Deep Learning Combined with Radiologist's Intervention Achieves Accurate Segmentation of Hepatocellular Carcinoma in Dual-Phase Magnetic Resonance Images.

Purpose: Segmentation of hepatocellular carcinoma (HCC) is crucial; however, manual segmentation is subjective and time-consuming. Accurate and automatic lesion contouring for HCC is desirable in clinical practice. In response to this need, our study introduced a segmentation approach for HCC combining deep convolutional neural networks (DCNNs) and radiologist intervention in magnetic resonance imaging (MRI). We sought to design a segmentation method with a deep learning method that automatically segments using manual location information for moderately experienced radiologists. In addition, we verified the viability of this method to assist radiologists in accurate and fast lesion segmentation. Method: In our study, we developed a semiautomatic approach for segmenting HCC using DCNN in conjunction with radiologist intervention in dual-phase gadolinium-ethoxybenzyl-diethylenetriamine penta-acetic acid- (Gd-EOB-DTPA-) enhanced MRI. We developed a DCNN and deep fusion network (DFN) trained on full-size images, namely, DCNN-F and DFN-F. Furthermore, DFN was applied to the image blocks containing tumor lesions that were roughly contoured by a radiologist with 10 years of experience in abdominal MRI, and this method was named DFN-R. Another radiologist with five years of experience (moderate experience) performed tumor lesion contouring for comparison with our proposed methods. The ground truth image was contoured by an experienced radiologist and reviewed by an independent experienced radiologist. Results: The mean DSC of DCNN-F, DFN-F, and DFN-R was 0.69 ± 0.20 (median, 0.72), 0.74 ± 0.21 (median, 0.77), and 0.83 ± 0.13 (median, 0.88), respectively. The mean DSC of the segmentation by the radiologist with moderate experience was 0.79 ± 0.11 (median, 0.83), which was lower than the performance of DFN-R. Conclusions: Deep learning using dual-phase MRI shows great potential for HCC lesion segmentation. The radiologist-aided semiautomated method (DFN-R) achieved improved performance compared to manual contouring by the radiologist with moderate experience, although the difference was not statistically significant.

Predicting One-year Recurrence of HCC based on Gadoxetic Acid-enhanced MRI by Machine Learning Models.

OBJECTIVE: Accurate prediction of recurrence risk after resction in patients with Hepatocellular Carcinoma (HCC) may help to individualize therapy strategies. This study aimed to develop machine learning models based on preoperative clinical factors and multiparameter Magnetic Resonance Imaging (MRI) characteristics to predict the 1-year recurrence after HCC resection. METHODS: Eighty-two patients with single HCC who underwent surgery were retrospectively analyzed. All patients underwent preoperative gadoxetic acidenhanced MRI examination. Preoperative clinical factors and MRI characteristics were collected for feature selection. Least Absolute Shrinkage and Selection Operator (LASSO) was applied to select the optimal features for predicting postoperative 1-year recurrence of HCC. Four machine learning algorithms, Multilayer Perception (MLP), random forest, support vector machine, and k-nearest neighbor, were used to construct the predictive models based on the selected features. A Receiver Operating Characteristic (ROC) curve was used to assess the performance of each model. RESULTS: Among the enrolled patients, 32 patients experienced recurrences within one year, while 50 did not. Tumor size, peritumoral hypointensity, decreasing ratio of liver parenchyma T1 value (ΔT1), and α-fetoprotein (AFP) levels were selected by using LASSO to develop the machine learning models. The area under the curve (AUC) of each model exceeded 0.72. Among the models, the MLP model showed the best performance with an AUC, accuracy, sensitivity, and specificity of 0.813, 0.742, 0.570, and 0.853, respectively. CONCLUSION: Machine learning models can accurately predict postoperative 1-year recurrence in patients with HCC, which may help to provide individualized treatment.

Prediction of immunocyte infiltration and prognosis in postoperative hepatitis B virus-related hepatocellular carcinoma patients using magnetic resonance imaging.

Background: The immune microenvironment (IME) is closely associated with prognosis and therapeutic response of hepatitis B virus-related hepatocellular carcinoma (HBV-HCC). Multi-parametric magnetic resonance imaging (MRI) enables non-invasive assessment of IME and predicts prognosis in HBV-HCC. We aimed to construct an MRI prediction model of the immunocyte-infiltration subtypes and explore its prognostic significance. Methods: HBV-HCC patients at the First Affiliated Hospital of Sun Yat-sen University (Guangzhou, China) with radical surgery (between 1 October and 30 December 2021) were prospectively enrolled. Patients with pathologically proven HCC (between 1 December 2013 and 30 October 2019) were retrospectively enrolled. Pearson correlation analysis was used to examine the relationship between the immunocyte-infiltration counts and MRI parameters. An MRI prediction model of immunocyte-infiltration subtypes was constructed in prospective cohort. Kaplan-Meier survival analysis was used to analyse its prognostic significance in the retrospective cohort. Results: Twenty-four patients were prospectively enrolled to construct the MRI prediction model. Eighty-nine patients were retrospectively enrolled to determine its prognostic significance. MRI parameters (relative enhancement, ratio of the apparent diffusion coefficient value of tumoral region to peritumoral region [rADC], T1 value) correlated significantly with the immunocyte-infiltration counts (leukocytes, T help cells, PD1+Tc cells, B lymphocytes). rADC differed significantly between high and low immunocyte-infiltration groups (1.47 ± 0.36 vs 1.09 ± 0.25, P = 0.009). The area under the curve of the MRI model was 0.787 (95% confidence interval 0.587-0.987). Based on the MRI model, the recurrence-free time was longer in the high immunocyte-infiltration group than in the low immunocyte-infiltration group (P = 0.026). Conclusions: MRI is a non-invasive method for assessing the IME and immunocyte-infiltration subtypes, and predicting prognosis in post-operative HBV-HCC patients.

Dual-energy CT improves differentiation of non-hypervascular pancreatic neuroendocrine neoplasms from CA 19-9-negative pancreatic ductal adenocarcinomas.

PURPOSE: To evaluate the utility of dual-energy CT (DECT) in differentiating non-hypervascular pancreatic neuroendocrine neoplasms (PNENs) from pancreatic ductal adenocarcinomas (PDACs) with negative carbohydrate antigen 19-9 (CA 19-9). METHODS: This retrospective study included 26 and 39 patients with pathologically confirmed non-hypervascular PNENs and CA 19-9-negative PDACs, respectively, who underwent contrast-enhanced DECT before treatment between June 2019 and December 2021. The clinical, conventional CT qualitative, conventional CT quantitative, and DECT quantitative parameters of the two groups were compared using univariate analysis and selected by least absolute shrinkage and selection operator regression (LASSO) analysis. Multivariate logistic regression analyses were performed to build qualitative, conventional CT quantitative, DECT quantitative, and comprehensive models. The areas under the receiver operating characteristic curve (AUCs) of the models were compared using DeLong's test. RESULTS: The AUCs of the DECT quantitative (based on normalized iodine concentrations [nICs] in the arterial and portal venous phases: 0.918; 95% confidence interval [CI] 0.852-0.985) and comprehensive (based on tumour location and nICs in the arterial and portal venous phases: 0.966; 95% CI 0.889-0.995) models were higher than those of the qualitative (based on tumour location: 0.782; 95% CI 0.665-0.899) and conventional CT quantitative (based on normalized conventional CT attenuation in the arterial phase: 0.665; 95% CI 0.533-0.797; all P < 0.05) models. The DECT quantitative and comprehensive models had comparable performances (P = 0.076). CONCLUSIONS: Higher nICs in the arterial and portal venous phases were associated with higher blood supply improving the identification of non-hypervascular PNENs.

A radiomics-based interpretable model to predict the pathological grade of pancreatic neuroendocrine tumors.

OBJECTIVES: To develop a computed tomography (CT) radiomics-based interpretable machine learning (ML) model to predict the pathological grade of pancreatic neuroendocrine tumors (pNETs) in a non-invasive manner. METHODS: Patients with pNETs who underwent contrast-enhanced abdominal CT between 2010 and 2022 were included in this retrospective study. Radiomics features were extracted, and five radiomics-based ML models, namely logistic regression (LR), random forest (RF), support vector machine (SVM), XGBoost, and GaussianNB, were developed. The performance of these models was evaluated using a time-independent testing set, and metrics such as sensitivity, specificity, accuracy, and the area under the receiver operating characteristic curve (AUC) were calculated. The accuracy of the radiomics model was compared to that of needle biopsy. The Shapley Additive Explanation (SHAP) tool and the correlation between radiomics and biological features were employed to explore the interpretability of the model. RESULTS: A total of 122 patients (mean age: 50 ± 14 years; 53 male) were included in the training set, whereas 100 patients (mean age: 48 ± 13 years; 50 male) were included in the testing set. The AUCs for LR, SVM, RF, XGBoost, and GaussianNB were 0.758, 0.742, 0.779, 0.744, and 0.745, respectively, with corresponding accuracies of 73.0%, 70.0%, 77.0%, 71.9%, and 72.9%. The SHAP tool identified two features of the venous phase as the most significant, which showed significant differences among the Ki-67 index or mitotic count subgroups (p < 0.001). CONCLUSIONS: An interpretable radiomics-based RF model can effectively differentiate between G1 and G2/3 of pNETs, demonstrating favorable interpretability. CLINICAL RELEVANCE STATEMENT: The radiomics-based interpretable model developed in this study has significant clinical relevance as it offers a non-invasive method for assessing the pathological grade of pancreatic neuroendocrine tumors and holds promise as an important complementary tool to traditional tissue biopsy. KEY POINTS: • A radiomics-based interpretable model was developed to predict the pathological grade of pNETs and compared with preoperative needle biopsy in terms of accuracy. • The model, based on CT radiomics, demonstrated favorable interpretability. • The radiomics model holds potential as a valuable complementary technique to preoperative needle biopsy; however, it should not be considered a replacement for biopsy.

Harnessing dual-energy CT for glycogen quantification: a phantom analysis.

Background: Non-invasive glycogen quantification in vivo could provide crucial information on biological processes for glycogen storage disorder. Using dual-energy computed tomography (DECT), this study aimed to assess the viability of quantifying glycogen content in vitro. Methods: A fast kilovolt-peak switching DECT was used to scan a phantom containing 33 cylinders with different proportions of glycogen and iodine mixture at varying doses. The virtual glycogen concentration (VGC) was then measured using material composition images. Additionally, the correlations between VGC and nominal glycogen concentration (NGC) were evaluated using least-square linear regression, then the calibration curve was constructed. Quantitative estimation was performed by calculating the linearity, conversion factor (inverse of curve slope), stability, sensitivity (limit of detection/limit of quantification), repeatability (inter-class correlation coefficient), and variability (coefficient of variation). Results: In all conditions, excellent linear relationship between VGC and NGC were observed (P<0.001, coefficient of determination: 0.989-0.997; residual root-mean-square error of glycogen: 1.862-3.267 mg/mL). The estimated conversion factor from VGC to NGC was 3.068-3.222. In addition, no significant differences in curve slope were observed among different dose levels and iodine densities. The limit of detection and limit of quantification had respective ranges of 6.421-15.315 and 10.95-16.46 mg/mL. The data demonstrated excellent scan-repeat scan agreement (inter-class correlation coefficient, 0.977-0.991) and small variation (coefficient of variation, 0.1-0.2%). Conclusions: The pilot phantom analysis demonstrated the feasibility and efficacy of detecting and quantifying glycogen using DECT and provided good quantitative performance with significant stability and reproducibility/variability. Thus, in the future, DECT could be used as a convenient method for glycogen quantification to provide more reliable information for clinical decision-making.

Clinical outcome is distinct between radiological stricture and endoscopic stricture in ileal Crohn's disease.

OBJECTIVES: Differences in clinical adverse outcomes (CAO) based on different intestinal stricturing definitions in Crohn's disease (CD) are poorly documented. This study aims to compare CAO between radiological strictures (RS) and endoscopic strictures (ES) in ileal CD and explore the significance of upstream dilatation in RS. METHODS: This retrospective double-center study included 199 patients (derivation cohort, n = 157; validation cohort, n = 42) with bowel strictures who simultaneously underwent endoscopic and radiologic examinations. RS was defined as a luminal narrowing with wall thickening relative to the normal gut on cross-sectional imaging (group 1 (G1)), which further divided into G1a (without upstream dilatation) and G1b (with upstream dilatation). ES was defined as an endoscopic non-passable stricture (group 2 (G2)). Strictures met the definitions of RS (with or without upstream dilatation) and ES were categorized as group 3 (G3). CAO referred to stricture-related surgery or penetrating disease. RESULTS: In the derivation cohort, G1b (93.3%) had the highest CAO occurrence rate, followed by G3 (32.6%), G1a (3.2%), and G2 (0%) (p < 0.0001); the same order was found in the validation cohort. The CAO-free survival time was significantly different among the four groups (p < 0.0001). Upstream dilatation (hazard ratio, 1.126) was a risk factor for predicting CAO in RS. Furthermore, when upstream dilatation was added to diagnose RS, 17.6% of high-risk strictures were neglected. CONCLUSIONS: CAO differs significantly between RS and ES, and clinicians should pay more attention to strictures in G1b and G3. Upstream dilatation has an important impact on the clinical outcome of RS but may not be an essential factor for RS diagnosis. CLINICAL RELEVANCE STATEMENT: This study explored the definition of intestinal stricture with the greatest significance for the clinical diagnosis and prognosis of patients with CD, and consequently provided effective auxiliary information for clinicians to formulate strategies for the treatment of CD intestinal strictures. KEY POINTS: • The retrospective double-center study showed that clinical adverse outcome is different between radiological strictures and endoscopic strictures in CD. • Upstream dilatation has an important impact on the clinical outcome of radiological strictures but may not be an essential factor for diagnosis of radiological strictures. • Radiological stricture with upstream dilatation and simultaneous radiological and endoscopic stricture were at increased risk for clinical adverse outcomes; thus, closer monitoring should be considered.

Evaluation of Hypoxia in Hepatocellular Carcinoma Using Quantitative MRI: Significances, Challenges, and Advances.

This review aimed to perform a scoping review of promising MRI methods in assessing tumor hypoxia in hepatocellular carcinoma (HCC). The hypoxic microenvironment and upregulated hypoxic metabolism in HCC are determining factors of poor prognosis, increased metastatic potential, and resistance to chemotherapy and radiotherapy. Assessing hypoxia in HCC is essential for personalized therapy and predicting prognoses. Oxygen electrodes, protein markers, optical imaging, and positron emission tomography can evaluate tumor hypoxia. These methods lack clinical applicability because of invasiveness, tissue depth, and radiation exposure. MRI methods, including blood oxygenation level-dependent, dynamic contrast-enhanced MRI, diffusion-weighted imaging, MRI spectroscopy, chemical exchange saturation transfer MRI, and multinuclear MRI, are promising noninvasive methods that evaluate the hypoxic microenvironment by observing biochemical processes in vivo, which may inform on therapeutic options. This review summarizes the recent challenges and advances in MRI techniques for assessing hypoxia in HCC and highlights the potential of MRI methods for examining the hypoxic microenvironment via specific metabolic substrates and pathways. Although the utilization of MRI methods for evaluating hypoxia in patients with HCC is increasing, rigorous validation is needed in order to translate these MRI methods into clinical use. Due to the limited sensitivity and specificity of current quantitative MRI methods, their acquisition and analysis protocols require further improvement. EVIDENCE LEVEL: 3. TECHNICAL EFFICACY: Stage 4.

Feasibility of Imaging-Based Quantitative Hepatic Function Assessment: Gd-EOB-DTPA-Enhanced Liver MR Imaging Combined with T1 Mapping.

OBJECTIVE: to investigate the feasibility of gadoxetic acid (Gd-EOB-DTPA) enhanced MRI combined with T1 mapping in quantitative hepatic function assessment. METHODS: this study retrospectively enrolled 94 patients with Gd-EOB-DTPA enhanced MRI combined with T1 mapping, divided into group A (grade A, n=73), group B (grade B, n=14) and group C (grade C, n=7) based on Child-Pugh classification. Liver T1 relaxation times on plain scan (T1P) and hepatocellular phase (T1E) were measured. Decrease in T1 (T1D) and the percentage of decrease in T1 (T1D%) were calculated as follows: T1D=T1P-T1E, T1D%= T1D/T1P×100%. The relationship between T1P, T1E, T1D, T1D% and liver function classification was analyzed. RESULTS: T1P, T1D, and T1D% in group A were significantly higher than those of group B and C. T1E in group A was lower than those of group B and C. T1D% was significantly different between group B and C. There was no significant difference in T1P, T1E, T1D between groups B and C. T1E was positively correlated with liver function levels, T1P and T1D had a negative correlation with liver function levels. T1P, T1E, T1D, T1D% were significantly different between cirrhotic and non-cirrhotic groups. T1D% of less than 70% suggests liver dysfunction. CONCLUSION: Gd-EOB-DTPA enhanced liver MRI combined with T1 mapping is feasible for quantitative assessment of hepatic function.

Use of dual-layer spectral detector computed tomography in the diagnosis of pancreatic neuroendocrine neoplasms.

PURPOSE: To explore the optimal energy level of dual-layer spectral detector computed tomography (DLCT) images of pancreatic neuroendocrine neoplasms (pNENs) and investigate the value in their detection. METHODS: This retrospective analysis included 134 pNEN patients with 136 lesions; they underwent contrast-enhanced DLCT scanning with histopathological confirmation of pNENs. Virtual monoenergetic images (VMI) of 40-100 keV, iodine concentration map (IC map), Z-effective atomic number map (Zeff map), and conventional images were analysed. The optimal energy level was obtained by comparing the signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR). The lesion detection rates of DLCT and conventional images were compared. Subjective image analysis was performed by two readers who assessed the image quality and lesion conspicuity on a 5-point scale. RESULTS: The SNR of VMIs from 40 to 80 keV (arterial phase, P < 0.001; venous phase, P < 0.05) and CNR from 40 to 60 keV (arterial and venous phases, each P < 0.05) were higher than that of conventional images; VMI40keV showed the highest SNR and CNR. There was a good inter-reader agreement between the two reviewers (Kappa values > 0.61); the scores of Zeff and IC maps were higher than those of conventional images and VMI40keV (P < 0.05). The detection performance of DLCT images was better than conventional images. CONCLUSIONS: The VMI40keV demonstrated the best CNR and SNR of pNENs compared to other VMIs. Zeff and IC maps improve objective image quality and reader preference compared to conventional images. These findings could possess important clinical implications in formulating treatment strategies.

Multi-frequency magnetic resonance elastography of the pancreas: measurement reproducibility and variance among healthy volunteers.

Background: Patients with chronic pancreatitis often have irreversible pancreatic insufficiency before a clinical diagnosis. Pancreatic cancer is a fatal malignant tumor in the advanced stages. Patients having high risk of pancreatic diseases must be screened early to obtain better outcomes using new imaging modalities. Therefore, this study aimed to investigate the reproducibility of tomoelastography measurements for assessing pancreatic stiffness and fluidity and the variance among healthy volunteers. Methods: Forty-seven healthy volunteers were prospectively enrolled and underwent two tomoelastography examinations at a mean interval of 7 days. Two radiologists blindly and independently measured the pancreatic stiffness and fluidity at the first examination to determine the reproducibility between readers. One radiologist measured the adjacent pancreatic slice at the first examination to determine the reproducibility among slices and measured the pancreas at the second examination to determine short-term repeatability. The stiffness and fluidity of the pancreatic head, body, and tail were compared to determine anatomical differences. The pancreatic stiffness and fluidity were compared based on sex, age, and body mass index (BMI). Results: Bland-Altman analyses (all P > 0.05) and intraclass correlation coefficients (all >0.9) indicated near perfect reproducibility among readers, slices, and examinations at short intervals. Neither stiffness (P = 0.477) nor fluidity (P = 0.368) differed among the pancreatic anatomical regions. The mean pancreatic stiffness was 1.45 ± 0.09 m/s; the mean pancreatic fluidity was 0.83 ± 0.06 rad. Stiffness and fluidity did not differ by sex, age, or BMI. Conclusion: Tomoelastography is a promising and reproducible tool for assessing pancreatic stiffness and fluidity in healthy volunteers.

Non-invasive diagnosis strategy of hepatocellular carcinoma in low-risk population.

AIMS: With prevalence of hepatocellular carcinoma (HCC) in low-risk population (LRP), establishing a non-invasive diagnostic strategy becomes increasingly urgent to spare unnecessary biopsies in this population. The purposes of this study were to find characterisics of HCC and to establish a proper non-invasive method to diagnose HCC in LRP. METHODS: A total of 681 patients in LRP (defined as the population without cirrhosis, chronic HBV infection or HCC history) were collected from 2 institutions. The images of computed tomography (CT) and magnetic resonance imaging (MRI) were manually analysed. We divided the patients into the training cohort (n = 324) and the internal validating cohort (n = 139) by admission time in the first institution. The cohort in the second institution was viewed as the external validation (n = 218). A multivariate logistic regression model incorporating both imaging and clinical independent risk predictors was developed. C-statistics was used to evaluate the diagnostic performance. RESULTS: Besides the major imaging features of HCC (non-rim enhancement, washout and enhancing capsule), tumor necrosis or severe ischemia (TNSI) on imaging and two clinical characteristics (gender and alpha fetoprotein) were also independently associated with HCC diagnosis (all P < 0.01). A clinical model (including 3 major features, TNSI, gender and AFP) was built to diagnose HCC and achieved good diagnostic performance (area under curve values were 0.954 in the training cohort, 0.931 in the internal validation cohort and 0.902 in the external cohort). CONCLUSIONS: The clinical model in this study developed a satisfied non-invasive diagnostic performance for HCC in LRP.

Tumor fibrosis correlates with the survival of patients with pancreatic adenocarcinoma and is predictable using clinicoradiological features.

OBJECTIVES: To evaluate the prognostic value of fibrosis for patients with pancreatic adenocarcinoma (PDAC) and preoperatively predict fibrosis using clinicoradiological features. Tumor fibrosis plays an important role in the chemoresistance of PDAC. However, the prognostic value of tumor fibrosis remains contradiction and accurate prediction of tumor fibrosis is required. METHODS: The study included 131 patients with PDAC who underwent first-line surgery. The prognostic value of fibrosis and rounded cutoff fibrosis points for median overall survival (OS) and disease-free survival (DFS) were determined using Cox regression and receiver operating characteristic (ROC) analyses. Then the whole cohort was randomly divided into training (n = 88) and validation (n = 43) sets. Binary logistic regression analysis was performed to select independent risk factors for fibrosis in the training set, and a nomogram was constructed. Nomogram performance was assessed using a calibration curve and decision curve analysis (DCA). RESULTS: Hazard ratios of fibrosis for OS and DFS were 1.121 (95% confidence interval [CI]: 1.082-1.161) and 1.110 (95% CI: 1.067-1.155). ROC analysis identified 40% as the rounded cutoff fibrosis point for median OS and DFS. Tumor diameter, carbohydrate antigen 19-9 level, and peripancreatic tumor infiltration were independent risk factors; areas under the nomogram curve were 0.810 and 0.804 in the training and validation sets, respectively. The calibration curve indicated good agreement of the nomogram, and DCA demonstrated good clinical usefulness. CONCLUSIONS: Tumor fibrosis was associated with poor OS and DFS in patients with PDAC. The nomogram incorporating clinicoradiological features was useful for preoperatively predicting tumor fibrosis. KEY POINTS: • Tumor fibrosis is correlated with poor prognosis in patients with pancreatic adenocarcinoma. • Tumor fibrosis can be categorized according to its association with overall survival and disease-free survival. • A nomogram incorporating carbohydrate antigen 19-9 level, tumor diameter, and peripancreatic tumor infiltration is useful for preoperatively predicting tumor fibrosis.

Deep Segmentation Feature-Based Radiomics Improves Recurrence Prediction of Hepatocellular Carcinoma.

Objective and Impact Statement. This study developed and validated a deep semantic segmentation feature-based radiomics (DSFR) model based on preoperative contrast-enhanced computed tomography (CECT) combined with clinical information to predict early recurrence (ER) of single hepatocellular carcinoma (HCC) after curative resection. ER prediction is of great significance to the therapeutic decision-making and surveillance strategy of HCC. Introduction. ER prediction is important for HCC. However, it cannot currently be adequately determined. Methods. Totally, 208 patients with single HCC after curative resection were retrospectively recruited into a model-development cohort (n=180) and an independent validation cohort (n=28). DSFR models based on different CT phases were developed. The optimal DSFR model was incorporated with clinical information to establish a DSFR-C model. An integrated nomogram based on the Cox regression was established. The DSFR signature was used to stratify high- and low-risk ER groups. Results. A portal phase-based DSFR model was selected as the optimal model (area under receiver operating characteristic curve (AUC): development cohort, 0.740; validation cohort, 0.717). The DSFR-C model achieved AUCs of 0.782 and 0.744 in the development and validation cohorts, respectively. In the development and validation cohorts, the integrated nomogram achieved C-index of 0.748 and 0.741 and time-dependent AUCs of 0.823 and 0.822, respectively, for recurrence-free survival (RFS) prediction. The RFS difference between the risk groups was statistically significant (P<0.0001 and P=0.045 in the development and validation cohorts, respectively). Conclusion. CECT-based DSFR can predict ER in single HCC after curative resection, and its combination with clinical information further improved the performance for ER prediction.

Prediction of Early Treatment Response to Initial Conventional Transarterial Chemoembolization Therapy for Hepatocellular Carcinoma by Machine-Learning Model Based on Computed Tomography.

PURPOSE: The treatment response to initial conventional transarterial chemoembolization (cTACE) is essential for the prognosis of patients with hepatocellular carcinoma (HCC). This study explored and verified the feasibility of machine-learning models based on clinical data and contrast-enhanced computed tomography (CT) image findings to predict early responses of HCC patients after initial cTACE treatment. PATIENTS AND METHODS: Overall, 110 consecutive unresectable HCC patients who were treated with cTACE for the first time were retrospectively enrolled. Clinical data and imaging features based on contrast-enhanced CT were collected for the selection of characteristics. Treatment responses were evaluated based on the modified Response Evaluation Criteria in Solid Tumors (mRECIST) by postoperative CT examination within 2 months after the procedure. Python (version 3.70) was used to develop machine learning models. Least absolute shrinkage and selection operator (LASSO) algorithm was applied to select features with the impact on predicting treatment response after the first TACE procedure. Six machine learning algorithms were used to build predictive models, including XGBoost, decision tree, support vector machine, random forest, k-nearest neighbor, and fully convolutional networks, and their performances were compared using receiver operator characteristic (ROC) curves to determine the best performing model. RESULTS: Following TACE, 31 patients (28.2%) were described as responsive to TACE, while 72 patients (71.8%) were nonresponsive to TACE. Portal vein tumor thrombosis type, albumin level, and distribution of tumors within the liver were selected for predictive model building. Among the models, the RF model showed the best performance, with area under the curve (AUC), accuracy, sensitivity, and specificity of 0.802, 0.784, 0.904, and 0.480, respectively. CONCLUSION: Machine learning models can provide an accurate prediction of the early response of initial TACE treatment for HCC, which can help in individualizing clinical decision-making and modification of further treatment strategies for patients with unresectable HCC.

Preoperative Prediction of Cytokeratin 19 Expression for Hepatocellular Carcinoma with Deep Learning Radiomics Based on Gadoxetic Acid-Enhanced Magnetic Resonance Imaging.

PURPOSE: Cytokeratin 19 (CK19) expression is a proven independent prognostic predictor of hepatocellular carcinoma (HCC). This study aimed to develop and validate the performance of a deep learning radiomics (DLR) model for CK19 identification in HCC based on preoperative gadoxetic acid-enhanced magnetic resonance imaging (MRI). PATIENTS AND METHODS: A total of 141 surgically confirmed HCCs with preoperative gadoxetic acid-enhanced MRI from two institutions were included. Prediction models were established based on hepatobiliary phase (HBP) images using a training set (n=102) and validated using time-independent (n=19) and external (n=20) test sets. A receiver operating characteristic curve was used to evaluate the performance for CK19 prediction. Recurrence-free survival (RFS) was also analyzed by incorporating the CK19 expression and other factors. RESULTS: For predicting CK19 expression, the area under the curve (AUC) of the DLR model was 0.820 (95% confidence interval [CI]: 0.732-0.907, P<0.001) with sensitivity, specificity, accuracy of 0.800, 0.766, and 0.775, respectively, and reached 0.781 in the external test set. Combined with alpha fetoprotein, the AUC increased to 0.833 (95% CI: 0.753-0.912, P<0.001) and the sensitivity was 0.960. Intratumoral hemorrhage and peritumoral hypointensity on HBP were independent risk factors for HCC recurrence by multivariate analysis. Based on predicted CK19 expression and the independent risk factors, a nomogram was developed to predict RFS and achieved C-index of 0.707. CONCLUSION: This study successfully established and verified an optimal DLR model for preoperative prediction of CK19-positive HCCs based on gadoxetic acid-enhanced MRI. The prediction of CK19 expression in HCC using a non-invasive method can help inform preoperative planning.

Predicting the recurrence risk of pancreatic neuroendocrine neoplasms after radical resection using deep learning radiomics with preoperative computed tomography images.

BACKGROUND: To establish and validate a prediction model for pancreatic neuroendocrine neoplasms (pNENs) recurrence after radical surgery with preoperative computed tomography (CT) images. METHODS: We retrospectively collected data from 74 patients with pathologically confirmed pNENs (internal group: 56 patients, Hospital I; external validation group: 18 patients, Hospital II). Using the internal group, models were trained with CT findings evaluated by radiologists, radiomics, and deep learning radiomics (DLR) to predict 5-year pNEN recurrence. Radiomics and DLR models were established for arterial (A), venous (V), and arterial and venous (A&V) contrast phases. The model with the optimal performance was further combined with clinical information, and all patients were divided into high- and low-risk groups to analyze survival with the Kaplan-Meier method. RESULTS: In the internal group, the areas under the curves (AUCs) of DLR-A, DLR-V, and DLR-A&V models were 0.80, 0.58, and 0.72, respectively. The corresponding radiomics AUCs were 0.74, 0.68, and 0.70. The AUC of the CT findings model was 0.53. The DLR-A model represented the optimum; added clinical information improved the AUC from 0.80 to 0.83. In the validation group, the AUCs of DLR-A, DLR-V, and DLR-A&V models were 0.77, 0.48, and 0.64, respectively, and those of radiomics-A, radiomics-V, and radiomics-A&V models were 0.56, 0.52, and 0.56, respectively. The AUC of the CT findings model was 0.52. In the validation group, the comparison between the DLR-A and the random models showed a trend of significant difference (P=0.058). Recurrence-free survival differed significantly between high- and low-risk groups (P=0.003). CONCLUSIONS: Using DLR, we successfully established a preoperative recurrence prediction model for pNEN patients after radical surgery. This allows a risk evaluation of pNEN recurrence, optimizing clinical decision-making.

Regional liver function analysis with gadoxetic acid-enhanced MRI and virtual hepatectomy: prediction of postoperative short-term outcomes for HCC.

OBJECTIVES: To explore the role of quantitative regional liver function assessed by preoperative gadoxetic acid-enhanced MRI with computer-aided virtual hepatectomy to predict short-term outcomes after major hepatectomy for HCC. METHODS: We retrospectively reviewed the records of 133 consecutive patients with HCC who underwent preoperative gadoxetic acid-enhanced MRI and indocyanine green (ICG) test. Forty-five patients received open major hepatectomy. Liver function reserve and the future liver remnant were evaluated by computer-aided virtual hepatectomy. Global liver functional parameters included the T1 relaxation time reduction rate (T1ratio) and functional liver volume (FV), whereas regional parameters included the rT1pos, rT1ratio, remnant FV (rFV), and remnant FV ratio (rFVratio) of the remnant liver. The functional parameters of the MRI and ICG were used to predict the short-term outcomes (liver failure and major complications) after major hepatectomy. RESULTS: The T1ratio and FV were correlated with the ICG test (rho = - 0.304 and - 0.449, p < 0.05). FV < 682.8 ml indicated preoperative ICG-R15 ≥ 14% with 0.765 value of the area under the curve (AUC). No patient who underwent major resection with good liver functional reserve (ICG < 14%) and enough future remnant volume (> 30% standard LV) developed liver failure. Low rT1ratio (< 66.5%) and high rT1pos (> 217.5 ms) may predict major complications (AUC = 0.831 and 0.756, respectively; p < 0.05). The rT1ratio was an independent risk factor for postoperative major complications (odds ratio [OR] = 0.845, 95% CI, 0.736-0.966; p < 0.05). CONCLUSION: Preoperative gadoxetic acid-enhanced MRI with computer-aided virtual hepatectomy may facilitate optimal assessment of regional liver functional reserve to predict short-term outcomes after major hepatectomy for HCC. KEY POINTS: • Preoperative gadoxetic acid-enhanced MRI with virtual hepatectomy and volumetric analysis can provide precise liver volume and regional functional assessment. • Quantitative regional liver function assessed by gadoxetic acid-enhanced MRI can predict the short-term outcomes after major hepatectomy in patients with HCC. • The regional liver function assessed by gadoxetic acid-enhanced MRI is an independent risk factor for postoperative major complications.

A Transfer Learning Radiomics Nomogram for Preoperative Prediction of Borrmann Type IV Gastric Cancer From Primary Gastric Lymphoma.

OBJECTIVE: This study aims to differentiate preoperative Borrmann type IV gastric cancer (GC) from primary gastric lymphoma (PGL) by transfer learning radiomics nomogram (TLRN) with whole slide images of GC as source domain data. MATERIALS AND METHODS: This study retrospectively enrolled 438 patients with histopathologic diagnoses of Borrmann type IV GC and PGL. They received CT examinations from three hospitals. Quantitative transfer learning features were extracted by the proposed transfer learning radiopathomic network and used to construct transfer learning radiomics signatures (TLRS). A TLRN, which integrates TLRS, clinical factors, and CT subjective findings, was developed by multivariate logistic regression. The diagnostic TLRN performance was assessed by clinical usefulness in the independent validation set. RESULTS: The TLRN was built by TLRS and a high enhanced serosa sign, which showed good agreement by the calibration curve. The TLRN performance was superior to the clinical model and TLRS. Its areas under the curve (AUC) were 0.958 (95% confidence interval [CI], 0.883-0.991), 0.867 (95% CI, 0.794-0.922), and 0.921 (95% CI, 0.860-0.960) in the internal and two external validation cohorts, respectively. Decision curve analysis (DCA) showed that the TLRN was better than any other model. TLRN has potential generalization ability, as shown in the stratification analysis. CONCLUSIONS: The proposed TLRN based on gastric WSIs may help preoperatively differentiate PGL from Borrmann type IV GC.Borrmann type IV gastric cancer, primary gastric lymphoma, transfer learning, whole slide image, deep learning.

A Case of a Huge Inferior Vena Cava Leiomyosarcoma: Precise Preoperative Evaluation with Gadobutrol-Enhanced MRI.

BACKGROUND: Leiomyosarcoma of the inferior vena cava (IVC) is a rare malignant tumour with poor prognosis. Surgical resection is the first line of treatment to achieve the best possible outcome. However, precise preoperative evaluation is essential to guide therapeutic decisions. Here, the preoperative evaluation potential of gadobutrol-enhanced magnetic resonance imaging (MRI) was assessed in the management of a 42-year-old patient with a large IVC mass. METHODS: The patient first underwent enhanced computed tomography (CT), but the relationship between the left renal vein and the mass in the dilated IVC was ambiguous, and it remained unclear whether the right hepatic vein was invaded by the lesion. To make a precise assessment of the tumour, the patient subsequently underwent high-resolution MRI angiography examination combined with high-concentration contrast medium gadobutrol. RESULTS: MRI demonstrated the integrity of the right hepatic vein and the left renal vein. Following a multidisciplinary consultation, a complicated surgery including complete resection of the mass, artificial vessel replacement of IVC, total hepatectomy, and bilateral nephrectomy with liver and kidney auto-transplantation was performed successfully. The surgical plan formulated after reviewing the MRI preoperatively was adhered to the course of the surgery. Postoperative CT re-examination showed that the blood flow of the artificial blood vessel and the right hepatic vein was unobstructed. Histopathological examination confirmed the tumour to be a leiomyosarcoma. CONCLUSION: Preoperative imaging diagnosis and assessment have important implications for the surgical planning of IVC leiomyosarcoma. High-resolution MRI angiography examination with high concentration contrast medium gadobutrol may be of particular benefit in IVC tumours.