Radiomics model based on CT images for distinguishing invasive lung adenocarcinoma with micropapillary or solid structure / 中国胸心血管外科临床杂志

@#Objective    To investigate the radiomics features to distinguish invasive lung adenocarcinoma with micropapillary or solid structure. Methods    A retrospective analysis was conducted on patients who received surgeries and pathologically confirmed invasive lung adenocarcinoma in our hospital from April 2016 to August 2019. The dataset was randomly divided into a training set [including a micropapillary/solid structure positive group (positive group) and a micropapillary/solid structure negative group (negative group)] and a testing set (including a positive group and a negative group) with a ratio of 7∶3. Two radiologists drew regions of interest on preoperative high-resolution CT images to extract radiomics features. Before analysis, the intraclass correlation coefficient was used to determine the stable features, and the training set data were balanced using synthetic minority oversampling technique. After mean normalization processing, further radiomics features selection was conducted using the least absolute shrinkage and selection operator algorithm, and a 5-fold cross validation was performed. Receiver operating characteristic (ROC) curves were depicted on the training and testing sets to evaluate the diagnostic performance of the radiomics model. Results    A total of 340 patients were enrolled, including 178 males and 162 females with an average age of 60.31±6.69 years. There were 238 patients in the training set, including 120 patients in the positive group and 118 patients in the negative group. There were 102 patients in the testing set, including 52 patients in the positive group and 50 patients in the negative group. The radiomics model contained 107 features, with the final 2 features selected for the radiomics model, that is, Original_ glszm_ SizeZoneNonUniformityNormalized and Original_ shape_ SurfaceVolumeRatio. The areas under the ROC curve of the training and the testing sets of the radiomics model were 0.863 (95%CI 0.815-0.912) and 0.857 (95%CI 0.783-0.932), respectively. The sensitivity was 91.7% and 73.7%, the specificity was 78.8% and 84.0%, and the accuracy was 85.3% and 78.4%, respectively. Conclusion    There are differences in radiomics features between invasive pulmonary adenocarcinoma with or without micropapillary and solid structures, and the radiomics model is demonstrated to be with good diagnostic value.

Construction of a machine learning model for identifying clinical high-risk carotid plaques based on radiomics / 中国胸心血管外科临床杂志

@#Objective     To construct a radiomics model for identifying clinical high-risk carotid plaques. Methods     A retrospective analysis was conducted on patients with carotid artery stenosis in China-Japan Friendship Hospital from December 2016 to June 2022. The patients were classified as a clinical high-risk carotid plaque group and a clinical low-risk carotid plaque group according to the occurrence of stroke, transient ischemic attack and other cerebrovascular clinical symptoms within six months. Six machine learning models including eXtreme Gradient Boosting, support vector machine, Gaussian Naive Bayesian, logical regression, K-nearest neighbors and artificial neural network were established. We also constructed a joint predictive model combined with logistic regression analysis of clinical risk factors. Results    Finally 652 patients were collected, including 427 males and 225 females, with an average age of 68.2 years. The results showed that the prediction ability of eXtreme Gradient Boosting was the best among the six machine learning models, and the area under the curve (AUC) in validation dataset was 0.751. At the same time, the AUC of eXtreme Gradient Boosting joint prediction model established by clinical data and carotid artery imaging data validation dataset was 0.823. Conclusion     Radiomics features combined with clinical feature model can effectively identify clinical high-risk carotid plaques.

Magnetic resonance imaging texture analysis of the temporomandibular joint for changes in the articular disc in individuals with migraine headache / Análise de textura por ressonância magnética da articulação temporomanbibular com alterações no disco articular em individuos com cefaleia migranea

Objective: the aim of this study was to analyse the performance of the technique of texture analysis (TA) with magnetic resonance imaging (MRI) scans of temporomandibular joints (TMJs) as a tool for identification of possible changes in individuals with migraine headache (MH) by relating the findings to the presence of internal derangements. Material and Methods: thirty MRI scans of the TMJ were selected for study, of which 15 were from individuals without MH or any other type of headache (control group) and 15 from those diagnosed with migraine. T2-weighted MRI scans of the articular joints taken in closed-mouth position were used for TA. The co-occurrence matrix was used to calculate the texture parameters. Fisher's exact test was used to compare the groups for gender, disc function and disc position, whereas Mann-Whitney's test was used for other parameters. The relationship of TA with disc position and function was assessed by using logistic regression adjusted for side and group. Results: the results indicated that the MRI texture analysis of articular discs in individuals with migraine headache has the potential to determine the behaviour of disc derangements, in which high values of contrast, low values of entropy and their correlation can correspond to displacements and tendency for non-reduction of the disc in these individuals. Conclusion: the TA of articular discs in individuals with MH has the potential to determine the behaviour of disc derangements based on high values of contrast and low values of entropy (AU)

Objetivo: o objetivo deste estudo foi analisar o desempenho da técnica de análise de textura (AT) em exames de ressonância magnética (RM) das articulações temporomandibulares (ATM) como ferramenta para identificação de possíveis alterações em indivíduos com cefaléia migrânea (CM) relacionando os achados com a presença de desarranjos internos. Material e Métodos: trinta exames de RM das ATM foram selecionados para estudo, sendo 15 de indivíduos sem cefaleia migrânea ou qualquer outro tipo de cefaléia (grupo controle) e 15 diagnosticados com CM. As imagens de RM ponderadas em T2 das articulações realizadas na posição de boca fechada foram usadas para AT. A matriz de co-ocorrência foi usada para calcular os parâmetros de textura. O teste exato de Fisher foi usado para comparar os grupos quanto ao sexo, função do disco e posição do disco, enquanto o teste de Mann-Whitney foi usado para os demais parâmetros. A relação da AT com a posição e função do disco foi avaliada por meio de regressão logística ajustada para lado e grupo. Resultados: a AT por RM dos discos articulares em indivíduos com cefaleia migrânea tem o potencial de determinar o comportamento dos desarranjos discais, em que altos valores de contraste, baixos valores de entropia e sua correlação podem corresponder a deslocamentos e tendência a não redução do disco nesses indivíduos. Conclusão: a análise de textura dos discos articulares em indivíduos com CM tem potencial para determinar o comportamento dos desarranjos do disco com base em altos valores de contraste e baixos valores de entropia. (AU)

Based on CT radiomics model for predicting the response to first-line chemotherapy of diffuse large B-cell lymphoma / 中华肿瘤杂志

Objective: To investigate the potential value of CT Radiomics model in predicting the response to first-line chemotherapy in diffuse large B-cell lymphoma (DLBCL). Methods: Pre-treatment CT images and clinical data of DLBCL patients treated at Shanxi Cancer Hospital from January 2013 to May 2018 were retrospectively analyzed and divided into refractory patients (73 cases) and non-refractory patients (57 cases) according to the Lugano 2014 efficacy evaluation criteria. The least absolute shrinkage and selection operator (LASSO) regression algorithm, univariate and multivariate logistic regression analyses were used to screen out clinical factors and CT radiomics features associated with efficacy response, followed by radiomics model and nomogram model. Receiver operating characteristic (ROC) curve, calibration curve and clinical decision curve were used to evaluate the models in terms of the diagnostic efficacy, calibration and clinical value in predicting chemotherapy response. Results: Based on pre-chemotherapy CT images, 850 CT texture features were extracted from each patient, and 6 features highly correlated with the first-line chemotherapy effect of DLBCL were selected, including 1 first order feature, 1 gray level co-occurence matrix, 3 grey level dependence matrix, 1 neighboring grey tone difference matrix. Then, the corresponding radiomics model was established, whose ROC curves showed AUC values of 0.82 (95% CI: 0.76-0.89) and 0.73 (95% CI: 0.60-0.86) in the training and validation groups, respectively. The nomogram model, built by combining validated clinical factors (Ann Arbor stage, serum LDH level) and CT radiomics features, showed an AUC of 0.95 (95% CI: 0.90-0.99) and 0.91 (95% CI: 0.82-1.00) in the training group and the validation group, respectively, with significantly better diagnostic efficacy than that of the radiomics model. In addition, the calibration curve and clinical decision curve showed that the nomogram model had good consistency and high clinical value in the assessment of DLBCL efficacy. Conclusion: The nomogram model based on clinical factors and radiomics features shows potential clinical value in predicting the response to first-line chemotherapy of DLBCL patients.

Radiomics-based prediction of gamma pass rates for different intensity-modulated radiation therapy techniques for pelvic tumors / 中华放射医学与防护杂志

Objective:To explore the feasibility of a classification prediction model for gamma pass rates (GPRs) under different intensity-modulated radiation therapy techniques for pelvic tumors using a radiomics-based machine learning approach, and compare the classification performance of four integrated tree models.Methods:With a retrospective collection of 409 plans using different IMRT techniques, the three-dimensional dose validation results were adopted based on modality measurements, with a GPR criterion of 3%/2 mm and 10% dose threshold. Then prediction were built models by extracting radiomics features based on dose documentation. Four machine learning algorithms were used, namely random forest (RF), adaptive boosting (AdaBoost), extreme gradient boosting (XGBoost), and light gradient boosting machine (LightGBM). Their classification performance was evaluated by calculating sensitivity, specificity, F1 score, and AUC value. Results:The RF, AdaBoost, XGBoost, and LightGBM models had sensitivities of 0.96, 0.82, 0.93, and 0.89, specificities of 0.38, 0.54, 0.62, and 0.62, F1 scores of 0.86, 0.81, 0.88, and 0.86, and AUC values of 0.81, 0.77, 0.85, and 0.83, respectively. XGBoost model showed the highest sensitivity, specificity, F1 score, and AUC value, outperforming the other three models. Conclusions:To build a GPR classification prediction model using a radiomics-based machine learning approach is feasible for plans using different intensity-modulated radiotherapy techniques for pelvic tumors, providing a basis for future multi-institutional collaborative research on GPR prediction.

Prediction of EGFR mutant subtypes in patients with non-small cell lung cancer by pre-treatment CT radiomics and machine learning / 中华放射医学与防护杂志

Objective:To evaluate the feasibility and clinical value of pre-treatment non-enhanced chest CT radiomics features and machine learning algorithm to predict the mutation status and subtype (19Del/21L858R) of epidermal growth factor receptor (EGFR) for patients with non-small cell lung cancer (NSCLC).Methods:This retrospective study enrolled 280 NSCLC patients from first and second affiliated hospital of University of South China who were confirmed by biopsy pathology, gene examination, and have pre-treatment non-enhanced CT scans. There are 136 patients were confirmed EGFR mutation. Primary lung gross tumor volume was contoured by two experienced radiologists and oncologists, and 851 radiomics features were subsequently extracted. Then, spearman correlation analysis and RELIEFF algorithm were used to screen predictive features. The two hospitals were training and validation cohort, respectively. Clinical-radiomics model was constructed using selected radiomics and clinical features, and compared with models built by radiomics features or clinical features respectively. In this study, machine learning models were established using support vector machine (SVM) and a sequential modeling procedure to predict the mutation status and subtype of EGFR. The area under receiver operating curve (AUC-ROC) was employed to evaluate the performances of established models.Results:After feature selection, 21 radiomics features were found to be efffective in predicting EGFR mutation status and subtype and were used to establish radiomics models. Three types models were established, including clinical model, radiomics model, and clinical-radiomics model. The clinical-radiomics model showed the best predictive efficacy, AUCs of predicting EGFR mutation status for training dataset and validation dataset were 0.956 (95% CI: 0.952-1.000) and 0.961 (95% CI: 0.924-0.998), respectively. The AUCs of predicting 19Del/L858R mutation subtype for training dataset and validation dataset were 0.926 (95% CI: 0.893-0.959), 0.938 (95% CI: 0.876-1.000), respectively. Conclusions:The constructed sequential models based on integration of CT radiomics, clinical features and machine learning can accurately predict the mutation status and subtype of EGFR.

Prediction of radiomics-based machine learning in dose verification of intensity-modulated pelvic radiotherapy / 中华放射医学与防护杂志

Objective:Based on radiomics characteristics, different machine learning classification models are constructed to predict the gamma pass rate of dose verification in intensity-modulated radiotherapy for pelvic tumors, and to explore the best prediction model.Methods:The results of three-dimensional dose verification based on phantom measurement were retrospectively analyzed in 196 patients with pelvic tumor intensity-modulated radiotherapy plans. The gamma pass rate standard was 3%/2 mm and 10% dose threshold. Prediction models were constructed by extracting radiomic features based on dose documentation. Four machine learning algorithms, random forest, support vector machine, adaptive boosting, and gradient boosting decision tree were used to calculate the AUC value, sensitivity, and specificity respectively. The classification performance of the four prediction models was evaluated.Results:The sensitivity and specificity of the random forest, support vector machine, adaptive boosting, and gradient boosting decision tree models were 0.93, 0.85, 0.93, 0.96, 0.38, 0.69, 0.46, and 0.46, respectively. The AUC values were 0.81 and 0.82 for the random forest and adaptive boosting models, respectively, and 0.87 for the support vector machine and gradient boosting decision tree models.Conclusions:Machine learning method based on radiomics can be used to construct a prediction model of gamma pass rate for specific dosimetric verification of pelvic intensity-modulated radiotherapy. The classification performance of the support vector machine model and gradient boosting decision tree model is better than that of the random forest model and adaptive boosting model.

Preoperative prediction of vessel invasion in locally advanced gastric cancer based on venous phase enhanced CT radiomics and machine learning / 中华放射学杂志

Objective:To evaluate the value of preoperative prediction of vessel invasion (VI) of locally advanced gastric cancer by machine learning model based on the venous phase enhanced CT radiomics features.Methods:A retrospective analysis of 296 patients with locally advanced gastric cancer confirmed by pathology in the First Affiliated Hospital of Zhengzhou University from July 2011 to December 2020 was performed. The patients were divided into VI positive group ( n=213) and VI negative group ( n=83) based on pathological results. The data were divided into training set ( n=207) and test set ( n=89) according to the ratio of 7∶3 with stratification sampling. The clinical characteristics of patients were recorded, and the independent risk factors of gastric cancer VI were screened by multivariate logistic regression. Pyradiomics software was used to extract radiomic features from the venous phase enhanced CT images, and the minimum absolute shrinkage and selection algorithm (LASSO) was used to screen the features, obtain the optimal feature subset, and establish the radiomics signature. Four machine learning algorithms, including extreme gradient boosting (XGBoost), logistic, naive Bayes (GNB), and support vector machine (SVM) models, were used to build prediction models for the radiomics signature and the screened clinical independent risk factors. The efficacy of the model in predicting gastric cancer VI was evaluated by the receiver operating characteristic curve. Results:The degree of differentiation (OR=13.651, 95%CI 7.265-25.650, P=0.003), Lauren′s classification (OR=1.349, 95%CI 1.011-1.799, P=0.042) and CA199 (OR=1.796, 95%CI 1.406-2.186, P=0.044) were independent risk factors for predicting the VI of locally advanced gastric cancer. Based on the venous phase enhanced CT images, 864 quantitative features were extracted, and 18 best constructed radiomics signature were selected by LASSO. In the training set, the area under the curve (AUC) of XGBoost, logistic, GNB and SVM models for predicting gastric cancer VI were 0.914 (95%CI 0.875-0.953), 0.897 (95%CI 0.853-0.940), 0.880 (95%CI 0.832-0.928) and 0.814 (95%CI 0.755-0.873), respectively, and in the test set were 0.870 (95%CI 0.769-0.971), 0.877 (95%CI 0.788-0.964), 0.859 (95%CI 0.755-0.961) and 0.773 (95%CI 0.647-0.898). The logistic model had the largest AUC in the test set. Conclusions:The machine learning model based on the venous phase enhanced CT radiomics features has high efficacy in predicting the VI of locally advanced gastric cancer before the operation, and the logistic model demonstrates the best diagnostic efficacy.

The value of radiomics based on contrast-enhanced spectral mammography of internal and peripheral regions combined with clinical factors in predicting benign and malignant breast lesions of breast imaging reporting and data system category 4 / 中华放射学杂志

Objective:To evaluate the value of radiomics based on contrast-enhanced spectral mammography (CESM) of internal and peripheral regions combined with clinical factors in predicting benign and malignant breast lesions of breast imaging reporting and data system category 4 (BI-RADS 4).Methods:A retrospective analysis was performed on the clinical and imaging data of patients with breast lesions who were treated in Yantai Yuhuangding Hospital (Center 1) Affiliated to Qingdao University from July 2017 to July 2020 and in Fudan University Cancer Hospital (Center 2) from June 2019 to July 2020. Center 1 included 835 patients, all female, aged 17-80 (49±12) years, divided into training set (667 cases) and test set (168 cases) according to the "train-test-split" function in Python software at a ratio of 8∶2; and 49 patients were included from Center 2 as external validation set, all female, aged 34-70 (51±8) years. The radiomics features were extracted from the intralesional region (ITR), the perilesional regions of 5, 10 mm (PTR 5 mm, PTR10 mm) and the intra-and perilesional regions of 5, 10 mm (IPTR 5 mm, IPTR 10 mm) and were selected by variance filtering, SelectKBest algorithm, and least absolute shrinkage and selection operator. Then five radiomics signatures were constructed including ITR signature, PTR 5 mm signature, PTR 10 mm signature, IPTR 5 mm signature, IPTR 10 mm signature. In the training set, univariable and multivariable logistic regressions were used to construct nomograms by selecting radiomics signatures and clinical factors with significant difference between benign and malignant BI-RADS type 4 breast lesions. The efficacy of nomogram in predicting benign and malignant BI-RADS 4 breast lesions was evaluated by the receiver operating characteristic curve and area under the curve (AUC). Decision curve and calibration curve were used to evaluate the net benefit and calibration capability of the nomogram.Results:The nomogram included ITR signature, PTR 5 mm signature, PTR 10 mm signature, IPTR 5 mm signature, age, and BI-RADS category 4 subclassification for differentiating malignant and benign BI-RADS category 4 breast lesions and obtained AUCs of 0.94, 0.92, and 0.95 in the training set, test set, and external validation set, respectively. The calibration curve showed good agreement between the predicted probabilities and actual results and the decision curve indicated a good net benefit of the nomogram for predicting malignant BI-RADS 4 lesions in the training set, test set, and external validation set.Conclusion:The nomogram constructed from the radiomics features of the internal and surrounding regions of CESM breast lesions combined with clinical factors is attributed to differentiate benign from malignant BI-RADS category 4 breast lesions.

Radiomics based on three-dimensional high-resolution MR vessel wall imaging for identification of culprit plaques in symptomatic patients with middle cerebral artery atherosclerosis / 中华放射学杂志

Objective:To investigate the value of radiomics based on three-dimensional high resolution MR vessel wall imaging (3D HRMR-VWI) for identifying culprit plaques in symptomatic patients with middle cerebral atherosclerosis.Methods:The clinical and imaging features of 117 patients (139 middle cerebral artery plaques) with cerebrovascular diseases in Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology from October 2018 to October 2020 were respectively reviewed. Stratified random sampling was used to divide 139 plaques into training set (97 plaques) and validation set (42 plaque) at the ratio of 7∶3. The plaques were divided into 69 culprit plaques and 70 non-culprit plaques based on plaque MR features and clinical symptoms. The clinical and imaging characteristics of culprit plaques and non-culprit plaques were compared by independent sample t-test, Mann-Whitney U test and χ 2 test, and factors with significant difference between two groups in univariate analysis were further analyzed by multivariate logistic regression to find out the independent predictors of culprit plaques. Radiomics features were extracted, screened and radiomics model was constructed using pre-and post-contrast 3D HRMR-VWI based on the training set. The combined model was constructed by combining the independent predictors and radiomics model. Receiver operating characteristic curve and area under curve (AUC) were used to evaluate the efficacy of each model, and DeLong test was used to compare the efficacy of different models. Results:Significant difference was found in intraplaque hemorrhage, lumen area of stenosis, stenosis diameter, stenosis rate, plaque burden and enhancement rate between culprit and non-culprit plaques (all P<0.05). Multivariate logistic regression analysis confirmed that only intraplaque hemorrhage was the independent predictor for culprit plaques (OR=7.045,95%CI 1.402-35.397, P=0.018). In the validation set, the AUC of the pre-contrast 3D HRMR-VWI model was lower than that of the post-contrast 3D HRMR-VWI model ( Z=-2.01, P=0.044). The AUC of pre+post-contrast 3D HRMR-VWI model was not significantly different from that of post-contrast 3D HRMR-VWI model ( Z=0.79, P=0.427). The AUC showed no significant difference between combined model and pre+post-contrast 3D HRMR-VWI model ( Z=-0.59, P>0.05). The combined model showed the best performance in predicting culprit plaques of middle cerebral artery (AUC=0.939), with the sensitivity, specificity and accuracy of 95.24%, 76.19% and 85.71%. Conclusion:Radiomics based on 3D HRMR-VWI has potential values in identifying culprit plaques in symptomatic patients with middle cerebral atherosclerosis.

Preoperative prediction of lymphatic matastasis of mesolow colorectal cancer by endorectal ultrasound and elastography-based radiomics model / 中华超声影像学杂志

Objective:To investigate whether radiomics based on ultrasound images can predict lym-phatic metastasis of rectal cancer before surgery.Methods:A total of 80 patients with rectal cancer who underwent endorectal ultrasound (TRUS) and endorectal elastography were confirmed by postoperative pathology in Zhejiang Cancer Hospital from January 2016 to December 2019 were retrospectively analyzed. The general characteristics (gender, age, tumor size, depth of tumor infiltration, tumor location, carcinoembryonic antigen, glycoantigen 199) of the lymph node metastasis group ( n=27) and the non-metastasis group ( n=53) were compared, and the clinical risk factors with statistically significant differences were screened out. The tumor maximum sagittal 2D TRUS images and endorectal elastography were manually outlined, and the radiomics features were extracted using the open source software pyradiomics 3.0.1, and the filtering and embedding methods were used to reduce the dimensionality of the data to select the important features and obtain the best parameters of the model. Then all samples were randomly divided into training and validation sets in the ratio of 8∶2, the models were trained using the best model parameters, which were tested and validated in the validation set, and the predictive efficacy of different models was evaluated according to the ROC curve. Results:The depth of tumor infiltration was statistically significant in predicting whether the lymph nodes metastasized or not (χ 2=11.555, P<0.05), and its area under ROC curve(AUC) value was 0.699. A total of 1 710 features were extracted from sagittal 2D TRUS images and endorectal elastography. After pre-processing and screening, 10 features were strongly correlated with lymph node metastasis status. The 10 features were used to construct the prediction models with AUC values of 0.703, 0.726 and 0.742 for the Logistic Regression Model, Random Forest Model and Support Vector Machine Model, respectively. And the AUC value of the ensemble averaging model in the validation set was 0.734. The imaging-omics prediction model outperformed the prediction model based on statistical analysis of clinical data (AUC: 0.734 vs 0.699, Z=1.984), with a statistically significant difference ( P<0.05). Conclusions:The endorectal ultrasound and endorectal elastography-based radiomics model constructed in this study is better than the model constructed based on statistical analysis of clinical data only, and it is valuable for preoperative lymph node metastasis prediction in rectal cancer.

Value of predictive liver metastasis in pancreatic neuroendocrine neoplasms based on ultrasonographic radiomics / 中华超声影像学杂志

Objective:To explore the predictive value of ultrasound-based radiomics for liver metastasis in pancreatic neuroendocrine tumors (pNEN).Methods:A retrospective analysis was conducted on clinical, pathological, and ultrasound data of 269 pNEN patients confirmed by pathology at Tianjin Medical University Cancer Institute and Hospital from January 2012 to June 2022, including 94 patients with liver metastasis and 175 without liver metastasis. The regions of interest (ROI) were delineated on the maximum diameter section of the tumor using ITKSNAP software, and radiomics features were extracted using Pyradiomics. Radiomics features with an intra-group correlation coefficient greater than 0.90 were retained, and the optimal features were selected using the maximum relevance minimum redundancy (MRMR) algorithm. The dataset was randomly divided into a training set and a validation set in a ratio of 7∶3, and the random forest algorithm (Rfs) was used to predict pNEN liver metastasis. Three models were constructed, including the clinical ultrasound model, the radiomics model, and the comprehensive model that combined clinical ultrasound and radiomics features. The predictive performance of different models for pNEN liver metastasis was analyzed using the ROC curve, and the predictive performance of different models was compared using the Delong test.Results:A total of 874 features were extracted from the ROI, and 12 highly robust radiomics features were retained for model construction based on inter- and intra-observer correlation grading and feature selection. The area under curve(AUC), sensitivity, specificity, and accuracy of the radiomics model, the clinical ultrasound model, and the comprehensive model for predicting liver metastasis in pNEN patients were 0.800, 0.574, 0.789, 0.714; 0.780, 0.596, 0.874, 0.777; and 0.890, 0.694, 0.874, 0.810, respectively. The Delong test showed that the comprehensive model had the best predictive performance, with an AUC superior to that of radiomics model ( Z=3.845, P=0.000 12) and clinical ultrasound model ( Z=3.506, P=0.000 45). Conclusions:The radiomics model based on ultrasound has good performance in predicting liver metastasis in pNEN, and the comprehensive model that combines clinical ultrasound and radiomics features can further improve the predictive performance of the model.

Diagnosis model of the white matter injury of premature infants based on ultrasonography radiomics / 中华超声影像学杂志

Objective:To investigate the value of the ultrasonography in the diagnosis of the white matter injury of premature infants based on gray-scale ultrasonography radiomics.Methods:A total of 256 premature infants in Huazhong University of Science and Technology Union Shenzhen Hospital and Shenzhen Hospital of Southern Medical University from August 2018 to April 2022 were analyzed retrospectively. The computer-generated random numbers were assigned to the training set and the verification set according to 6∶4 ratio. On the basis of standardized collection of craniocerebral ultrasound images, the radiomics features were extracted from imaging by Pyradiomics 3.0.1 software package, the Mann-Whitney U test and the least absolute shrinkage and selection operator (LASSO) and stepwise regression were used to select the optimal features. Then the Logistic regression was used to build radiomics model. According to MRI, ROC curve was utilized to evaluate the performance of the model. The craniocerebral ultrasound images in the validation set were independently diagnosed by a senior physician and a junior physician, and then the above two physicians diagnosed again with the help of the radiomics, and the diagnostic abilities of this model were compared with those of the junior and senior physicians with and without radiomics assist. Results:A total of 5 optimal features were selected to develop radiomics model. The sensitivity, specificity, accuracy and the area under the ROC curve (AUC) in the training and validation sets were 0.861, 0.775, 0.799, 0.818; 0.929, 0.824, 0.853, 0.876, respectively. The sensitivity, specificity, accuracy and AUC in the senior sonographer, the junior sonographer, and both of them with radiomics assist for the dagnosis in the validation set were 0.929, 0.892, 0.902, 0.910; 0.714, 0.743, 0.735, 0.729; 0.929, 0.919, 0.922, 0.924; 0.857, 0.824, 0.833, 0.841, respectively. Performance of radiomics model reached the level of the senior sonographer (AUC: 0.876 vs 0.910, P=0.284), which was significantly better than the performance of the junior sonographer(AUC: 0.876 vs 0.729, P=0.001). Performance of the junior sonographer with radiomics assist was significantly better than the performance of the junior sonographer(AUC: 0.841 vs 0.729, P=0.003). Performance of the senior sonographer with radiomics assist was comparable to that of the senior sonographer(AUC: 0.924 vs 0.910, P=0.156). Conclusions:The ultrasound diagnosis method based on radiomics technology shows good diagnostic performance for the white matter injury of premature infants. It is helpful to improve the diagnostic ability of junior sonographer. It is expected to assist the sonographers in diagnosis and provide objective, consistent and accurate results for clinical practice.

Machine learning based on automated breast volume scanner radiomics for differential diagnosis of benign and malignant BI-RADS 4 lesions / 中华超声影像学杂志

Objective:To evaluate the performance of machine learning (ML) based on automated breast volume scanner (ABVS) radiomics in distinguishing benign and malignant BI-RADS 4 lesions.Methods:Between May to December 2020, patients with BI-RADS 4 lesions from the Affiliated Hospital of Southwest Medical University (Center 1) and Guangdong Provincial Hospital of Traditional Chinese Medicine (Center 2) were prospectively collected and divided into training cohort (Center 1) and external validation cohort (Center 2). The radiomics features of BI-RADS 4 lesions were extracted from the axial, sagittal and coronal ABVS images by MaZda software. In the training cohort, 7 feature selection methods and thirteen ML algorithms were combined in pairs to construct different ML models, and the 6 models with the best performance were verified in the external validation cohort to determine the final ML model. Finally, the diagnostic performance and confidence (5-point scale) of radiologists (R1, R2 and R3, with 3, 6 and 10 years of experience, respectively) with or without the model were evaluated.Results:①A total of 251 BI-RADS 4 lesions were enrolled, including 178 lesions (91 benign, 87 malignant) in the training cohort and 73 lesions (44 benign, 29 malignant) in the external validation cases. ②In the training cohort, the 6 ML models (DNN-RFE, AdaBoost-RFE, LR-RFE, LDA-RFE, Bagging-RFE and SVM-RFE) had the best diagnostic performance, and their AUCs were 0.972, 0.969, 0.968, 0.968, 0.967 and 0.962, respectively. ③In the external validation cohort, the DNN-RFE still had the best performance with the AUC, accuracy, sensitivity, specificity, PPV and NPV were 0.886, 0.836, 0.934, 0.776, 86.8% and 82.5%, respectively. ④Before model assistance, R1 had the worst diagnostic performance with the accuracy, sensitivity, specificity, PPV and NPV were 0.680, 0.750, 0.640, 57% and 81%, respectively. After model assistance, the diagnostic performance of R1 was significantly improved ( P=0.039), and its diagnostic sensitivity, specificity, accuracy, PPV and NPV improved to 0.730, 0.810, 0.930, 68% and 94%; while the improvement of R2 and R3 were not significantly ( P=0.811, 0.752). Meanwhile, the overall diagnostic confidence of the 3 radiologists increased from 2.8 to 3.3 ( P<0.001). Conclusions:The performance of ML based on ABVS radiomics in distinguishing between benign and malignant BI-RADS 4 lesions is good, which may improve the diagnostic performance of inexperienced radiologists and enhance diagnostic confidence.

Preoperative prediction of Ki-67 level in hepatocellular carcinoma based on radiomics signatures during Kupffer phase of Sonazoid contrast enhanced ultrasound / 中华超声影像学杂志

Objective:To evaluate the value of Sonazoid contrast enhanced ultrasound (CEUS) in preoperative prediction of proliferating cell nuclear antigen 67 (Ki-67) level of hepatocellular carcinoma (HCC) by establishing predictive model based on radiomics features of Kupffer phase.Methods:From October 2020 to August 2021, patients with histologically confirmed HCC lesion and who underwent Sonazoid CEUS examination 1 week before surgery were prospectively enrolled. The radiomics signatures were extracted from the whole tumor region on gray scale images and Kupffer phase images. Two predictive radiomics models were constructed using radiomic method. The predictive performance of 2 models was compared.Results:A total of 50 patients with histologically confirmed single HCC lesions were prospectively enrolled in this study. Among them, histological results revealed 24 HCC lesions with high level representation of Ki-67 (>20%) and 26 HCC lesions with low level representation of Ki-67 (≤20%). Two radiomics predictive models were established based on gray scale images and Kupffer phase images respectively. While compared with model based on B-mode ultrasound images, model based on Kupffer phase images showed significantly higher area under receiver operating characteristic curve (0.753 vs 0.535, P=0.017), accuracy (0.720 vs 0.580, P=0.023) and sensitivity (0.458 vs 0.250, P=0.043). Calibration plot indicated that Kupffer phase model showed better consistency with the actual Ki-67 level than gray scale model. Conclusions:The radiomics model based on Kupffer phase features of Sonazoid CEUS is a preoperative and noninvasive prediction the presentation level of Ki-67 in HCC lesions.

Application value of CT and MRI radiomics based on machine-learning method in diagnosing pancreatic cancer / 中华胰腺病杂志

Objective:To investigate the application value of CT and MRI imageomics based on machine learning method in the diagnosis of pancreatic cancer.Methods:The clinical data of 62 patients with surgically resected and pathologically confirmed pancreatic cancer, who underwent enhanced CT scan, MRI plain or enhanced scan in Shanghai General Hospital between January 2014 and December 2021 were collected. According to the chronological order of surgery, 49 patients from January 2014 to December 2020 were enrolled in the training set and 13 patients from January 2021 to December 2021 were enrolled in the validation set. 3D-slicer 4.8.1 software was used to draw the region of interest in each layer of CT and MRI images for cancerous and paracancerous tissue segment. Image features were extracted by Python and the optimal feature set from the training set data was obtained by using Lasso regression model. The machine learning decision tree model was constructed. The receiver operating characteristic curve(ROC) curve was drawn, and the area under the curve (AUC) was calculated to evaluate the value of these three kinds of imageomics models in the diagnosis of pancreatic cancer.Results:The 1 767 CT features and 1 674 MRI features were obtained from enhanced CT scan, MRI plain scan and enhanced MRI scan, respectively. For the differential diagnosis model of cancerous tissue and paracancerous tissue, the enhanced CT scan data model obtained the optimal feature set involving 6 features, the MRI plain scan model obtained the optimal feature set involving 16 features, and the enhanced MRI scan model obtained the optimal feature set involving 15 features. The diagnostic model based on enhanced CT scan had an AUC of 0.98 in the training set and 1 in the verification group. The AUC of the MRI plain scan and enhanced MRI scan models in both the training set and the validation set was 1. The specificity and sensitivity of machine learning decision tree model based on the three kinds of imageomics models in the diagnosis of cancerous tissue and paracancerous tissue were 100%. For the differential diagnosis model of splenic artery wrapping, the enhanced CT scan model didn′t obtain the optimal features and had no diagnostic efficacy. The MRI plain scan model and enhanced MRI scan model obtained the optimal feature set involving 5 and 4 features, respectively. The AUC of the MRI plain scan model in the training set and the validation set were 0.862 and 0.750, respectively, with diagnostic sensitivity of 93.8% and 50.0%, and specificity of 78.6% and 100%, respectively. The AUC of the enhanced MRI scan model in the training set and the validation set were 0.950 and 0.861, respectively, with diagnostic sensitivity of 90.0% and 93.6%, and specificity of 100% and 78.6%, respectively.Conclusions:Based on the radiomics of CT enhanced, MRI plain scan and enhanced MRI scan, the machine learning diagnostic model has an accuracy of more than 90% in differentiating pancreatic cancer from paracancerous tissue. For the differentiation of splenic artery wrapping in pancreatic cancer, the diagnostic model based on enhanced MRI scan haS the best diagnostic efficiency.

Construction of a clinical-radiomic nomogram model for the differential diagnosis of adrenal adenoma and small diameter pheochromocytoma / 现代泌尿外科杂志

【Objective】 To effectively differentiate adrenal adenoma (AA) and small diameter pheochromocytoma (PCC) by establishing a clinical-radiomic nomogram model before surgery. 【Methods】 A total of 132 pathologically confirmed patients (45 PCC cases, 87 AA cases) were enrolled. After the features of enhanced CT were assessed, the radiomic features and related clinical indicators were extracted. Based on multiple Logistic regression, a clinical-radiomic nomogram with radiomic features and independent clinical predictors was developed. The area under the receiver operating characteristic (ROC) curve (AUC) was used for internal evaluation to compare the diagnostic effectiveness of the three models. The clinical effectiveness of the nomogram was verified with decision curve analysis (DCA). 【Results】 One of the 108 candidate features was used to construct the radiological feature score. Individualized clinical-radiomic nomogram included independent clinical factors (24 h urinary vanmandelic acid/renin/angiotensin I) and original first-order median radiological feature scores. Internal evaluation of the prediction model showed that the AUC was 0.945 (95%CI:0.906-0.984), superior to the single clinical model or radiological model in precise differentiation. DCA showed that the nomogram had the best clinical use. 【Conclusion】 The clinical-radiomic nomogram model can effectively differentiate adrenal adenoma from small diameter pheochromocytoma, which can improve the preoperative differential diagnosis and preparation.

Research Progress in Artificial Intelligence for Diagnosis， Prediction of Recurrence and Prognosis in Hepatocellular Carcinoma / 中山大学学报(医学科学版)

With the rapid development of artificial intelligence （AI） technology in the field of medicine， AI models show great potential in the diagnosis， prognosis and efficacy prediction of hepatocellular carcinoma （HCC）. AI techniques include computational search algorithms， machine learning （ML） and deep learning （DL） models. Based on histopathology， radiomics and related molecular markers， the ML or DL algorithm is used to extract key information and then establish the diagnosis or prediction model， which may serve as a tool to aid in clinical decision-making. Further technical support， large-scale clinical validation and regulatory approvals are still needed due to the limitations on the application of AI models. This review summarizes the advances of AI in HC diagnosis， prediction of recurrence and prognosis， and highlights the radiomics， histopathology and molecular marker data.

Application value of radiomics-based machine learning model in identifying the degree of pulmonary ground-glass nodule infiltration / 中国胸心血管外科临床杂志

@#Objective    To establish a machine learning model based on computed tomography (CT) radiomics for preoperatively predicting invasive degree of lung ground-glass nodules (GGNs). Methods    We retrospectively analyzed the clinical data of GGNs patients whose solid component less than 3 cm in the Department of Thoracic Surgery of Shanghai Pulmonary Hospital from March 2021 to July 2021 and the First Hospital of Lanzhou University from January 2019 to May 2022. The lesions were divided into pre-invasiveness and invasiveness according to postoperative pathological results, and the patients were randomly divided into a training set and a test set in a ratio of 7∶3. Radiomic features (1 317) were extracted from CT images of each patient, the max-relevance and min-redundancy (mRMR) was used to screen the top 100 features with the most relevant categories, least absolute shrinkage and selection operator (LASSO) was used to select radiomic features, and the support vector machine (SVM) classifier was used to establish the prediction model. We calculated the area under the curve (AUC), sensitivity, specificity, accuracy, negative predictive value, positive predictive value to evaluate the performance of the model, drawing calibration and decision curves of the prediction model to evaluate the accuracy and clinical benefit of the model, analyzed the performance in the training set and subgroups with different nodule diameters, and compared the prediction performance of this model with Mayo and Brock models. Two primary thoracic surgeons were required to evaluate the invasiveness of GGNs to investigate the clinical utility of the mode. Results    A total of 400 patients were divided into the training set (n=280) and the test set (n=120) according to the admission criteria. There were 267 females and 133 males with an ……

Advances in the application of radiomics in predicting outcomes after transarterial chemoembolization for hepatocellular carcinoma / 中华全科医师杂志

Hepatocellular carcinoma is one of the common malignant tumors, and most patients with hepatocellular carcinoma are already in the middle stage at the time of clinical detection, transarterial chemoembolization is the treatment of choice for mid-stage hepatocellular carcinoma.Due to the high degree of tumor heterogeneity, accurately predicting the outcome of patients with hepatocellular carcinoma after transarterial chemoembolization remains one of the difficulties in clinical practice.As an emerging technology, radiomics can not only reflect tumor heterogeneity non-invasively, but also monitor, evaluate and predict tumor progression by analyzing changes in the tumor microenvironment to guide patients′ personalized treatment and prolong their survival time.This article reviews the progress of the application of radiomics in predicting the efficacy of transarterial chemoembolization for hepatocellular carcinoma.