Feasibility and effectiveness of automatic deep learning network and radiomics models for differentiating tumor stroma ratio in pancreatic ductal adenocarcinoma.

OBJECTIVE: This study aims to compare the feasibility and effectiveness of automatic deep learning network and radiomics models in differentiating low tumor stroma ratio (TSR) from high TSR in pancreatic ductal adenocarcinoma (PDAC). METHODS: A retrospective analysis was conducted on a total of 207 PDAC patients from three centers (training cohort: n = 160; test cohort: n = 47). TSR was assessed on hematoxylin and eosin-stained specimens by experienced pathologists and divided as low TSR and high TSR. Deep learning and radiomics models were developed including ShuffulNetV2, Xception, MobileNetV3, ResNet18, support vector machine (SVM), k-nearest neighbor (KNN), random forest (RF), and logistic regression (LR). Additionally, the clinical models were constructed through univariate and multivariate logistic regression. Kaplan-Meier survival analysis and log-rank tests were conducted to compare the overall survival time between different TSR groups. RESULTS: To differentiate low TSR from high TSR, the deep learning models based on ShuffulNetV2, Xception, MobileNetV3, and ResNet18 achieved AUCs of 0.846, 0.924, 0.930, and 0.941, respectively, outperforming the radiomics models based on SVM, KNN, RF, and LR with AUCs of 0.739, 0.717, 0.763, and 0.756, respectively. Resnet 18 achieved the best predictive performance. The clinical model based on T stage alone performed worse than deep learning models and radiomics models. The survival analysis based on 142 of the 207 patients demonstrated that patients with low TSR had longer overall survival. CONCLUSIONS: Deep learning models demonstrate feasibility and superiority over radiomics in differentiating TSR in PDAC. The tumor stroma ratio in the PDAC microenvironment plays a significant role in determining prognosis. CRITICAL RELEVANCE STATEMENT: The objective was to compare the feasibility and effectiveness of automatic deep learning networks and radiomics models in identifying the tumor-stroma ratio in pancreatic ductal adenocarcinoma. Our findings demonstrate deep learning models exhibited superior performance compared to traditional radiomics models. KEY POINTS: • Deep learning demonstrates better performance than radiomics in differentiating tumor-stroma ratio in pancreatic ductal adenocarcinoma. • The tumor-stroma ratio in the pancreatic ductal adenocarcinoma microenvironment plays a protective role in prognosis. • Preoperative prediction of tumor-stroma ratio contributes to clinical decision-making and guiding precise medicine.

Solitary lung adenocarcinoma: follow-up CT, pathological-molecular characteristics, and surgical prognosis for different morphological classifications.

OBJECTIVE: To investigate the dynamic changes during follow-up computed tomography (CT), histological subtypes, gene mutation status, and surgical prognosis for different morphological presentations of solitary lung adenocarcinomas (SLADC). MATERIALS AND METHODS: This retrospective study compared dynamic tumor changes and volume doubling time (VDT) in 228 patients with SLADC (morphological types I-IV) who had intermittent growth during follow-ups. The correlation between the morphological classification and histological subtypes, gene mutation status, and surgical prognosis was evaluated. RESULTS: Among the 228 patients, 66 (28.9%) were classified as type I, 123 (53.9%) as type II, 16 (7%) as type III, and 23 (10.1%) as type IV. Type I had the shortest VDT (254 days), followed by types IV (381 days) and III (501 days), and then type II (993 days) (p < 0.05 each). Type I had a greater proportion of solid/micropapillary-predominant pattern than type II, and the lepidic-predominant pattern was more common in type II and III than in type I (p < 0.05 each). Furthermore, type II and IV SLADCs were correlated with positive epidermal growth factor receptor mutation (p < 0.05 each). Lastly, the Kaplan-Meier curves showed that the disease-free survival was longest for patients with type II tumors, followed by those with type III and IV tumors, and then those with type I tumors (p < 0.001 each). CONCLUSION: A good understanding of the natural progression and pathological-molecular characteristics of different morphological SLADC types can help make accurate diagnoses, develop individual treatment strategies, and predict patient outcomes. CRITICAL RELEVANCE STATEMENT: A good understanding of the natural progression and pathological-molecular characteristics of different morphological solitary lung adenocarcinoma types can help make accurate diagnoses, develop individual treatment strategies, and predict patient outcomes. KEY POINTS: • Type I-IV solitary lung adenocarcinomas exhibit varying natural progression on serial CT scans. • Morphological classification of solitary lung adenocarcinomas predicts histological subtype, gene status, and surgical prognosis. • This classification of solitary lung adenocarcinomas may help improve diagnostic, therapeutic, and prognosticating abilities.

Dual-energy CT of the pancreas: comparison between virtual non-contrast images and true non-contrast images in the detection of pancreatic lesion.

PURPOSE: To evaluate the image quality and diagnostic performance for pancreatic lesion between true non-contrast (TNC) and virtual non-contrast (VNC) images obtained from the dual-energy computed tomography (DECT). METHODS: One hundred six patients with pancreatic mass underwent contrast-enhanced DECT examinations were retrospectively included in this study. VNC images of the abdomen were generated from late arterial (aVNC) and portal (pVNC) phases. For quantitative analysis, the attenuation differences and reproducibility of abdominal organs were compared between TNC and aVNC/pVNC measurements. Qualitatively image quality was assessed by two radiologists using a five-point scale, and they independently compared the detection accuracy of pancreatic lesions between TNC and aVNC/pVNC images. The volume CT dose index (CTDIvol) and size-specific dose estimates (SSDE) were recorded to evaluate the potential dose reduction when using VNC reconstruction to replace the unenhanced phase. RESULTS: A total of 78.38% (765/976) of the attenuation measurement pairs were reproducible between TNC and aVNC images, and 71.0% (693/976) between TNC and pVNC images. In triphasic examinations, a total of 108 pancreatic lesions were found in 106 patients, and no significant difference in detection accuracy was found between TNC and VNC images (p = 0.587-0.957). Qualitatively, image quality was rated diagnostic (score ≥ 3) in all the VNC images. Calculated CTDIvol and SSDE reduction of about 34% could be achieved by omitting the non-contrast phase. CONCLUSION: VNC images of DECT provide diagnostic image quality and accurate pancreatic lesions detection, which are a promising alternative to unenhanced phase with a substantial reduction of radiation exposure in clinical routine.

One 3D VOI-based deep learning radiomics strategy, clinical model and radiologists for predicting lymph node metastases in pancreatic ductal adenocarcinoma based on multiphasic contrast-enhanced computer tomography.

Purpose: We designed to construct one 3D VOI-based deep learning radiomics strategy for identifying lymph node metastases (LNM) in pancreatic ductal adenocarcinoma on the basis of multiphasic contrast-enhanced computer tomography and to assist clinical decision-making. Methods: This retrospective research enrolled 139 PDAC patients undergoing pre-operative arterial phase and venous phase scanning examination between 2015 and 2021. A primary group (training group and validation group) and an independent test group were divided. The DLR strategy included three sections. (1) Residual network three dimensional-18 (Resnet 3D-18) architecture was constructed for deep learning feature extraction. (2) Least absolute shrinkage and selection operator model was used for feature selection. (3) Fully connected network served as the classifier. The DLR strategy was applied for constructing different 3D CNN models using 5-fold cross-validation. Radiomics scores (Rad score) were calculated for distinguishing the statistical difference between negative and positive lymph nodes. A clinical model was constructed by combining significantly different clinical variables using univariate and multivariable logistic regression. The manifestation of two radiologists was detected for comparing with computer-developed models. Receiver operating characteristic curves, the area under the curve, accuracy, precision, recall, and F1 score were used for evaluating model performance. Results: A total of 45, 49, and 59 deep learning features were selected via LASSO model. No matter in which 3D CNN model, Rad score demonstrated the deep learning features were significantly different between non-LNM and LNM groups. The AP+VP DLR model yielded the best performance in predicting status of lymph node in PDAC with an AUC of 0.995 (95% CI:0.989-1.000) in training group; an AUC of 0.940 (95% CI:0.910-0.971) in validation group; and an AUC of 0.949 (95% CI:0.914-0.984) in test group. The clinical model enrolled the histological grade, CA19-9 level and CT-reported tumor size. The AP+VP DLR model outperformed AP DLR model, VP DLR model, clinical model, and two radiologists. Conclusions: The AP+VP DLR model based on Resnet 3D-18 demonstrated excellent ability for identifying LNM in PDAC, which could act as a non-invasive and accurate guide for clinical therapeutic strategies. This 3D CNN model combined with 3D tumor segmentation technology is labor-saving, promising, and effective.

Dual-energy CT with virtual monoenergetic images and iodine maps improves tumor conspicuity in patients with pancreatic ductal adenocarcinoma.

OBJECTIVES: To evaluate the value of monoenergetic images (MEI [+]) and iodine maps in dual-source dual-energy computed tomography (DECT) for assessing pancreatic ductal adenocarcinoma (PDAC), including the visually isoattenuating PDAC. MATERIALS AND METHODS: This retrospective study included 75 PDAC patients, who underwent contrast-enhanced DECT examinations. Conventional polyenergetic image (PEI) and 40-80 keV MEI (+) (10-keV increments) were reconstructed. The tumor contrast, contrast-to-noise ratio (CNR) of the tumor and peripancreatic vessels, the signal-to-noise ratio (SNR) of the pancreas and tumor, and the tumor diameters were quantified. On iodine maps, the normalized iodine concentration (NIC) in the tumor and parenchyma was compared. For subjective analysis, two radiologists independently evaluated images on a 5-point scale. RESULTS: All the quantitative parameters were maximized at 40-keV MEI (+) and decreased gradually with increasing energy. The tumor contrast, SNR of pancreas and CNRs in 40-60 keV MEI (+) were significantly higher than those in PEI (p < 0.05). For visually isoattenuating PDAC, 40-50 keV MEI (+) provided significantly higher tumor CNR compared to PEI (p < 0.05). The reproducibility in tumor measurements was highest in 40-keV MEI (+) between the two radiologists. The tumor and parenchyma NIC were 1.28 ± 0.65 and 3.38 ± 0.72 mg/mL, respectively (p < 0.001). 40-50 keV MEI (+) provided the highest subjective scores, compared to PEI (p < 0.001). CONCLUSIONS: Low-keV MEI (+) of DECT substantially improves the subjective and objective image quality and consistency of tumor measurements in patients with PDAC. Combining the low-keV MEI (+) and iodine maps may yield diagnostically adequate tumor conspicuity in visually isoattenuating PDAC.

Comparison of Multiple Radiomics Models for Identifying Histological Grade of Pancreatic Ductal Adenocarcinoma Preoperatively Based on Multiphasic Contrast-Enhanced Computed Tomography: A Two-Center Study in Southwest China.

BACKGROUND: We designed and validated the value of multiple radiomics models for diagnosing histological grade of pancreatic ductal adenocarcinoma (PDAC), holding a promise of assisting in precision medicine and providing clinical therapeutic strategies. METHODS: 198 PDAC patients receiving surgical resection and pathological confirmation were enrolled and classified as 117 low-grade PDAC and 81 high-grade PDAC group. An external validation group was used to assess models' performance. Available radiomics features were selected using GBDT algorithm on the basis of the arterial and venous phases, respectively. Five different machine learning models were built including k-nearest neighbour, logistic regression, naive bayes model, support vector machine, and random forest using ten times tenfold cross-validation. Multivariable logistic regression analysis was applied to establish clinical model and combined model. The models' performance was assessed according to its predictive performance, calibration curves, and decision curves. A nomogram was established for visualization. Survival analysis was conducted for stratifying the overall survival prior to treatment. RESULTS: In the training group, the RF model demonstrated the optimal predictive ability and robustness with an AUC of 0.943; the SVM model achieved the secondary performance, followed by Bayes model. In the external validation group, these three models (Bayes, RF, SVM) also achieved the top three predictive ability. A clinical model was built by selected clinical features with an AUC of 0.728, and combined model was established by an RF model and a clinical model with an AUC of 0.961. The log-rank test revealed that the low-grade group survived longer than the high-grade group. CONCLUSIONS: The multiphasic CECT radiomics models offered an accurate and noninvasive perspective to differentiate histological grade in PDAC and advantages of machine learning models including RF, SVM and Bayes were more remarkable.

A Comparative Study of 2 Different Segmentation Methods of ADC Histogram for Differentiation Genetic Subtypes in Lower-Grade Diffuse Gliomas.

BACKGROUND: To evaluate the diagnostic performance of apparent diffusion coefficient (ADC) histogram parameters for differentiating the genetic subtypes in lower-grade diffuse gliomas and explore which segmentation method (ROI-1, the entire tumor ROI; ROI2, the tumor ROI excluding cystic and necrotic portions) performs better. MATERIALS AND METHODS: We retrospectively evaluated 56 lower-grade diffuse gliomas and divided them into three categories: IDH-wild group (IDHwt, 16cases); IDH mutant with the intact 1p or 19q group (IDHmut/1p19q+, 18cases); and IDH mutant with the 1p/19q codeleted group (IDHmut/1p19q-, 22cases). Histogram parameters of ADC maps calculated with the two different ROI methods: ADCmean, min, max, mode, P5, P10, P25, P75, P90, P95, kurtosis, skewness, entropy, StDev, and inhomogenity were compared between these categories using the independent t test or Mann-Whitney U test. For statistically significant results, a receiver operating characteristic (ROC) curves were constructed, and the optimal cutoff value was determined by maximizing Youden's index. Area under the curve (AUC) results were compared using the method of Delong et al. RESULTS: The inhomogenity from the two different ROI methods for distinguishing IDHwt gliomas from IDHmut gliomas both showed the biggest AUC (0.788, 0.930), the optimal cutoff value was 0.229 (sensitivity, 81.3%; specificity, 75.0%) for the ROI-1 and 0.186 (sensitivity, 93.8%; specificity, 82.5%) for the ROI-2, and the AUC of the inhomogenity from the ROI-2 was significantly larger than that from another segmentation, but no significant differences were identified between the AUCs of other same parameters from the two different ROI methods. For the differentiaiton of IDHmut/1p19q- tumors and IDHmut/1p19q+ tumors, with the ROI-1, the ADCmode showed the biggest AUC (AUC: 0.784; sensitivity, 61.1%; specificity, 90.9%), with the ROI-2, and the skewness performed best (AUC, 0.821; sensitivity, 81.8%; specificity, 77.8%), but no significant differences were identified between the AUCs of the same parameters from the two different ROI methods. CONCLUSION: ADC values analyzed by the histogram method could help to classify the genetic subtypes in lower-grade diffuse gliomas, no matter which ROI method was used. Extracting cystic and necrotic portions from the entire tumor lesions is preferable for evaluating the difference of the intratumoral heterogeneity and classifying IDH-wild tumors, but not significantly beneficial to predicting the 1p19q genotype in the lower-grade gliomas.

[Apparent Diffusion Coefficient Histogram Analysis:Differentiation of Genetic Subtypes of Diffuse Lower-grade Gliomas].

Objective To explore the utility of apparent diffusion coefficient(ADC)histogram analysis for differentiating genetic subtypes of diffuse lower-grade gliomas. Methods A total of 55 patients with WHO grade Ⅱ/Ⅲ diffuse lower-grade gliomas who underwent preoperative routine brain magnetic resonance imaging and diffusion weighted imaging in our center were retrospectively evaluated.Among whom there were 14 patients with isocitrate dehydrogenase(IDH)wild-type gliomas(IDH wt group),19 patients with IDH-mutant 1p19q intact gliomas(IDH mut1p19q int group),and 22 patients with IDH-mutant 1p19q co-deleted gliomas(IDH mut1p19q del group).The whole-lesion ADC values derived from histogram analysis(including ADCmean,ADCminimum,ADC5%,ADC10%,ADC25%,ADC50%,ADC75%,ADC90%,ADC95%,ADCmaximum,mode,range,skewness,kurtosis,standard deviation,inhomogeneity,and entrophy)were measured for each patient.All parameters between the different genetic subtypes were compared by using the Student's t test or Mann-Whitney U test.Receiver operating curve(ROC)analysis was used to assess the diagnostic performance of ADC histogram in distinguishing the different genetic subtypes. Results Compared with IDH wt group,the ADC75%(P=0.021),ADC90%(P=0.015),ADC95%(P=0.014),ADCmaximum (P=0.035),range(P=0.009),standard deviation(P=0.001)and inhomogeneity(P=0.001)were significantly lower in IDH mut group;in contrast,the ADCminimum (P=0.031)and kurtosis(P=0.020)of IDH mut group were significantly higher than those in IDH wt group.The ADCmean(P=0.010),ADC5%(P=0.016),ADC10%(P=0.012),ADC25%(P=0.007),ADC50%(P=0.005),ADC75%(P=0.015),and mode(P=0.002)were significantly higher in IDH mut1p19q int group than in IDH mut1p19q del group.Inhomogeneity achieved the highest area under ROC(AUC)(0.811)in differentiating IDH mut gliomas and IDH wt gliomas,with a cutoff value of 0.229;the sensitivity and specificity were 85.7% and 73.2%.The mode achieved the highest AUC(0.744)in differentiating IDH mut1p19q int gliomas and IDH mut1p19q del gliomas,with a cutoff value was 1448.75×10 -6 mm 2/s;the sensitivity and specificity were 57.9% and 90.9%.Conclusion ADC histograms analysis may be helpful to differentiate genetic subtypes in lower-grade gliomas.