ABSTRACT
Objective:To explore the value of different machine learning models based on Gd-EOB-DTPA enhanced MRI hepatobiliary phase radiomics features in preoperative prediction of microvascular invasion (MVI) of hepatocellular carcinoma (HCC).Methods:The data of 132 patients with HCC confirmed by pathology in the First Affiliated Hospital of Soochow University from January 2015 to May 2020 were retrospectively analyzed, including 72 cases of positive MVI and 60 cases of negative MVI. According to the proportion of 7∶3, the cases were randomly divided into training set and validation set. The radiomics features of hepatobiliary phase images for HCC were extracted by PyRadiomics software. The clinical and radiomics features of the training set were screened by the least absolute shrinkage and selection operator (LASSO) regression with 5 fold cross-validation, and then the optimal feature subset was obtained. Six machine learning algorithms, including decision tree, extreme gradient boosting, random forest, support vector machine (SVM), generalized linear model (GLM) and neural network, were used to build the prediction models, and the ROC curves were used to evaluate the prediction ability of the models. DeLong test was used to compare the differences of area under the curve (AUC) for 6 machine learning algorithms.Results:Totally 14 features selected by LASSO regression were obtained to form the optimal feature subset, including 2 clinical features (maximum tumor diameter and alpha-fetoprotein) and 12 radiomics features. The AUCs of decision tree, extreme gradient boosting, random forest, SVM, GLM and neural network based on the optimal feature subset were 0.969, 1.000, 1.000, 0.991, 0.966, 1.000 in the training set and 0.781, 0.890, 0.920, 0.806, 0.684, 0.703 in the validation set, respectively. There were significant differences in the AUCs between extreme gradient boosting and GLM or neural network ( Z=2.857, 3.220, P=0.004, 0.001). The differences in AUCs between random forest and SVM, GLM, or neural network were significant ( Z=2.371, 3.190, 3.967, P=0.018, 0.001,<0.001). The difference in AUCs between SVM and GLM was statistically significant ( Z=2.621 , P=0.009). There were no significant differences in the AUCs among the other machine learning models ( P>0.05). Conclusion:Machine learning models based on Gd-EOB-DTPA enhanced MRI hepatobiliary phase radiomics features can be used to preoperatively predict MVI of HCC, particularly the extreme gradient boosting and random forest models have high prediction efficiency.
ABSTRACT
Objective:To explore the value of spectral CT radiomics quantitative features on differentiating lung cancer nodule from inflammatory nodule.Methods:The spectral CT imaging data of 96 lung cancer nodules and 45 inflammatory nodules from the First Affiliated Hospital of Soochow University were analyzed retrospectively. According to a ratio of two to one, patients were randomly assigned to the training group and validation group, including 64 lung cancer nodules and 30 inflammatory nodules in the training group, 32 lung cancer nodules and 15 inflammatory nodules in the validation group. MaZda software was used for radiomic feature extraction from the 70 keV monochromatic images in arterial phase and venous phase for lung cancer nodules and inflammatory nodules in the training group. Fisher coefficients (Fisher), classification error probability combined average correlation coefficients (POE+ACC) and mutual information (MI) were used to select 10 optimal features for the optimal feature subsets. The optimal feature subsets were analyzed by using linear discriminant analysis (LDA) and nonlinear discriminant analysis (NDA) to calculate the area under the receiver operating characteristic (ROC) curve (AUC), accuracy, sensitivity, specificity, precise and F1 score in differentiating lung cancer nodule from inflammatory nodule. The prediction model was established using the optimal feature subsets in the training group with artificial neural network (ANN). Then the established prediction model was used to differentiate lung cancer nodule from inflammatory nodule in the validation group. Delong test was used to compare the differences in the AUC of different optimal feature subsets.Results:In arterial phase, the optimal feature subset obtained from MI-NDA had the highest AUC of 0.888 [95% confidence interval (CI) 0.806-0.943], accuracy rate of 88.3%, sensitivity of 87.5% and specificity of 90.0%, on the differential diagnosis of lung cancer nodule and inflammatory nodule in the training group. There was no significant difference in AUC between MI-NDA and Fisher-NDA or (POE+ACC)-NDA method ( Z=1.941, P=0.052; Z=1.683, P=0.092). In venous phase, the optimal feature subset obtained from (POE+ACC)-NDA had the highest AUC of 0.846 (95%CI 0.757-0.912), accuracy rate of 87.2%, sensitivity of 92.2% and specificity of 76.7%, on the differential diagnosis of lung cancer nodule and inflammatory nodule in the training group. There was no significant difference in AUC between(POE+ACC)-NDA and MI-NDA method ( Z=1.354, P=0.18), but significant difference between (POE+ACC)-NDA and Fisher-NDA method ( Z=2.423, P=0.015). In the validation group and training group, the optimal feature subset selected by MI-NDA method had the highest AUC of 0.888(95%CI 0.806-0.943) and 0.871(95%CI 0.741-0.951). Conclusion:Spectral CT radiomics quantitative features have great value on the differential diagnosis of lung cancer nodule and inflammatory nodule.
ABSTRACT
Objective To explore the application value of spectral CT quantitative analysis in differentiating adenocarcinoma or squamous carcinoma from inflammatory myofibroblastic tumor (IMT). Methods A total of 115 patients with 62 adenocarcinomas, 33 squamous carcinomas and 20 IMTs underwent spectral CT scans to obtain spectral images at arterial phase (AP) and venous phase (VP). The imaging data were analyzed retrospectively. The iodine concentration of adenocarcinoma, squamous carcinomas and IMT were measured. The normalized iodine concentration in AP (NICAP), normalized iodine concentration in VP (NICVP) and normalized iodine concentration difference between AP and VP (ICD) were calculated. The above quantitative parameters among three groups were analyzed with analysis of variance and ROC curve. Results NICAP (0.15 ± 0.04), NICVP (0.37 ± 0.08) and ICD(0.23 ± 0.06)of the adenocarcinoma were lower than those of IMT (0.21 ± 0.05,0.50 ± 0.06,0.28 ± 0.08). There were significant differences in NICAP, NICVP and ICD between adenocarcinoma and IMT (P<0.05). NICAP (0.13 ± 0.03), NICVP (0.35±0.06) and ICD (0.22±0.05) of the squamous carcinoma were lower than those of IMT (0.21± 0.05,0.50±0.06,0.28±0.08). The differences in NICAP, NICVP and ICD were significant between squamous carcinoma and IMT (P<0.05). There were no significant differences in NICAP, NICVP and ICD between adenocarcinoma and squamous carcinoma (P>0.05). The best spectral quantitative parameter for differentiating the adenocarcinoma from IMT was NICVP, which yielded a sensitivity of 92.3% and a specificity of 86.7%with the threshold of 0.425. NICVP was also the best spectral quantitative parameter for differentiating squamous carcinomas from IMT. With the threshold of 0.44, a sensitivity of 84.6% and a specificity of 92.3% were found. Conclusion Spectral CT imaging with the quantitative iodine concentration analysis may help to increase the accuracy of differentiating adenocarcinoma and squamous carcinoma from IMT.
ABSTRACT
Objective To explore the application value of spectral CT quantitative analysis in differentiating adenocarcinoma or squamous carcinoma from inflammatory myofibroblastic tumor (IMT). Methods A total of 115 patients with 62 adenocarcinomas, 33 squamous carcinomas and 20 IMTs underwent spectral CT scans to obtain spectral images at arterial phase (AP) and venous phase (VP). The imaging data were analyzed retrospectively. The iodine concentration of adenocarcinoma, squamous carcinomas and IMT were measured. The normalized iodine concentration in AP (NICAP), normalized iodine concentration in VP (NICVP) and normalized iodine concentration difference between AP and VP (ICD) were calculated. The above quantitative parameters among three groups were analyzed with analysis of variance and ROC curve. Results NICAP (0.15 ± 0.04), NICVP (0.37 ± 0.08) and ICD(0.23 ± 0.06)of the adenocarcinoma were lower than those of IMT (0.21 ± 0.05,0.50 ± 0.06,0.28 ± 0.08). There were significant differences in NICAP, NICVP and ICD between adenocarcinoma and IMT (P<0.05). NICAP (0.13 ± 0.03), NICVP (0.35±0.06) and ICD (0.22±0.05) of the squamous carcinoma were lower than those of IMT (0.21± 0.05,0.50±0.06,0.28±0.08). The differences in NICAP, NICVP and ICD were significant between squamous carcinoma and IMT (P<0.05). There were no significant differences in NICAP, NICVP and ICD between adenocarcinoma and squamous carcinoma (P>0.05). The best spectral quantitative parameter for differentiating the adenocarcinoma from IMT was NICVP, which yielded a sensitivity of 92.3% and a specificity of 86.7%with the threshold of 0.425. NICVP was also the best spectral quantitative parameter for differentiating squamous carcinomas from IMT. With the threshold of 0.44, a sensitivity of 84.6% and a specificity of 92.3% were found. Conclusion Spectral CT imaging with the quantitative iodine concentration analysis may help to increase the accuracy of differentiating adenocarcinoma and squamous carcinoma from IMT.