Robustly federated learning model for identifying high-risk patients with postoperative gastric cancer recurrence.

The prediction of patient disease risk via computed tomography (CT) images and artificial intelligence techniques shows great potential. However, training a robust artificial intelligence model typically requires large-scale data support. In practice, the collection of medical data faces obstacles related to privacy protection. Therefore, the present study aims to establish a robust federated learning model to overcome the data island problem and identify high-risk patients with postoperative gastric cancer recurrence in a multicentre, cross-institution setting, thereby enabling robust treatment with significant value. In the present study, we collect data from four independent medical institutions for experimentation. The robust federated learning model algorithm yields area under the receiver operating characteristic curve (AUC) values of 0.710, 0.798, 0.809, and 0.869 across four data centres. Additionally, the effectiveness of the algorithm is evaluated, and both adaptive and common features are identified through analysis.

A transfer learning nomogram for predicting prostate cancer and benign conditions on MRI.

BACKGROUND: Deep learning has been used to detect or characterize prostate cancer (PCa) on medical images. The present study was designed to develop an integrated transfer learning nomogram (TLN) for the prediction of PCa and benign conditions (BCs) on magnetic resonance imaging (MRI). METHODS: In this retrospective study, a total of 709 patients with pathologically confirmed PCa and BCs from two institutions were included and divided into training (n = 309), internal validation (n = 200), and external validation (n = 200) cohorts. A transfer learning signature (TLS) that was pretrained with the whole slide images of PCa and fine-tuned on prebiopsy MRI images was constructed. A TLN that integrated the TLS, the Prostate Imaging-Reporting and Data System (PI-RADS) score, and the clinical factor was developed by multivariate logistic regression. The performance of the TLS, clinical model (CM), and TLN were evaluated in the validation cohorts using the receiver operating characteristic (ROC) curve, the Delong test, the integrated discrimination improvement (IDI), and decision curve analysis. RESULTS: TLS, PI-RADS score, and age were selected for TLN construction. The TLN yielded areas under the curve of 0.9757 (95% CI, 0.9613-0.9902), 0.9255 (95% CI, 0.8873-0.9638), and 0.8766 (95% CI, 0.8267-0.9264) in the training, internal validation, and external validation cohorts, respectively, for the discrimination of PCa and BCs. The TLN outperformed the TLS and the CM in both the internal and external validation cohorts. The decision curve showed that the TLN added more net benefit than the CM. CONCLUSIONS: The proposed TLN has the potential to be used as a noninvasive tool for PCa and BCs differentiation.

Development and validation of a deep learning radiomics nomogram for preoperatively differentiating thymic epithelial tumor histologic subtypes.

OBJECTIVES: Using contrast-enhanced computed tomography (CECT) and deep learning technology to develop a deep learning radiomics nomogram (DLRN) to preoperative predict risk status of patients with thymic epithelial tumors (TETs). METHODS: Between October 2008 and May 2020, 257 consecutive patients with surgically and pathologically confirmed TETs were enrolled from three medical centers. We extracted deep learning features from all lesions using a transformer-based convolutional neural network and created a deep learning signature (DLS) using selector operator regression and least absolute shrinkage. The predictive capability of a DLRN incorporating clinical characteristics, subjective CT findings and DLS was evaluated by the area under the curve (AUC) of a receiver operating characteristic curve. RESULTS: To construct a DLS, 25 deep learning features with non-zero coefficients were selected from 116 low-risk TETs (subtypes A, AB, and B1) and 141 high-risk TETs (subtypes B2, B3, and C). The combination of subjective CT features such as infiltration and DLS demonstrated the best performance in differentiating TETs risk status. The AUCs in the training, internal validation, external validation 1 and 2 cohorts were 0.959 (95% confidence interval [CI]: 0.924-0.993), 0.868 (95% CI: 0.765-0.970), 0.846 (95% CI: 0.750-0.942), and 0.846 (95% CI: 0.735-0.957), respectively. The DeLong test and decision in curve analysis revealed that the DLRN was the most predictive and clinically useful model. CONCLUSIONS: The DLRN comprised of CECT-derived DLS and subjective CT findings showed a high performance in predicting risk status of patients with TETs. CLINICAL RELEVANCE STATEMENT: Accurate risk status assessment of thymic epithelial tumors (TETs) may aid in determining whether preoperative neoadjuvant treatment is necessary. A deep learning radiomics nomogram incorporating enhancement CT-based deep learning features, clinical characteristics, and subjective CT findings has the potential to predict the histologic subtypes of TETs, which can facilitate decision-making and personalized therapy in clinical practice. KEY POINTS: • A non-invasive diagnostic method that can predict the pathological risk status may be useful for pretreatment stratification and prognostic evaluation in TET patients. • DLRN demonstrated superior performance in differentiating the risk status of TETs when compared to the deep learning signature, radiomics signature, or clinical model. • The DeLong test and decision in curve analysis revealed that the DLRN was the most predictive and clinically useful in differentiating the risk status of TETs.

Identifying Solitary Granulomatous Nodules from Solid Lung Adenocarcinoma: Exploring Robust Image Features with Cross-Domain Transfer Learning.

PURPOSE: This study aimed to find suitable source domain data in cross-domain transfer learning to extract robust image features. Then, a model was built to preoperatively distinguish lung granulomatous nodules (LGNs) from lung adenocarcinoma (LAC) in solitary pulmonary solid nodules (SPSNs). METHODS: Data from 841 patients with SPSNs from five centres were collected retrospectively. First, adaptive cross-domain transfer learning was used to construct transfer learning signatures (TLS) under different source domain data and conduct a comparative analysis. The Wasserstein distance was used to assess the similarity between the source domain and target domain data in cross-domain transfer learning. Second, a cross-domain transfer learning radiomics model (TLRM) combining the best performing TLS, clinical factors and subjective CT findings was constructed. Finally, the performance of the model was validated through multicentre validation cohorts. RESULTS: Relative to other source domain data, TLS based on lung whole slide images as source domain data (TLS-LW) had the best performance in all validation cohorts (AUC range: 0.8228-0.8984). Meanwhile, the Wasserstein distance of TLS-LW was 1.7108, which was minimal. Finally, TLS-LW, age, spiculated sign and lobulated shape were used to build the TLRM. In all validation cohorts, The AUC ranges were 0.9074-0.9442. Compared with other models, decision curve analysis and integrated discrimination improvement showed that TLRM had better performance. CONCLUSIONS: The TLRM could assist physicians in preoperatively differentiating LGN from LAC in SPSNs. Furthermore, compared with other images, cross-domain transfer learning can extract robust image features when using lung whole slide images as source domain data and has a better effect.

Distinguishing common renal cell carcinomas from benign renal tumors based on machine learning: comparing various CT imaging phases, slices, tumor sizes, and ROI segmentation strategies.

OBJECTIVES: To determine whether a CT-based machine learning (ML) can differentiate benign renal tumors from renal cell carcinomas (RCCs) and improve radiologists' diagnostic performance, and evaluate the impact of variable CT imaging phases, slices, tumor sizes, and region of interest (ROI) segmentation strategies. METHODS: Patients with pathologically proven RCCs and benign renal tumors from our institution between 2008 and 2020 were included as the training dataset for ML model development and internal validation (including 418 RCCs and 78 benign tumors), and patients from two independent institutions and a public database (TCIA) were included as the external dataset for individual testing (including 262 RCCs and 47 benign tumors). Features were extracted from three-phase CT images. CatBoost was used for feature selection and ML model establishment. The area under the receiver operating characteristic curve (AUC) was used to assess the performance of the ML model. RESULTS: The ML model based on 3D images performed better than that based on 2D images, with the highest AUC of 0.81 and accuracy (ACC) of 0.86. All three radiologists achieved better performance by referring to the classifier's decision, with accuracies increasing from 0.82 to 0.87, 0.82 to 0.88, and 0.76 to 0.87. The ML model achieved higher negative predictive values (NPV, 0.82-0.99), and the radiologists achieved higher positive predictive values (PPV, 0.91-0.95). CONCLUSIONS: A ML classifier based on whole-tumor three-phase CT images can be a useful and promising tool for differentiating RCCs from benign renal tumors. The ML model also perfectly complements radiologist interpretations. KEY POINTS: • A machine learning classifier based on CT images could be a reliable way to differentiate RCCs from benign renal tumors. • The machine learning model perfectly complemented the radiologists' interpretations. • Subtle variances in ROI delineation had little effect on the performance of the ML classifier.

Deep learning nomogram for predicting lymph node metastasis using computed tomography image in cervical cancer.

BACKGROUND: Deep learning (DL) has been used on medical images to grade, differentiate, and predict prognosis in many tumors. PURPOSE: To explore the effect of computed tomography (CT)-based deep learning nomogram (DLN) for predicting cervical cancer lymph node metastasis (LNM) before surgery. MATERIAL AND METHODS: In total, 418 patients with stage IB-IIB cervical cancer were retrospectively enrolled for model exploration (n = 296) and internal validation (n = 122); 62 patients from another independent institution were enrolled for external validation. A convolutional neural network (CNN) was used for DL features extracting from all lesions. The least absolute shrinkage and selection operator (Lasso) logistic regression was used to develop a deep learning signature (DLS). A DLN incorporating the DLS and clinical risk factors was proposed to predict LNM individually. The performance of the DLN was evaluated on internal and external validation cohorts. RESULTS: Stage, CT-reported pelvic lymph node status, and DLS were found to be independent predictors and could be used to construct the DLN. The combination showed a better performance than the clinical model and DLS. The proposed DLN had an area under the curve (AUC) of 0.925 in the training cohort, 0.771 in the internal validation cohort, and 0.790 in the external validation cohort. Decision curve analysis and stratification analysis suggested that the DLN has potential ability to generate a personalized probability of LNM in cervical cancer. CONCLUSION: The proposed CT-based DLN could be used as a personalized non-invasive tool for preoperative prediction of LNM in cervical cancer, which could facilitate the choice of clinical treatment methods.

Predicting lymphovascular invasion in clinically node-negative breast cancer detected by abbreviated magnetic resonance imaging: Transfer learning vs. radiomics.

Objective: To compare the performance of abbreviated breast magnetic resonance imaging (AB-MRI)-based transfer learning (TL) algorithm and radionics analysis for lymphovascular invasion (LVI) prediction in patients with clinically node-negative invasive breast cancer (IBC). Methods: Between November 2017 and October 2020, 233 clinically node-negative IBCs detected by AB-MRI were retrospectively enrolled. One hundred thirty IBCs from center 1 (37 LVI-positive and 93 LVI-negative) were assigned as the training cohort and 103 from center 2 (25 LVI-positive and 78 LVI-negative) as the validation cohort. Based on AB-MRI, a TL signature (TLS) and a radiomics signature (RS) were built with the least absolute shrinkage and selection operator (LASSO) logistic regression. Their diagnostic performances were validated and compared using areas under the receiver operating curve (AUCs), net reclassification improvement (NRI), integrated discrimination improvement (IDI), decision curve analysis (DCA), and stratification analysis. A convolutional filter visualization technique was used to map the response areas of LVI on the AB-MRI. Results: In the validation cohort, compared with RS, the TLS showed better capability in discriminating LVI-positive from LVI-negative lesions (AUC: 0.852 vs. 0.726, p < 0.001; IDI = 0.092, p < 0.001; NRI = 0.554, p < 0.001). The diagnostic performance of TLS was not affected by the menstrual state, molecular subtype, or contrast agent type (all p > 0.05). Moreover, DCA showed that the TLS added more net benefit than RS for clinical utility. Conclusions: An AB-MRI-based TLS was superior to RS for preoperative LVI prediction in patients with clinically node-negative IBC.

A brain-like classification method for computed tomography images based on adaptive feature matching dual-source domain heterogeneous transfer learning.

Transfer learning can improve the robustness of deep learning in the case of small samples. However, when the semantic difference between the source domain data and the target domain data is large, transfer learning easily introduces redundant features and leads to negative transfer. According the mechanism of the human brain focusing on effective features while ignoring redundant features in recognition tasks, a brain-like classification method based on adaptive feature matching dual-source domain heterogeneous transfer learning is proposed for the preoperative aided diagnosis of lung granuloma and lung adenocarcinoma for patients with solitary pulmonary solid nodule in the case of small samples. The method includes two parts: (1) feature extraction and (2) feature classification. In the feature extraction part, first, By simulating the feature selection mechanism of the human brain in the process of drawing inferences about other cases from one instance, an adaptive selected-based dual-source domain feature matching network is proposed to determine the matching weight of each pair of feature maps and each pair of convolution layers between the two source networks and the target network, respectively. These two weights can, respectively, adaptive select the features in the source network that are conducive to the learning of the target task, and the destination of feature transfer to improve the robustness of the target network. Meanwhile, a target network based on diverse branch block is proposed, which made the target network have different receptive fields and complex paths to further improve the feature expression ability of the target network. Second, the convolution kernel of the target network is used as the feature extractor to extract features. In the feature classification part, an ensemble classifier based on sparse Bayesian extreme learning machine is proposed that can automatically decide how to combine the output of base classifiers to improve the classification performance. Finally, the experimental results (the AUCs were 0.9542 and 0.9356, respectively) on the data of two center data show that this method can provide a better diagnostic reference for doctors.

A CT-based deep learning model for subsolid pulmonary nodules to distinguish minimally invasive adenocarcinoma and invasive adenocarcinoma.

OBJECTIVE: To develop and validate a deep learning nomogram (DLN) model constructed from non-contrast computed tomography (CT) images for discriminating minimally invasive adenocarcinoma (MIA) from invasive adenocarcinoma (IAC) in patients with subsolid pulmonary nodules (SSPNs). MATERIALS AND METHODS: In total, 365 consecutive patients who presented with SSPNs and were pathologically diagnosed with MIA or IAC after surgery, were recruited from two medical institutions from 2016 to 2019. Deep learning features were selected from preoperative CT images using convolutional neural network. Deep learning signature (DLS) was developed via the least absolute shrinkage and selection operator (LASSO). New DLN integrating clinical variables, subjective CT findings, and DLS was constructed. The diagnostic efficiency and discriminative capability were analyzed using the receiver operating characteristic method and decision curve analysis (DCA). RESULTS: In total, 18 deep learning features with non-zero coefficients were enrolled to develop the DLS, which was statistically different between the MIA and IAC groups. Independent predictors of DLS and lobulated sharp were used to build the DLN. The areas under the curves of the DLN were 0.889 (95% confidence interval (CI): 0.824-0.936), 0.915 (95% CI: 0.846-0.959), and 0.914 (95% CI: 0.848-0.958) in the training, internal validation, and external validation cohorts, respectively. After stratification analysis and DCA, the DLN showed potential generalization ability. CONCLUSION: The DLN incorporating the DLS and subjective CT findings have strong potential to distinguish MIA from IAC in patients with SSPNs, and will facilitate the suitable treatment method selection for the management of SSPNs.

Computed Tomography-Based Radiomics Nomogram: Potential to Predict Local Recurrence of Gastric Cancer After Radical Resection.

OBJECTIVE: Accurate prediction of postoperative recurrence risk of gastric cancer (GC) is critical for individualized precision therapy. We aimed to investigate whether a computed tomography (CT)-based radiomics nomogram can be used as a tool for predicting the local recurrence (LR) of GC after radical resection. MATERIALS AND METHODS: 342 patients (194 in the training cohort, 78 in the internal validation cohort, and 70 in the external validation cohort) with pathologically proven GC from two centers were included. Radiomics features were extracted from the preoperative CT imaging. The clinical model, radiomics signature, and radiomics nomogram, which incorporated the radiomics signature and independent clinical risk factors, were developed and verified. Furthermore, the performance of these three models was assessed by using the area under the curve (AUC) of receiver operating characteristic (ROC) curve analysis and decision curve analysis (DCA). RESULTS: The radiomics signature, which was comprised of two selected radiomics features, namely, contrast_GLCM and dissimilarity_GLCM, showed better performance than the clinical model in predicting the LR of GC, with AUC values of 0.83 in the training cohort, 0.84 in the internal validation cohort, and 0.73 in the external cohort, respectively. By integrating the independent clinical risk factors (N stage, bile acid duodenogastric reflux and nodular or irregular outer layer of the gastric wall) into the radiomics signature, the radiomics nomogram achieved the highest accuracy in predicting LR, with AUC values of 0.89, 0.89 and 0.80 in the three cohorts, respectively. DCA in the validation cohort showed that radiomics nomogram added more net benefit than the clinical model within the range of 0.01-0.98. CONCLUSION: The CT-based radiomics nomogram has the potential to predict the LR of GC after radical resection.

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 Heterogeneity Radiomic Nomogram for Preoperative Differentiation of Primary Gastric Lymphoma From Borrmann Type IV Gastric Cancer.

OBJECTIVE: This study aimed to preoperatively differentiate primary gastric lymphoma from Borrmann type IV gastric cancer by heterogeneity nomogram based on routine contrast-enhanced computed tomographic images. METHODS: We enrolled 189 patients from 2 hospitals (90 in the training cohort and 99 in the validation cohort). Subjective findings, including high-enhanced mucosal sign, high-enhanced serosa sign, nodular or an irregular outer layer of the gastric wall, and perigastric fat infiltration, were assessed to construct a subjective finding model. A deep learning model was developed to segment tumor areas, from which 1680 three-dimensional heterogeneity radiomic parameters, including first-order entropy, second-order entropy, and texture complexity, were extracted to build a heterogeneity signature by least absolute shrinkage and selection operator logistic regression. A nomogram that integrates heterogeneity signature and subjective findings was developed by multivariate logistic regression. The diagnostic performance of the nomogram was assessed by discrimination and clinical usefulness. RESULTS: High-enhanced serosa sign and nodular or an irregular outer layer of the gastric wall were identified as independent predictors for building the subjective finding model. High-enhanced serosa sign and heterogeneity signature were significant predictors for differentiating the 2 groups (all, P < 0.05). The area under the curve with heterogeneity nomogram was 0.932 (95% confidence interval, 0.863-0.973) in the validation cohort. Decision curve analysis and stratified analysis confirmed the clinical utility of the heterogeneity nomogram. CONCLUSIONS: The proposed heterogeneity radiomic nomogram on contrast-enhanced computed tomographic images may help differentiate primary gastric lymphoma from Borrmann type IV gastric cancer preoperatively.

MIP image derived from abbreviated breast MRI: potential to reduce unnecessary sub-nipple biopsies during nipple-sparing mastectomy for breast cancer.

OBJECTIVE: To determine the value of a maximum-intensity projection (MIP) image derived from abbreviated breast MRI for excluding occult nipple-areolar complex (NAC) involvement in patients with breast cancer. METHODS: This prospective study included breast cancer patients with clinically normal NACs between April 2016 and May 2019. Abbreviated breast MRI was performed, and an MIP image was generated for each patient. MIP images were examined for the following features: asymmetric nipple enhancement, tumor-nipple distance (TND), tumor diameter, lesion type, location, and multifocality. Independent predictive MIP features for occult NAC involvement were identified by univariable and multivariable logistic regression analyses. Models based on independent predictive MIP features were developed, and their diagnostic performances were evaluated using ROC analysis. The utility of an MIP image for excluding occult NAC involvement was assessed by considering NPVs across patient subgroups. RESULTS: Eight hundred forty-three patients (67 NAC-positive and 776 NAC-negative) were enrolled. On MIP images, asymmetric nipple enhancement (odds ratio, 6.098; p < 0.001) and TND (odds ratio, 0.564; p = 0.003) were independent predictors of occult NAC involvement. A parallel test model of "asymmetric nipple enhancement or TND ≤ 15 mm" yielded the highest AUC value (0.838) among prediction models. The NPV of MIP images for excluding occult NAC involvement was 99.5%, which was applicable across various patient subgroups. CONCLUSIONS: A single MIP image derived from abbreviated breast MRI has utility for excluding occult NAC involvement in breast cancer patients and reducing the number of unnecessary sub-nipple biopsies in nipple-sparing mastectomy. KEY POINTS: • On MIP images derived from abbreviated breast MRI, asymmetric nipple enhancement and tumor-nipple distance were independent predictors for occult nipple involvement in patients with breast cancer. • Negative findings on MIP image can help select patients at minimal risk of occult nipple involvement, for whom unnecessary intraoperative sub-nipple biopsies in nipple-sparing mastectomy can be omitted.

A CT-based radiomics nomogram for prediction of lung adenocarcinomas and granulomatous lesions in patient with solitary sub-centimeter solid nodules.

PURPOSE: To develop a radiomics nomogram based on computed tomography (CT) images that can help differentiate lung adenocarcinomas and granulomatous lesions appearing as sub-centimeter solid nodules (SCSNs). MATERIALS AND METHODS: The records of 214 consecutive patients with SCSNs that were surgically resected and histologically confirmed as lung adenocarcinomas (n = 112) and granulomatous lesions (n = 102) from 2 medical institutions between October 2011 and June 2019 were retrospectively analyzed. Patients from center 1 ware enrolled as training cohort (n = 150) and patients from center 2 were included as external validation cohort (n = 64), respectively. Radiomics features were extracted from non-contrast chest CT images preoperatively. The least absolute shrinkage and selection operator (LASSO) regression model was used for radiomics feature extraction and radiomics signature construction. Clinical characteristics, subjective CT findings, and radiomics signature were used to develop a predictive radiomics nomogram. The performance was examined by assessment of the area under the receiver operating characteristic curve (AUC). RESULTS: Lung adenocarcinoma was significantly associated with an irregular margin and lobulated shape in the training set (p = 0.001, < 0.001) and external validation set (p = 0.016, = 0.018), respectively. The radiomics signature consisting of 22 features was significantly associated with lung adenocarcinomas of SCSNs (p < 0.001). The radiomics nomogram incorporated the radiomics signature, gender and lobulated shape. The AUCs of combined model in the training and external validation dataset were 0.885 (95% confidence interval [CI]: 0.823-0.931), 0.808 (95% CI: 0.690-0.896), respectively. Decision curve analysis (DCA) demonstrated that the radiomics nomogram was clinically useful. CONCLUSION: A radiomics signature based on non-enhanced CT has the potential to differentiate between lung adenocarcinomas and granulomatous lesions. The radiomics nomogram incorporating the radiomics signature and subjective findings may facilitate the individualized, preoperative treatment in patients with SCSNs.

Segmentation of small ground glass opacity pulmonary nodules based on Markov random field energy and Bayesian probability difference.

BACKGROUND: Image segmentation is an important part of computer-aided diagnosis (CAD), the segmentation of small ground glass opacity (GGO) pulmonary nodules is beneficial for the early detection of lung cancer. For the segmentation of small GGO pulmonary nodules, an integrated active contour model based on Markov random field energy and Bayesian probability difference (IACM_MRFEBPD) is proposed in this paper. METHODS: First, the Markov random field (MRF) is constructed on the computed tomography (CT) images, then the MRF energy is calculated. The MRF energy is used to construct the region term. It can not only enhance the contrast between pulmonary nodule and the background region, but also solve the problem of intensity inhomogeneity using local spatial correlation information between neighboring pixels in the image. Second, the Gaussian mixture model is used to establish the probability model of the image, and the model parameters are estimated by the expectation maximization (EM) algorithm. So the Bayesian posterior probability difference of each pixel can be calculated. The probability difference is used to construct the boundary detection term, which is 0 at the boundary. Therefore, the blurred boundary problem can be solved. Finally, under the framework of the level set, the integrated active contour model is constructed. RESULTS: To verify the effectiveness of the proposed method, the public data of the lung image database consortium and image database resource initiative (LIDC-IDRI) and the clinical data of the Affiliated Jiangmen Hospital of Sun Yat-sen University are used to perform experiments, and the intersection over union (IOU) score is used to evaluate the segmentation methods. Compared with other methods, the proposed method achieves the best results with the highest average IOU of 0.7444, 0.7503, and 0.7450 for LIDC-IDRI test set, clinical test set, and all test sets, respectively. CONCLUSIONS: The experiment results show that the proposed method can segment various small GGO pulmonary nodules more accurately and robustly, which is helpful for the accurate evaluation of medical imaging.

Solitary solid pulmonary nodules: a CT-based deep learning nomogram helps differentiate tuberculosis granulomas from lung adenocarcinomas.

OBJECTIVES: To evaluate the differential diagnostic performance of a computed tomography (CT)-based deep learning nomogram (DLN) in identifying tuberculous granuloma (TBG) and lung adenocarcinoma (LAC) presenting as solitary solid pulmonary nodules (SSPNs). METHODS: Routine CT images of 550 patients with SSPNs were retrospectively obtained from two centers. A convolutional neural network was used to extract deep learning features from all lesions. The training set consisted of data for 218 patients. The least absolute shrinkage and selection operator logistic regression was used to create a deep learning signature (DLS). Clinical factors and CT-based subjective findings were combined in a clinical model. An individualized DLN incorporating DLS, clinical factors, and CT-based subjective findings was constructed to validate the diagnostic ability. The performance of the DLN was assessed by discrimination and calibration using internal (n = 140) and external validation cohorts (n = 192). RESULTS: DLS, gender, age, and lobulated shape were found to be independent predictors and were used to build the DLN. The combination showed better diagnostic accuracy than any single model evaluated using the net reclassification improvement method (p < 0.05). The areas under the curve in the training, internal validation, and external validation cohorts were 0.889 (95% confidence interval [CI], 0.839-0.927), 0.879 (95% CI, 0.813-0.928), and 0.809 (95% CI, 0.746-0.862), respectively. Decision curve analysis and stratification analysis showed that the DLN has potential generalization ability. CONCLUSIONS: The CT-based DLN can preoperatively distinguish between LAC and TBG in patients presenting with SSPNs. KEY POINTS: • The deep learning nomogram was developed to preoperatively differentiate TBG from LAC in patients with SSPNs. • The performance of the deep learning feature was superior to that of the radiomics feature. • The deep learning nomogram achieved superior performance compared to the deep learning signature, the radiomics signature, or the clinical model alone.

Radiomics nomogram for preoperative differentiation of lung tuberculoma from adenocarcinoma in solitary pulmonary solid nodule.

PURPOSE: To investigate the preoperative differential diagnostic performance of a radiomics nomogram in tuberculous granuloma (TBG) and lung adenocarcinoma (LAC) appearing as solitary pulmonary solid nodules (SPSNs). METHOD: We retrospectively recruited 426 patients with SPSNs from two centers and assigned them to training (n = 123), internal validation (n = 121), and external validation cohorts (n = 182). A model of deep learning (DL) was built for tumor segmentation from routine computed tomography (CT) images and extraction of 3D radiomics features. We used the least absolute shrinkage and selection operator (LASSO) logistic regression to build a radiomics signature. A clinical model was developed with clinical factors, including age, gender, and CT-based subjective findings (eg, lesion size, lesion location, lesion margin, lobulated sharp, and spiculation sign). We constructed individualized radiomics nomograms incorporating the radiomics signature and clinical factors to validate the diagnostic ability. RESULTS: Three factors - radiomics signature, age, and spiculation sign - were found to be independent predictors and were used to build the radiomics nomogram, which showed better diagnostic accuracy than any single model (all net reclassification improvement p < 0.05). The area under curve yielded was 0.9660 (95% confidence interval [CI], 0.9390-0.9931), 0.9342 (95% CI, 0.8944-0.9739), and 0.9064 (95% CI, 0.8639-0.9490) for the training, internal validation, and external validation cohorts, respectively. Decision curve analysis (DCA) and stratification analysis showed the nomogram has potential for generalizability. CONCLUSION: The radiomics nomogram we developed can preoperatively distinguish between LAC and TBG in patient with a SPSN.

A radiomics model to predict the invasiveness of thymic epithelial tumors based on contrast­enhanced computed tomography.

In the present study, we aimed to construct a radiomics model using contrast­enhanced computed tomography (CT) to predict the pathological invasiveness of thymic epithelial tumors (TETs). We retrospectively reviewed the records of 179 consecutive patients (89 females) with histologically confirmed TETs from two hospitals. The 82 low­ and 97 high­risk TETs were assigned to training (90 tumors), internal validation (49 tumors) and external validation (40 tumors) cohorts. Radiomics features extracted from preoperative contrast­enhanced chest CT were selected using least absolute shrinkage and selection operator logistic regression. Three prediction models were developed using multivariate logistic regression analysis. Their performance and clinical utility were assessed using receiver operating characteristic curves and the DeLong test, respectively. Eight radiomics features with non­zero coefficients were used to develop a radiomics score, which significantly differed between low­ and high­risk TETs (P<0.001). The subjective finding, infiltration, was independently associated with high­risk TETs. Prediction models based on infiltration alone, the radiomics signature alone, and both these parameters showed diagnostic accuracies of 72.2% [area under curve (AUC), 0.731; 95% confidence interval (CI): 0.627­0.819; sensitivity, 85.7%; specificity, 60.4%], 88.9% (AUC, 0.944; 95% CI: 0.874­0.981; sensitivity, 92.9%; specificity, 85.4%), and 90.0% (AUC, 0.953; 95% CI: 0.887­0.987; sensitivity, 92.9%; specificity, 87.5%), respectively. Decision­curve analysis showed that the combined model added more net benefit than the single­parameter models. In conclusion, a radiomics signature based on contrast­enhanced CT has the potential to differentiate between low­ and high­risk TETs. The model incorporating the radiomics signature and subjective finding may facilitate the individualized, preoperative prediction of the pathological invasiveness of TETs.

Whole-Lesion Computed Tomography-Based Entropy Parameters for the Differentiation of Minimally Invasive and Invasive Adenocarcinomas Appearing as Pulmonary Subsolid Nodules.

OBJECTIVE: The aim of this study was to investigate the differentiation of computed tomography (CT)-based entropy parameters between minimally invasive adenocarcinoma (MIA) and invasive adenocarcinoma (IAC) lesions appearing as pulmonary subsolid nodules (SSNs). METHODS: This study was approved by the institutional review board in our hospital. From July 2015 to November 2018, 186 consecutive patients with solitary peripheral pulmonary SSNs that were pathologically confirmed as pulmonary adenocarcinomas (74 MIA and 112 IAC lesions) were included and subdivided into the training data set and the validation data set. Chest CT scans without contrast enhancement were performed in all patients preoperatively. The subjective CT features of the SSNs were reviewed and compared between the MIA and IAC groups. Each SSN was semisegmented with our in-house software, and entropy-related parameters were quantitatively extracted using another in-house software developed in the MATLAB platform. Logistic regression analysis and receiver operating characteristic analysis were performed to evaluate the diagnostic performances. Three diagnostic models including subjective model, entropy model, and combined model were built and analyzed using area under the curve (AUC) analysis. RESULTS: There were 119 nonsolid nodules and 67 part-solid nodules. Significant differences were found in the subjective CT features among nodule type, lesion size, lobulated shape, and irregular margin between the MIA and IAC groups. Multivariate analysis revealed that part-solid type and lobulated shape were significant independent factors for IAC (P < 0.0001 and P < 0.0001, respectively). Three entropy parameters including Entropy-0.8, Entropy-2.0-32, and Entropy-2.0-64 were identified as independent risk factors for the differentiation of MIA and IAC lesions. The median entropy model value of the MIA group was 0.266 (range, 0.174-0.590), which was significantly lower than the IAC group with value 0.815 (range, 0.623-0.901) (P < 0.0001). Multivariate analysis revealed that the combined model had an excellent diagnostic performance with sensitivity of 88.2%, specificity of 73.0%, and accuracy of 82.1%. The AUC value of the combined model was significantly higher (AUC, 0.869) than that of the subjective model (AUC, 0.809) or the entropy model alone (AUC, 0.836) (P < 0.0001). CONCLUSIONS: The CT-based entropy parameters could help assess the aggressiveness of pulmonary adenocarcinoma via quantitative analysis of intratumoral heterogeneity. The MIA can be differentiated from IAC accurately by using entropy-related parameters in peripheral pulmonary SSNs.

Preoperative prediction of lymphovascular invasion in invasive breast cancer with dynamic contrast-enhanced-MRI-based radiomics.

BACKGROUND: Lymphovascular invasion (LVI) status facilitates the selection of optimal therapeutic strategy for breast cancer patients, but in clinical practice LVI status is determined in pathological specimens after resection. PURPOSE: To explore the use of dynamic contrast-enhanced (DCE)-magnetic resonance imaging (MRI)-based radiomics for preoperative prediction of LVI in invasive breast cancer. STUDY TYPE: Prospective. POPULATION: Ninety training cohort patients (22 LVI-positive and 68 LVI-negative) and 59 validation cohort patients (22 LVI-positive and 37 LVI-negative) were enrolled. FIELD STRENGTH/SEQUENCE: 1.5 T and 3.0 T, T1 -weighted DCE-MRI. ASSESSMENT: Axillary lymph node (ALN) status for each patient was evaluated based on MR images (defined as MRI ALN status), and DCE semiquantitative parameters of lesions were calculated. Radiomic features were extracted from the first postcontrast DCE-MRI. A radiomics signature was constructed in the training cohort with 10-fold cross-validation. The independent risk factors for LVI were identified and prediction models for LVI were developed. Their prediction performances and clinical usefulness were evaluated in the validation cohort. STATISTICAL TESTS: Mann-Whitney U-test, chi-square test, kappa statistics, least absolute shrinkage and selection operator (LASSO) regression, logistic regression, receiver operating characteristic (ROC) analysis, DeLong test, and decision curve analysis (DCA). RESULTS: Two radiomic features were selected to construct the radiomics signature. MRI ALN status (odds ratio, 10.452; P < 0.001) and the radiomics signature (odds ratio, 2.895; P = 0.031) were identified as independent risk factors for LVI. The value of the area under the curve (AUC) for a model combining both (0.763) was higher than that for MRI ALN status alone (0.665; P = 0.029) and similar to that for the radiomics signature (0.752; P = 0.857). DCA showed that the combined model added more net benefit than either feature alone. DATA CONCLUSION: The DCE-MRI-based radiomics signature in combination with MRI ALN status was effective in predicting the LVI status of patients with invasive breast cancer before surgery. LEVEL OF EVIDENCE: 1 Technical Efficacy Stage: 2 J. Magn. Reson. Imaging 2019;50:847-857.