Radiomics-based early prediction and treatment efficacy evaluation is critical for personalized treatment strategies in patients with colorectal cancer liver metastases(CCLM).Owing to the high artificial intelligence(AI)participation,repeatability,and reliable perform-ance,deep learning(DL)based on convolutional neural networks enhances the predictive efficacy of the models,enabling its potential clinic-al application more promising.Subsequent to the gradual construction of a multimodal fusion model and multicenter large sample database,radiomics and DL will become increasingly essential in the management of CCLM.This review focuses on the main steps of radiomics and DL,and summarizes the value of its application in early state prediction and treatment efficacy evaluation of different treatment modalities in CCLM,we also look forward to the potential of its in-depth application in the clinical management of CCLM.

Objective To observe the value of lumbar spine bone marrow MR T1 mapping radiomics for predicting clinical risk of acute lymphoblastic leukemia(ALL)in children.Methods Lumbar bone marrow T1 mappings were prospectively acquired from 77 newly diagnosed ALL children.The volume of interest(VOI)of L3 vertebral body was segmented using 3D Slicer software and 2 060 radiomics features were extracted,and the best features were screened.The children were divided into training and testing sets at the ratio of 8:2.Logistic regression(LR),support vector machine(SVM)and random forest(RF)were used to established radiomics models based on the best features,respectively,which were trained in training set and verified in testing set.The clinical risk was evaluated according to newly diagnosed risk and the response to chemotherapy after MR examination.Receiver operating characteristic(ROC)curve was drawn,and the area under the curve(AUC)was calculated to evaluate the efficacy of each model for predicting clinical risk of ALL in children.Results There were 52 cases in low-medium risk group and 25 in high risk group.The training set consisted of 44 cases of low-medium risk and 17 of high risk,while the testing set consisted of 8 cases of low-medium risk and 8 of high risk.Twelve best features were selected to establish radiomics models.The sensitivity and accuracy of RF model in training set were both 100％,but its sensitivity(50.00％)and accuracy(75.00％)in testing set were both low,which indicating overfitting.The AUC(0.95)of LR model was slightly higher than that of SVM model(0.92)in testing set,but no significantly difference was found(P＞0.05),and the accuracy of these two models was consistent.Conclusion Both lumbar bone marrow T1 mapping LR and SVM radiomics models could be used to predict clinical risk of ALL in children,and LR model had better predictive efficacy.

Purpose To use CT-arterial phase images of pre-treatment ovarian cancer patients,combined with deep learning algorithms and machine learning to build a model to predict the efficacy of neoadjuvant chemotherapy in ovarian cancer.Materials and Methods A total of 302 consecutive patients who underwent surgery and were pathologically diagnosed with ovarian cancer from March 2013 to August 2019 in Tianjin Medical University were retrospectively collected.All patients were partitioned into training and test sets according to the ratio of 7∶3.In the python environment,VGG13 model was integrated via combining deep learning network and machine learning,and features were filtered via least absolute shrinkage and selection operator algorithm to build a prediction model for classification and prediction of CT images.The area under the curve(AUC),accuracy,sensitivity,specificity,and Fl-Score were calculated,respectively.Results The AUC,accuracy,sensitivity,specificity,and Fl-Score of the model in the training set were 0.87,0.81,0.80,0.82 and 0.79,and 0.90,0.84,0.93,0.77 and 0.83 in the test set,respectively.The AUC of five-fold cross-validation were 0.86,0.88,0.88,0.90 and 0.87,respectively.Conclusion Predictive model based on CT images combined with deep learning and machine learning methods can provide a new clinical perspective for developing chemotherapy regimens for ovarian cancer.

Objective To establish a method to characterize the size and density of the female breast based on three-dimensional images of breast cone-beam computed tomography (CBCT), and describe the breast characteristics of women in a region of China, and to explore its value in dosimetric assessment for breast CBCT examinees. Methods We retrospectively surveyed the breast CBCT images of 203 women in a grade A tertiary hospital in a southwestern city of China from January 2021 to March 2023. The effective diameter of the breast at the chest-wall (D_eff), chest wall-to-nipple length (CNL), the effective diameter of the breast at half of CNL (D_h/2), breast volume (BV), glandular volume (GV), and volumetric breast density (VBD) were measured using the specific tools of the Koning Imaging Viewer system. The differences between groups were assessed using the Kruskal-Wallis H test. The correlation between variables was assessed using the Spearman’s correlation coefficient. Results The median values of D_eff, D_h/2, CNL, BV, GV, and VBD of the surveyed population were 11.9 cm, 8.3 cm, 6.5 cm, 327.7 cm³, 47.0 cm³, and 15.4%, respectively. GV and VBD had significant negative correlations with age. D_eff, D_h/2, CNL, and BV were significantly negatively correlated with VBD. Conclusion We established a quantitative method to analyze female breast characteristics based on three-dimensional breast CBCT images, and preliminarily characterized the female breast in a region of China, which can provide methodological support for the investigation of female breast characteristics in various regions of China in the future.

Mammography has played an essential role in the screening and treatment of breast cancer. However, the application of X-rays will also increase the risks of breast cancer while improving its detection rate. Moreover, the risks will increase with an increase in the radiation dose. Since the glandular tissue in breasts is sensitive to radiation, the evaluation of the average glandular dose (AGD) in mammography has attracted considerable international attention. Compared to relatively mature dosimetric studies on traditional two-dimension mammography and digital breast tomosynthesis, the method for the dose evaluation of the new cone beam CT for breasts are still subjected to research. This paper reviews and explores the current status of studies on the assessment method and relevant influencing factors of AGD under different types of mammography equipment.

Objective:To investigate the computed tomography(CT)and magnetic resonance imaging(MRI)features of retroperitoneal ded-ifferentiated liposarcoma(DDL),and improve the understanding of DDL and the accuracy of preoperative diagnosis.Methods:Clinical and imaging features of 25 patients with retroperitoneal DDL from Tianjin Medical University Cancer Institute&Hospital,confirmed by patho-logy from January 2012 to June 2022,were retrospectively analyzed.Results:Among 25 cases of retroperitoneal DDL,19 and 6 had single and multiple tumors,respectively and 10 and 15 were oval and irregular shaped tumors,respectively.Most lesions had unclear boundaries,with 15 cases invading the surrounding tissues and organs.Small vessel shadows were visible in 15 cases,while calcifications or ossifications were observed in 7 cases,and cystic necrosis was observed in only 3 cases.Enhanced scanning exhibits a centripetal and progressive con-tinuous augmentation characteristic defined as"slow in and slow out."According to its manifestations in CT and MRI,it can be divided into two types:type I(soft tissue mass type),where the tumor has a soft tissue component with no fat content(14 cases);and type Ⅱ(fat con-taining),where the tumors exhibit both soft tissue and adipose components,most of which are clearly defined and rarely present in a mosa-ic shape.In abnormal fat areas,cord-like fibrous septa can be seen.Among them,the intratumoral fat composition<50%was Ⅱa type(10 cases).Intratumor fat composition≥50%was type Ⅱb(1 case).Conclusions:Combined with imaging classification,a comprehensive ana-lysis of the CT and MRI imaging characteristics of retroperitoneal DDL is of great value for its preoperative qualitative diagnosis.

Objective:To establish a clinical diagnostic scoring model for preoperative predicting hepatocellular carcinoma (HCC) microvascular invasion (MVI) based on gadolinium-ethoxybenzyl-diethylenetriamine pentacetic acid (Gd-EOB-DTPA) enhanced MRI, and verify its effectiveness.Methods:From January 2014 to December 2020, a total of 251 cases with pathologically confirmed HCC from Tianjin First Central Hospital and Jilin University First Hospital were retrospectively collected to serve as the training set, while 57 HCC patients from Tianjin Medical University Cancer Hospital were recruited as an independent external validation set. The HCC patients were divided into MVI positive and MVI negative groups according to the pathological results. The tumor maximum diameters and apparent diffusion coefficient (ADC) values were measured. On the Gd-EOB-DTPA MRI images, tumor morphology, peritumoral enhancement, peritumoral low intensity (PTLI), capsule, intratumoral artery, intratumoral fat, intratumoral hemorrhage, and intratumoral necrosis were observed. Univariate analysis was performed using the χ 2 test or the independent sample t-test. The independent risk factors associated with MVI were obtained in the training set using a multivariate logistic analysis. Points were assigned to each factor according to the weight value to establish a preoperative score model for predicting MVI. The receiver operating characteristic (ROC) curve was used to determine the score threshold and to verify the efficacy of this scoring model in predicting MVI in the independent external validation set. Results:The training set obtained 98 patients in the MVI positive group and 153 patients in the MVI negative group, while the external validation set obtained 16 patients in the MVI positive group and 41 patients in the MVI negative group. According to logistic analysis, tumor maximum diameter>3.66 cm (OR 3.654, 95%CI 1.902-7.018), hepatobiliary PTLI (OR 9.235, 95%CI 4.833-16.896) and incomplete capsule (OR 6.266, 95%CI 1.993-9.345) were independent risk factors for MVI in HCC, which were assigned scores of 3, 4 and 2, respectively. The total score ranged from 0 to 9. In the external validation set, ROC curve analysis showed that the area under the curve of the scoring model was 0.918 (95%CI 0.815-0.974, P=0.001). When the score>4 was used as the threshold, the accuracy, sensitivity, and specificity of the model in predicting MVI were 84.2%, 81.3%, and 85.4%, respectively. Conclusions:A scoring model based on Gd-EOB-DTPA-enhanced MRI provided a convenient and reliable way to predict MVI preoperatively.

Objective:To explore the predictive value of qualitative and quantitative parameters of dual-layer spectral detector CT plain scan on the invasiveness of pure ground-glass pulmonary nodules (pGGNs).Methods:Clinical and imaging data of 113 patients (119 pGGNs) with pathology-proven lung adenocarcinoma who underwent preoperative dual-layer spectral detector CT plain scan in Tianjin Medical University Cancer Institute and Hospital from November 2019 to December 2020 were retrospectively analyzed. According to invasiveness, pGGNs were divided into non-invasive adenocarcinoma (non-IA) group ( n=66) and IA group ( n=53). The non-IA group included atypical adenomatous hyperplasia ( n=10), adenocarcinoma in situ ( n=26) and minimally invasive adenocarcinoma ( n=30). The qualitative parameters were nodule shape, lung-tumor interface, lobulation, spiculation, pleural retraction, bubblelike lucency, air bronchogram and vascular abnormality. The quantitative parameters included nodule size, effective atomic number (Z eff), CT value on 120 kVp images (CT 120 kVp) and virtual monoenergetic images from 40 keV to 200 keV (CT 40 keV-CT 200 keV), and slope of energy spectrum curve (λHU). The χ 2 test, Mann-Whitney U test and independent sample t test were used to analyze the parameter differences between non-IA group and IA group. Multivariate logistic regression analysis was performed to screen out independent predictors. Receiver operating characteristic (ROC) curve was used to assess the diagnostic efficacy of single predictor and combined independent factors for the invasiveness of pGGN. Results:Significant differences were found in nodule shape, lobulation, air bronchogram, vascular abnormality, nodules size, Z eff, CT 120 kVp and CT 40 keV-CT 200 keV between non-IA and IA groups ( P<0.05). The maltivariate logistic regression analysis showed that nodule size [odds ratio 9.269, 95% confidence interval (CI) 1.640-52.395, P=0.012] CT 200 keV (odds ratio 1.012, 95%CI 1.006-1.019, P<0.001) as well as vascular abnormality sign (odds ratio 4.940, 95%CI 1.358-17.969, P=0.015) were independent predictors of pGGN invasiveness. ROC curve analysis of a single independent predictor and a combination of the three factors showed that the area under the curve (AUC) of the combination of three factors predicting the invasiveness of pGGN was significantly higher than the AUC of vascular abnormality sign ( Z=4.01, P<0.001) and CT 200 keV ( Z=3.25, P=0.001), while there was no significant difference in AUC between the combination of the three factors and nodule size ( Z=1.94, P=0.052). The AUC of the combination of the three independent predictors was 0.909, and the sensitivity and specificity for predicting pGGN invasion were 81.1% and 86.4%, respectively, using a threshold of 0.505. Conclusion:The combination of qualitative and quantitative parameters of dual-layer spectral detector CT plain scan shows a high predictive value for the invasiveness of pGGNs.

Objective:To study the predictive value of combining alpha-feto protein (AFP) with contrast-enhanced MRI imaging features in predicting incidence of microvascular invasion (MVI) in patients with hepatocellular carcinoma.Methods:The data of 206 patients with hepatocellular carcinoma treated at Tianjin Medical University Cancer Institute and Hospital from January 2017 to April 2019 were retrospectively analyzed. There were 179 males and 27 females, with an average age of 58.7 years. The roles of preoperative MRI imaging features and clinical data on predicting the incidence of MVI in patients with hepatocellular carcinoma were evaluated by univariate and multivariate logistic regression analyses. Multivariable regression analysis was then used to plot a nomogram.Results:There were 86 patients (41.7%) with MVI positivity and 120 patients (58.3%) with MVI negativity. Multivariate logistic regression analysis showed that AFP >400 μg/L ( OR=3.318, 95% CI: 1.243-8.855, P=0.017), two-trait predictor of venous invasion (TTPVI) ( OR=13.111, 95% CI: 6.797-28.119, P<0.001), diffusion weighted imaging/T 2 weighted imaging (DWI/T 2WI) mismatch ( OR=17.233, 95% CI: 4.731-44.490, P<0.001), and rim enhancement( OR=5.665, 95% CI: 2.579-18.152, P=0.013) predicted increased risks of MVI in patients with hepatocellular carcinoma. The constructed nomogram directly predicted the risk of MVI in these patients. Conclusions:AFP>400 μg/L, TTPVI, DWI/T 2WI mismatch and rim enhancement were independent risk factors in predicting MVI in patients with hepatocellular carcinoma. This predictive model of MVI which was based on multivariate logistic regression analysis was helpful to clinicians in making individualized treatment plans for patients with hepatocellular carcinoma.

Objective:To investigate the value of logistic regression model based on the features of cone-beam breast CT (CBBCT) for the identification of benign and malignant masses in dense breast.Methods:The data of 106 patients (130 masses) with dense breast who underwent contrast-enhanced CBBCT examination and obtained pathological results from May 2011 to August 2020 were retrospectively analyzed as the training set. From August 2020, the data of 49 patients (54 masses) who met the same criteria were prospectively and consecutively collected and used as the validation set. Taking pathological results as the gold standard, the training set was divided into benign and malignant groups. The t-test, χ 2 test and Fisher′s exact test were used to compare the differences in CBBCT image characteristics between the two groups in the training set. A binary logistic regression model was established by multivariate analysis. ROC curves were used to assess the diagnostic efficacy of the model as a whole in the training and validation sets and the diagnostic efficacy of each feature in the model, and the cut-off value of the intensity (ΔCT) value was determined. The H-L method was used to test the goodness of fit of the model. Decision curve analysis (DCA) was drawn to validate the clinical power of the model. Results:Univariate analysis showed that the breast parenchymal background enhancement (BPE), shape, margin, lobulation, spiculation, density, calcifications, ΔCT value, enhancement pattern, non-mass enhancement, ipsilateral increased vascularity (IIV), and peripheral vascular signs had statistical difference between benign group and malignant group ( P<0.05). BPE, margin, ΔCT value and IIV were included in the multivariate analysis, the equation was logit( P′)=-8.510+0.830×BPE+0.822×margin+1.919× ΔCT+1.896 × IIV. The are a under curve of the model in the training set was 0.879 ( P<0.001) and in the validation set was 0.851 ( P=0.001). The are a under curve of BPE, margin, ΔCT value, and IIV in the diagnosis of malignant mass were 0.645, 0.711, 0.712, 0.775 (all P<0.05); the best cut-off value of ΔCT was 50.38 HU. The fit of this model was good ( P = 0.776). The DCA curve showed that when the risk threshold was 0.05-0.97, the net benefit rate was>0, and this model had some clinical value. Conclusion:The logistic regression model based on the features of CBBCT is helpful to distinguish benign and malignant masses in dense breasts.