Can computed tomography-based radiomics potentially discriminate between anterior mediastinal cysts and type B1 and B2 thymomas?

Development and Validation of Contrast-Enhanced CT-Based Deep Transfer Learning and Combined Clinical-Radiomics Model to Discriminate Thymomas and Thymic Cysts: A Multicenter Study

RATIONALE AND OBJECTIVES

This study aims to evaluate the feasibility and effectiveness of deep transfer learning (DTL) and clinical-radiomics in differentiating thymoma from thymic cysts.

MATERIALS AND METHODS

Clinical and imaging data of 196 patients pathologically diagnosed with thymoma and thymic cysts were retrospectively collected from center 1. (training cohort: n = 137; internal validation cohort: n = 59). An independent external validation cohort comprised 68 thymoma and thymic cyst patients from center 2. Region of interest (ROI) delineation was performed on contrast-enhanced chest computed tomography (CT) images, and eight DTL models including Densenet 169, Mobilenet V2, Resnet 101, Resnet 18, Resnet 34, Resnet 50, Vgg 13, Vgg 16 were constructed. Radiomics features were extracted from the ROI on the CT images of thymoma and thymic cyst patients, and feature selection was performed using intra-observer correlation coefficient (ICC), Spearman correlation analysis, and least absolute shrinkage and selection operator (LASSO) algorithm. Univariate analysis and multivariable logistic regression (LR) were used to select clinical-radiological features. Six machine learning classifiers, including LR, support vector machine (SVM), k-nearest neighbors (KNN), Light Gradient Boosting Machine (LightGBM), Adaptive Boosting (AdaBoost), and Multilayer Perceptron (MLP), were used to construct Radiomics and Clinico-radiologic models. The selected features from the Radiomics and Clinico-radiologic models were fused to build a Combined model. Receiver operating characteristic curve (ROC), calibration curve, and decision curve analysis (DCA) were used to evaluate the discrimination, calibration, and clinical utility of the models, respectively. The Delong test was used to compare the AUC between different models. K-means clustering was used to subdivide the lesions of thymomas or thymic cysts into subregions, and traditional radiomics methods were used to extract features and compare the ability of Radiomics and DTL models to reflect intratumoral heterogeneity using correlation analysis.

RESULTS

The Densenet 169 based on DTL performed the best, with AUC of 0.933 (95% CI: 0.875-0.991) in the internal validation cohort and 0.962 (95% CI: 0.923-1.000) in the external validation cohort. The AdaBoost classifier achieved AUC of 0.965 (95% CI: 0.923-1.000) and 0.959 (95% CI: 0.919-1.000) in the internal and external validation cohorts, respectively, for the Radiomics model. The LightGBM classifier achieved AUC of 0.805 (95% CI: 0.690-0.920) and 0.839 (95% CI: 0.736-0.943) in the Clinico-radiologic model. The AUC of the Combined model in the internal and external validation cohorts was 0.933 (95% CI: 0.866-1.000) and 0.945 (95% CI: 0.897-0.994), respectively. The results of the Delong test showed that the Radiomics model, DTL model, and Combined model outperformed the Clinico-radiologic model in both internal and external validation cohorts (p-values were 0.002, 0.004, and 0.033 in the internal validation cohort, while in the external validation cohort, the p-values were 0.014, 0.006, and 0.015, respectively). But there was no statistical difference in performance among the three models (all p-values <0.05). Correlation analysis showed that radiomics performed better than DTL in quantifying intratumoral heterogeneity differences between thymoma and thymic cysts.

CONCLUSION

The developed DTL model and the Combined model based on radiomics and clinical-radiologic features achieved excellent diagnostic performance in differentiating thymic cysts from thymoma. They can serve as potential tools to assist clinical decision-making, particularly when endoscopic biopsy carries a high risk.

Therapeutic Decision Making in Prevascular Mediastinal Tumors Using CT Radiomics and Clinical Features: Upfront Surgery or Pretreatment Needle Biopsy?

The study aimed to develop machine learning (ML) classification models for differentiating patients who needed direct surgery from patients who needed core needle biopsy among patients with prevascular mediastinal tumor (PMT). Patients with PMT who received a contrast-enhanced computed tomography (CECT) scan and initial management for PMT between January 2010 and December 2020 were included in this retrospective study. Fourteen ML algorithms were used to construct candidate classification models via the voting ensemble approach, based on preoperative clinical data and radiomic features extracted from the CECT. The classification accuracy of clinical diagnosis was 86.1%. The first ensemble learning model was built by randomly choosing seven ML models from a set of fourteen ML models and had a classification accuracy of 88.0% (95% CI = 85.8 to 90.3%). The second ensemble learning model was the combination of five ML models, including NeuralNetFastAI, NeuralNetTorch, RandomForest with Entropy, RandomForest with Gini, and XGBoost, and had a classification accuracy of 90.4% (95% CI = 87.9 to 93.0%), which significantly outperformed clinical diagnosis (p < 0.05). Due to the superior performance, the voting ensemble learning clinical-radiomic classification model may be used as a clinical decision support system to facilitate the selection of the initial management of PMT.

Multiparametric Evaluation of Radiomics Features and Dual-Energy CT Iodine Maps for Discrimination and Outcome Prediction of Thymic Masses

RATIONALE AND OBJECTIVES

To investigate the diagnostic value of radiomics features and dual-source dual-energy CT (DECT) based material decomposition in differentiating low-risk thymomas, high-risk thymomas, and thymic carcinomas.

MATERIALS AND METHODS

This retrospective study included 32 patients (16 males, mean age 66 ± 14 years) with pathologically confirmed thymic masses who underwent contrast-enhanced DECT between 10/2014 and 01/2023. Two experienced readers evaluated all patients regarding conventional radiomics features, as well as DECT-based features, including attenuation (HU), iodine density (mg/mL), and fat fraction (%). Data comparisons were performed using analysis of variance and chi-square statistic tests. Receiver operating characteristic curve analysis and Cox-regression tests were used to discriminate between low-risk/high-risk thymomas and thymic carcinomas.

RESULTS

Of the 32 thymic tumors, 12 (38%) were low-risk thymomas, 11 (34%) were high-risk thymomas, and 9 (28%) were thymic carcinomas. Values differed significantly between low-risk thymoma, high-risk thymoma, and thymic carcinoma regarding DECT-based features (p ≤ 0.023) and 30 radiomics features (p ≤ 0.037). The area under the curve to differentiate between low-risk/high-risk thymomas and thymic cancer was 0.998 (95% CI, 0.915-1.000; p < 0.001) for the combination of DECT imaging parameters and radiomics features, yielding a sensitivity of 100% and specificity of 96%. During a follow-up of 60 months (IQR, 35-60 months), the multiparametric approach including radiomics features, DECT parameters, and clinical parameters showed an excellent prognostic power to predict all-cause mortality (c-index = 0.978 [95% CI, 0.958-0.998], p = 0.003).

CONCLUSION

A multiparametric approach including conventional radiomics features and DECT-based features facilitates accurate, non-invasive discrimination between low-risk/high-risk thymomas and thymic carcinomas.

Deep learning-based radiomic nomogram to predict risk categorization of thymic epithelial tumors: A multicenter study

PURPOSE

The study was aimed to develop and evaluate a deep learning-based radiomics to predict the histological risk categorization of thymic epithelial tumors (TETs), which can be highly informative for patient treatment planning and prognostic assessment.

METHOD

A total of 681 patients with TETs from three independent hospitals were included and separated into derivation cohort and external test cohort. Handcrafted and deep learning features were extracted from preoperative contrast-enhanced CT images and selected to build three radiomics signatures (radiomics signature [Rad_Sig], deep learning signature [DL_Sig] and deep learning radiomics signature [DLR_Sig]) to predict risk categorization of TETs. A deep learning-based radiomic nomogram (DLRN) was then depicted to visualize the classification evaluation. The performance of predictive models was compared using the receiver operating characteristic and decision curve analysis (DCA).

RESULTS

Among three radiomics signatures, DLR_Sig demonstrated optimum performance with an AUC of 0.883 for the derivation cohort and 0.749 for the external test cohort. Combining DLR_Sig with age and gender, DLRN was depict and exhibited optimum performance among all radiomics models with an AUC of 0.965, accuracy of 0.911, sensitivity of 0.921 and specificity of 0.902 in the derivation cohort, and an AUC of 0.786, accuracy of 0.774, sensitivity of 0.778 and specificity of 0.771 in the external test cohort. The DCA showed that DLRN had greater clinical benefit than other radiomics signatures.

CONCLUSIONS

Our study developed and validated a DLRN to accurately predict the risk categorization of TETs, which has potential to facilitate individualized treatment and improve patient prognosis evaluation.

Pan-mediastinal neoplasm diagnosis via nationwide federated learning: a multicentre cohort study

BACKGROUND

Mediastinal neoplasms are typical thoracic diseases with increasing incidence in the general global population and can lead to poor prognosis. In clinical practice, the mediastinum's complex anatomic structures and intertype confusion among different mediastinal neoplasm pathologies severely hinder accurate diagnosis. To solve these difficulties, we organised a multicentre national collaboration on the basis of privacy-secured federated learning and developed CAIMEN, an efficient chest CT-based artificial intelligence (AI) mediastinal neoplasm diagnosis system.

METHODS

In this multicentre cohort study, 7825 mediastinal neoplasm cases and 796 normal controls were collected from 24 centres in China to develop CAIMEN. We further enhanced CAIMEN with several novel algorithms in a multiview, knowledge-transferred, multilevel decision-making pattern. CAIMEN was tested by internal (929 cases at 15 centres), external (1216 cases at five centres and a real-world cohort of 11 162 cases), and human-AI (60 positive cases from four centres and radiologists from 15 institutions) test sets to evaluate its detection, segmentation, and classification performance.

FINDINGS

In the external test experiments, the area under the receiver operating characteristic curve for detecting mediastinal neoplasms of CAIMEN was 0·973 (95% CI 0·969-0·977). In the real-world cohort, CAIMEN detected 13 false-negative cases confirmed by radiologists. The dice score for segmenting mediastinal neoplasms of CAIMEN was 0·765 (0·738-0·792). The mediastinal neoplasm classification top-1 and top-3 accuracy of CAIMEN were 0·523 (0·497-0·554) and 0·799 (0·778-0·822), respectively. In the human-AI test experiments, CAIMEN outperformed clinicians with top-1 and top-3 accuracy of 0·500 (0·383-0·633) and 0·800 (0·700-0·900), respectively. Meanwhile, with assistance from the computer aided diagnosis software based on CAIMEN, the 46 clinicians improved their average top-1 accuracy by 19·1% (0·345-0·411) and top-3 accuracy by 13·0% (0·545-0·616).

INTERPRETATION

For mediastinal neoplasms, CAIMEN can produce high diagnostic accuracy and assist the diagnosis of human experts, showing its potential for clinical practice.

FUNDING

National Key R&D Program of China, National Natural Science Foundation of China, and Beijing Natural Science Foundation.

Conventional and radiomic features to predict pathology in the preoperative assessment of anterior mediastinal masses

OBJECTIVES

The aim of this study was to differentiate benign from malignant tumors in the anterior mediastinum based on computed tomography (CT) imaging characteristics, which could be useful in preoperative planning. Additionally, our secondary aim was to differentiate thymoma from thymic carcinoma, which could guide the use of neoadjuvant therapy.

MATERIALS AND METHODS

Patients referred for thymectomy were retrospectively selected from our database. Twenty-five conventional characteristics were evaluated by visual analysis, and 101 radiomic features were extracted from each CT. In the step of model training, we applied support vector machines to train classification models. Model performance was assessed using the area under the receiver operating curves (AUC).

RESULTS

Our final study sample comprised 239 patients, 59 (24.7 %) with benign mediastinal lesions and 180 (75.3 %) with malignant thymic tumors. Among the malignant masses, there were 140 (58.6 %) thymomas, 23 (9.6 %) thymic carcinomas, and 17 (7.1 %) non-thymic lesions. For the benign versus malignant differentiation, the model that integrated both conventional and radiomic features achieved the highest diagnostic performance (AUC = 0.715), in comparison to the conventional (AUC = 0.605) and radiomic-only (AUC = 0.678) models. Similarly, regarding thymoma versus thymic carcinoma differentiation, the model that integrated both conventional and radiomic features also achieved the highest diagnostic performance (AUC = 0.810), in comparison to the conventional (AUC = 0.558) and radiomic-only (AUC = 0.774) models.

CONCLUSION

CT-based conventional and radiomic features with machine learning analysis could be useful for predicting pathologic diagnoses of anterior mediastinal masses. The diagnostic performance was moderate for differentiating benign from malignant lesions and good for differentiating thymomas from thymic carcinomas. The best diagnostic performance was achieved when both conventional and radiomic features were integrated in the machine learning algorithms.

Machine-learning classifiers based on non-enhanced computed tomography radiomics to differentiate anterior mediastinal cysts from thymomas and low-risk from high-risk thymomas: A multi-center study

Background

This study aimed to investigate the diagnostic value of machine-learning (ML) models with multiple classifiers based on non-enhanced CT Radiomics features for differentiating anterior mediastinal cysts (AMCs) from thymomas, and high-risk from low risk thymomas.

Methods

In total, 201 patients with AMCs and thymomas from three centers were included and divided into two groups: AMCs vs. thymomas, and high-risk vs low-risk thymomas. A radiomics model (RM) was built with 73 radiomics features that were extracted from the three-dimensional images of each patient. A combined model (CM) was built with clinical features and subjective CT finding features combined with radiomics features. For the RM and CM in each group, five selection methods were adopted to select suitable features for the classifier, and seven ML classifiers were employed to build discriminative models. Receiver operating characteristic (ROC) curves were used to evaluate the diagnostic performance of each combination.

Results

Several classifiers combined with suitable selection methods demonstrated good diagnostic performance with areas under the curves (AUCs) of 0.876 and 0.922 for the RM and CM in group 1 and 0.747 and 0.783 for the RM and CM in group 2, respectively. The combination of support vector machine (SVM) as the feature-selection method and Gradient Boosting Decision Tree (GBDT) as the classification algorithm represented the best comprehensive discriminative ability in both group. Comparatively, assessments by radiologists achieved a middle AUCs of 0.656 and 0.626 in the two groups, which were lower than the AUCs of the RM and CM. Most CMs exhibited higher AUC value compared to RMs in both groups, among them only a few CMs demonstrated better performance with significant difference in group 1.

Conclusion

Our ML models demonstrated good performance for differentiation of AMCs from thymomas and low-risk from high-risk thymomas. ML based on non-enhanced CT radiomics may serve as a novel preoperative tool.

The Studies of Prognostic Factors and the Genetic Polymorphism of Methylenetetrahydrofolate Reductase C667T in Thymic Epithelial Tumors

Objective

To describe the clinical features of a cohort of patients with thymic epithelial tumors (TETs) and to analyze their prognostic factors. In particular, we investigated the correlation between the genetic polymorphism of methylenetetrahydrofolate reductase (MTHFR) C667T and the incidence of TETs.

Methods

Pathological records were reviewed from the database of the Second Affiliated Hospital of Jiaxing University, from January 2010 to December 2020, and 84 patients with TETs were recruited for this study. Univariate and multivariate analyses were performed to determine the prognostic factors. The genetic polymorphism of MTHFR C667T was examined in the patients with TETs and in a group of healthy individuals. The correlation between MTHFR transcriptional levels and methylation was analyzed using The Cancer Genome Atlas (TCGA) thymoma dataset from the cBioPortal platform.

Results

Kaplan–Meier univariate survival analysis showed that sex, age, the maximum tumor diameter, surgery, chemotherapy, radiotherapy, WHO histological classification, Masaoka–Koga stage, and 8th UICC/AJCC TNM staging, were statistically significantly correlated with the prognosis of patients with TETs. The Masaoka–Koga stage and 8th UICC/AJCC TNM staging were strongly correlated with each other in this study (r=0.925, P<0.001). Cox multivariate survival analysis showed that the maximum tumor diameter, Masaoka–Koga stage, and 8th UICC/AJCC TNM staging were independent prognostic factors affecting the overall survival (OS) of patients with TETs (P<0.05). The MTHFR C667T genotype (χ^{2 =} 7.987, P=0.018) and allele distribution (χ^{2 =} 5.750, P=0.016) were significantly different between the patients and healthy controls. CT heterozygous and TT homozygous genotypes at this MTHFR polymorphism significantly increased the risk of TETs (odds ratio [OR] =4.721, P=0.008). Kaplan–Meier univariate survival analysis showed that there was no correlation between different genotypes and the prognosis of TETs (CC versus CT + TT, χ^{2 =}0.003, P=0.959). Finally, a negative correlation between the transcriptional and methylation levels of MTHFR was observed in the TCGA thymoma dataset (r=-0.24, P=0.010).

Conclusions

The Masaoka–Koga stage, 8th UICC/AJCC TNM staging, and maximum tumor diameter were independent prognostic factors for TETs. Reduced methylation levels of MTHFR and particular polymorphic variants may contribute to the susceptibility to developing TETs.