose levels (<7mmol/ L). Venous blood glucose levels should be controlled to lower than 7mmol/L in patients with diabetes. After intravenous injection of 18F-FDG 259～444 MBq in calm state, lying in the dark for 50～60 min, patients should urinate before PET/CT scanning. Whole body scan ranged from mid-thigh to the top of the head. Image reconstruction used the OSEM. The PET and CT images transferred to the workstation for image fusion. After the whole body PET/CT scanning, thin-section CT scanning on the same machine were performed in the nodule.3. Imaging analysis. PET/CT fusing images, PET images and CT images of same patient were analyzed frame by frame by the specific software of workstation. Two experienced nuclear medicine physicians reviewed PET scans and scored the intensity of 18F-FDG uptake using a two point scoring scale (0, no higher than mediastinal blood pool; 1, higher than mediastinal blood pool). Two radiologists visually confirmed lesions received from thin-section CT and record the location of the lesion, nodule diameter, border, and the presence of calcification, spiculated sign, lobulated sign, vessel convergence sign, pleural indentation sign and air bronchogram.4. Diagnostic basis. Solitary pulmonary nodule diagnosed based on histopathologic findings or clinical and radiologic follow-up. The time of follow-up was defined as the time between PET imaging and histologic diagnosis or the date of the last radiologic follow-up. Radiologic follow-up typically consisted of repeat chest CT scans. Lesions were classified as benign in case of benign pathologic findings, the disappearance of the lesion at radiologic follow-up, or decreased or no change in size within an observation period of at least 2 years. All malignant lesions were identified by pathology.5. Validation of the new model. Validated the new model by using 128 patients data of SPN.The new model is defined by the equations:P=1／(1+e-x):x=-5.124+0.058 x age+2.244 x nodule speculation+3.881 x TPSS.[e (the base of natural logarithms), age (years), nodule spiculation(l:exist nodule spiculation sign; 0:others), TPSS (1:higher than mediastinum; 0:others)]When P≥ 0.8475 diagnosed with lung cancer, the diagnostic sensitivity, specificity, positive predictive value, negative predictive value of the model were calculated. The diagnostic accuracy of the new model and the TPSS received the paired ROC curve analysis by using statistical software Medcalc. Performed McNemar test for comparing the sensitivity and specificity of these two methods by using SPSS 13.0 software. P<0.05 was considered statistically significant for all tests.6. Comparison of the new model and the two existing SPN models.The Mayo model is defined by the equations:Probability of malignancy= 1/(1+e-x);x=-6.8272+0.0391 × age+0.7917 x smoke+1.338 8× cancer+0.1274 x diameter+1.0407 × spiculation+0.7838 × upper;where e is the base of natural logarithms, age is the patient’s age in years, smoke = if the patient is a current or former smoker (otherwise= 0), cancer= if the patient has a history of an extrathoracic cancer that was diagnosed>5 years ago (otherwise = 0), diameter is the diameter of the nodule in millimeters, spiculation= if the edge of the nodule has spicules (otherwise= 0), and location= if the nodule is located in an upper lobe (otherwise= 0).The Li model is defined by the equations:P=1/(1+e-x);x=-4.496+0.07 × Age+0.676 × diameter+0.736 × spiculation+1.267× family history of cancer-1.615 × calcification-1.408 × border;where e is the natural logarithm, and 1 for yes and 0 for no in the last four elements (ie, spiculation, family cancer history, calcification and border). Used 0.463 as a diagnostic criterion.The diagnostic accuracy of three models (new model, Mayo model, Li model) received the paired ROC curve analysis by using statistical software Medcalc. Performed McNemar test in order to comparing the sensitivity and specificity of the new model and Li model by using SPSS 13.0 software. P< 0.05 was considered statistically significant for all tests. Results1. Clinical information of patients with SPN. In total,128 eligible patients with SPN were identified, of whom 81 proven to have malignant nodules and 47 have benign lesions. All patients with malignancy were diagnosed by histopathology. The diagnosis of a benign lesion was based on the histopathologic results in 38 patients and on the stabilization or spontaneous decrease in size of the lesion on a follow-up CT scan in 9 patients.2. Diagnostic potency of the TPSS for charactering SPN. The ROC-AUC value of the TPSS were 0.71 (95% CI:95% CI:0.62～0.79). Its diagnostic sensitivity and specificity of lung cancer was 92.6% (75/81) and 48.9%(23/47) respectively, while the positive predictive value and negative predictive value was 75.8%(75/99) and 79.3%(23/29).3. Validation of the new model. Validated the new model by using 128 patients data of SPN. Calculated the probability of malignancy and drew ROC curve, resulting AUC-ROC of the model was 0.86 (95% CI:0.79～0.92). Judged SPN to be malignant nodules when the probability of malignancy P> 0.8475. The diagnostic sensitivity and specificity of the new model were 90.1% (73/81) and 63.8%(30/47), positive predictive value and negative predictive value were 81.1%(73/90) and 78.9%(30/38).4. Diagnostic potency comparison between the new model and the TPSS. The ROC-AUC value of the model and the TPSS were 0.86 (95% CI:0.79～0.92) and 0.71 (95%CI:0.62-0.79). Diagnostic potency of the model outperformed the TPSS (z= 4.344, P< 0.0001). When the model taken optimum operating point P= 0.8475 (ie P > 0.8475 diagnosed with lung cancer), the diagnostic sensitivity of lung cancer of the model and the TPSS was 90.1%(73/81) compared to 92.6%(75/81) respectively (x2 = 0.500, P= 0.5), while the specificity was 63.8%(30/47) more than 48.9%(23/47) (x2= 5.143,P= 0.016).5. Diagnostic potency comparison of the new model with the Mayo model and the Li model. We validated the Mayo model and the Li model by means of calculating the probability of malignancy and then drawing ROC curves. The ROC-AUC value of the new model and the Mayo model was 0.86 (95% CI:0.79-0.91) and 0.74 (95% CI: 0.65-0.81) separately. Diagnostic potency of the new model outperformed the Mayo model (z= 2.875, P= 0.004). The accuracy between the new model and Li model was no significant difference (z= 1.687, P= 0.09) since their ROC-AUC value were 0.86 (95% CI:0.79-0.91) and 0.81 (95% CI:0.73-0.87). When the Li model taken optimum operating point P= 0.463 (ie P≥ 0.463 as a diagnostic cutoff of lung cancer) diagnostic sensitivity of the new model and Li model were 90.1%(73/81) and 88.9% (72/81) (x2= 0, P= 1), while the specificity were 63.8%(30/47) and 55.3%(26/47) respectively (%2= 0.75, P= 0.388), the difference between them was both not statistically significant.ConclusionThis study verified the new diagnostic SPN model based on 18F-FDG PET/ CT and clinical information and two existing SPN models, and then compared their diagnostic performance. The new model performed a better diagnostic potency than that of TPSS and Mayo model, but similar to Li model. It was simpler and facilitated to use, not only provided a new way for the clinical diagnosis of SPN, but also offered reference and inspiration for future research of SPN diagnostic model. |
PDF Full Text Request