This retrospective study was approved by the institutional review board and waived the requirement for written informed consent.
We retrospectively included a cohort of 328 patients who underwent 18F-FDG PET/CT in our institution with pathologically confirmed neuroblastoma between January 2018 and December 2019. The inclusion criteria consisted of (1) patients with neuroblastoma who underwent bone marrow aspirates or biopsies and were assessed BMI using morphologic criteria in conjunction with appropriate antibodies; (2) 18F-FDG PET/CT scan performed within 30 days before the bone marrow aspirates or biopsies. The exclusion criteria included the following: (1) patients who received tumor-related treatments such as chemotherapy, radiotherapy and surgical excision prior to 18F-FDG PET/CT examination; (2) patients with incomplete clinical data; (3) patients were greater than 18 years of age at diagnosis. Finally, a total of 133 patients (58 males and 75 females; median age, 3.2 years; range, 1.7–4.7 years) were retrospectively included in this study. According to the result of bone marrow aspirates or biopsies, there were 65 patients with BMI and 68 patients without BMI. The flow chart for patient selection is shown in Fig. 1.
Stratified sampling according to the BMI stratification was implemented to balance the positive and negative cases, and the final 133 cases were randomly divided into the training set and test set according to a ratio of 7:3, which resulted in 93 cases being divided into the training set and 40 cases being divided into the test set.
The baseline data of each patient were obtained by reviewing the medical records, which included the following aspects: (1) clinical information, (2) laboratory indicators, (3) PET metabolic parameters, and (4) pathological-related information. We defined these data as clinical characteristics. Laboratory indicators including neuron-specific enolase (NSE), serum ferritin, lactate dehydrogenase (LDH), urine vanillylmandelic acid (VMA) and homovanillic acid (HVA) were acquired within two weeks before therapy.
PET/CT imaging acquisition
All patients in the cohort underwent whole body 18F-FDG PET/CT (Biograph mCT-64 PET/CT; Siemens, Knoxville) scans according to European Association of Nuclear Medicine guidelines for tumor imaging [17, 18]. Patients were instructed to ban from intense exercises for at least 24 h before PET/CT scan and fast at least 6 h before 18F-FDG injection. A mean dose of 3 mCi (mean 0.14 mCi/kg) was administrated considering the patients are children. A low-dose CT scan (CT scanning parameters: tube voltage 120 keV, thickness 2 mm, matrix size 512 × 512) for viewing anatomic structures and attenuation correction was performed an hour after the injection. PET scan with three-dimension image mode and 2 min per bed setting followed immediately after CT acquisition. PET images were reconstructed with the time-of-flight ordered subsets-expectation maximization algorithm. All corrections for quantitative interpretation, including detector sensitivity normalization, dead time, random, scatter, attenuation and decay correction were applied during reconstruction. A Gaussian smoothing filter with a full width at half-maximum of 5 mm was applied to the PET images. The PET images’ parameters were as follows: pixel size 4.07 mm × 4.07 mm, 3 mm slice thickness, and matrix size 200 × 200.
Tumor segmentation, radiomics features extraction and selection
Image segmentation was performed semi-automatically with a commonly used open-source software (3D Slicer, Version 4.10.1) by reader 1 (W.W. with 7 years of experience in pediatric oncologic radiology). An example of ROI segmentation is shown in Fig. 2. The intraclass correlation coefficient (ICC) was used to assess the reproducibility of the selected features. A total of 30 cases (15 with BMI and 15 without BMI) of CT images and PET images randomly selected from the whole cohort were independently performed repeat segmentation by reader 1 and reader 2 (Y.K. with 10 years of experience in pediatric oncologic radiology). The readers were blinded to the clinical information when performing the segmentation.
Radiomics features were extracted from both CT and PET images using Pyradiomics in Python (version 3.7.8), an open-source python package for the extraction of radiomics features from medical imaging. A fixed bin width (0.3 standardized uptake value for PET image and 25 Hounsfield Units [HUs] for CT image) had been chosen to discretize gray value discretization for texture features extraction [19, 20]. Furthermore, filters including wavelet, square, and logarithm et al. were applied to the original CT and PET images for calculating high-dimensional features.
The features with ICC > 0.8 were considered reliable and maintained for subsequent analysis [21,22,23]. Then, the Pearson’s correction coefficients and Spearman’s rank correlation coefficient were calculated to examine redundant and collinear features, and features with mutual correlation coefficients > 0.9 were removed . Finally, the least absolute shrinkage and selection operator (LASSO) regression with fivefold cross-validation was applied for feature selection.
Radiomics model, clinical model and radiomics nomogram construction and evaluation
Using the most optimal features to construct the radiomics signature. The radiomics score (Rad score) was calculated for each patient via the combination of the selected features with their respective weight coefficients.
The univariate logistic regression analysis was used to assess the difference in clinical characteristics between BMI and without BMI in the training set. Then, variables with p < 0.05 in the univariate logistic regression analysis were applied to multivariate logistic regression analysis to elucidate the independent clinical risk factors. Meanwhile, multivariate logistic regression analysis was applied to build the clinical model was built based on the independent clinical risk factors.
The clinical-radiomics model, which incorporated the independent clinical risk factors and Rad score, was constructed using multivariable logistic regression analysis and finally presented as a radiomics nomogram in the training set. Logistic regression is a classical statistical model that internally has a linear regression, which is topped up by a sigmoid function such that the output of the model is a probability estimate between 0 and 1 .
The predictive performance of the clinical-radiomics model was evaluated by receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). A calibration curve, obtained by plotting the actual probability against the nomogram-predicted probability, was used to evaluate the calibration of the nomogram. DCA was employed to evaluate the clinical utility of the radiomics nomogram.
Moreover, we applied two other machine learning methods, naive bayes and neural network, to build clinical-radiomics models, and then compared the predictive performance of the models obtained by each machine learning method based on such matrices as the area under the curve (AUC), accuracy, precision, F1-score and recall.
Categorical variables were expressed as counts (n) and percentages (%), while continuous variables were expressed as mean values ± standard deviation or medians with the interquartile ranges. Statistical analyses were performed using R (version 4.0.3) and IMB SPSS Statistics (version 26.0). Two-sided with p < 0.05 was considered statistically significant. Univariate analysis was used to compare differences in the clinical factors between the training and test sets, using the independent t-test or Mann–Whitney U test for quantitative data, and the chi-squared test for categorical variables. Clinical independent predictors were screened using univariate and multivariate logistic regression analysis. The DeLong test was used to compare the AUC values of different models. The nomogram and calibration curve were depicted using the “rms (R)” package. DCA was performed using the “rmda (R)” package.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
This article is autogenerated using RSS feeds and has not been created or edited by OA JF.
Click here for Source link (https://www.springeropen.com/)