CT Radiomics Model for Discriminating the Risk Stratification of Gastrointestinal Stromal Tumors: A Multi-Class Classification and Multi-Center Study

CT-based radiomics research for discriminating the risk stratification of pheochromocytoma using different machine learning models: a multi-center study

OBJECTIVES

The purpose of this study was to explore and verify the value of various machine learning models in preoperative risk stratification of pheochromocytoma.

METHODS

A total of 155 patients diagnosed with pheochromocytoma through surgical pathology were included in this research (training cohort: n = 105; test cohort: n = 50); the risk stratification scoring system classified a PASS score of < 4 as low risk and a PASS score of ≥ 4 as high risk. From CT images captured during the non-enhanced, arterial, and portal venous phase, radiomic features were extracted. After reducing dimensions and selecting features, Logistic Regression (LR), Extra Trees, and K-Nearest Neighbor (KNN) were utilized to construct the radiomics models. By adopting ROC curve analysis, the optimal radiomics model was selected. Univariate and multivariate logistic regression analyses of clinical radiological features were used to determine the variables and establish a clinical model. The integration of radiomics and clinical features resulted in the creation of a combined model. ROC curve analysis was used to evaluate the performance of the model, while decision curve analysis (DCA) was employed to assess its clinical value.

RESULTS

3591 radiomics features were extracted from the region of interest in unenhanced and dual-phase (arterial and portal venous phase) CT images. 13 radiomics features were deemed to be valuable. The LR model demonstrated the highest prediction efficiency and robustness among the tested radiomics models, with an AUC of 0.877 in the training cohort and 0.857 in the test cohort. Ultimately, the composite of clinical features was utilized to formulate the clinical model. The combined model demonstrated the best discriminative ability (AUC, training cohort: 0.887; test cohort: 0.874). The DCA of the combined model showed the best clinical efficacy.

CONCLUSION

The combined model integrating radiomics and clinical features had an outstanding performance in differentiating the risk of pheochromocytoma and could offer a non-intrusive and effective approach for making clinical decisions.

Computed tomography radiogenomics: A potential tool for prediction of molecular subtypes in gastric stromal tumor

BACKGROUND

Preoperative knowledge of mutational status of gastrointestinal stromal tumors (GISTs) is essential to guide the individualized precision therapy.

AIM

To develop a combined model that integrates clinical and contrast-enhanced computed tomography (CE-CT) features to predict gastric GISTs with specific genetic mutations, namely KIT exon 11 mutations or KIT exon 11 codons 557-558 deletions.

METHODS

A total of 231 GIST patients with definitive genetic phenotypes were divided into a training dataset and a validation dataset in a 7:3 ratio. The models were constructed using selected clinical features, conventional CT features, and radiomics features extracted from abdominal CE-CT images. Three models were developed: ModelCT sign, modelCT sign + rad, and model CTsign + rad + clinic. The diagnostic performance of these models was evaluated using receiver operating characteristic (ROC) curve analysis and the Delong test.

RESULTS

The ROC analyses revealed that in the training cohort, the area under the curve (AUC) values for modelCT sign, modelCT sign + rad, and modelCT sign + rad + clinic for predicting KIT exon 11 mutation were 0.743, 0.818, and 0.915, respectively. In the validation cohort, the AUC values for the same models were 0.670, 0.781, and 0.811, respectively. For predicting KIT exon 11 codons 557-558 deletions, the AUC values in the training cohort were 0.667, 0.842, and 0.720 for modelCT sign, modelCT sign + rad, and modelCT sign + rad + clinic, respectively. In the validation cohort, the AUC values for the same models were 0.610, 0.782, and 0.795, respectively. Based on the decision curve analysis, it was determined that the modelCT sign + rad + clinic had clinical significance and utility.

CONCLUSION

Our findings demonstrate that the combined modelCT sign + rad + clinic effectively distinguishes GISTs with KIT exon 11 mutation and KIT exon 11 codons 557-558 deletions. This combined model has the potential to be valuable in assessing the genotype of GISTs.

Prediction of high Ki-67 proliferation index of gastrointestinal stromal tumors based on CT at non-contrast-enhanced and different contrast-enhanced phases

OBJECTIVES

To evaluate and analyze radiomics models based on non-contrast-enhanced computed tomography (CT) and different phases of contrast-enhanced CT in predicting Ki-67 proliferation index (PI) among patients with pathologically confirmed gastrointestinal stromal tumors (GISTs).

METHODS

A total of 383 patients with pathologically proven GIST were divided into a training set (n = 218, vendor 1) and 2 validation sets (n = 96, vendor 2; n = 69, vendors 3-5). Radiomics features extracted from the most recent non-contrast-enhanced and three contrast-enhanced CT scan prior to pathological examination. Random forest models were trained for each phase to predict tumors with high Ki-67 proliferation index (Ki-67>10%) and were evaluated using the area under the receiver operating characteristic curve (AUC) and other metrics on the validation sets.

RESULTS

Out of 107 radiomics features extracted from each phase of CT images, four were selected for analysis. The model trained using the non-contrast-enhanced phase achieved an AUC of 0.792 in the training set and 0.822 and 0.711 in the two validation sets, similar to models trained on different contrast-enhanced phases (p > 0.05). Several relevant features, including NGTDM Busyness and tumor size, remained predictive in non-contrast-enhanced and different contrast-enhanced images.

CONCLUSION

The results of this study indicate that a radiomics model based on non-contrast-enhanced CT matches that of models based on different phases of contrast-enhanced CT in predicting the Ki-67 PI of GIST. GIST may exhibit similar radiological patterns irrespective of the use of contrast agent, and such radiomics features may help quantify these patterns to predict Ki-67 PI of GISTs.

CLINICAL RELEVANCE STATEMENT

GIST may exhibit similar radiomics patterns irrespective of contrast agent; thus, radiomics models based on non-contrast-enhanced CT could be an alternative for risk stratification in GIST patients with contraindication to contrast agent.

KEY POINTS

• Performance of radiomics models in predicting Ki-67 proliferation based on different CT phases is evaluated. • Non-contrast-enhanced CT-based radiomics models performed similarly to contrast-enhanced CT in risk stratification in GIST patients. • NGTDM Busyness remains stable to contrast agents in GISTs in radiomics models.

Exploring deep learning radiomics for classifying osteoporotic vertebral fractures in X-ray images

Purpose

To develop and validate a deep learning radiomics (DLR) model that uses X-ray images to predict the classification of osteoporotic vertebral fractures (OVFs).

Material and methods

The study encompassed a cohort of 942 patients, involving examinations of 1076 vertebrae through X-ray, CT, and MRI across three distinct hospitals. The OVFs were categorized as class 0, 1, or 2 based on the Assessment System of Thoracolumbar Osteoporotic Fracture. The dataset was divided randomly into four distinct subsets: a training set comprising 712 samples, an internal validation set with 178 samples, an external validation set containing 111 samples, and a prospective validation set consisting of 75 samples. The ResNet-50 architectural model was used to implement deep transfer learning (DTL), undergoing -pre-training separately on the RadImageNet and ImageNet datasets. Features from DTL and radiomics were extracted and integrated using X-ray images. The optimal fusion feature model was identified through least absolute shrinkage and selection operator logistic regression. Evaluation of the predictive capabilities for OVFs classification involved eight machine learning models, assessed through receiver operating characteristic curves employing the "One-vs-Rest" strategy. The Delong test was applied to compare the predictive performance of the superior RadImageNet model against the ImageNet model.

Results

Following pre-training separately on RadImageNet and ImageNet datasets, feature selection and fusion yielded 17 and 12 fusion features, respectively. Logistic regression emerged as the optimal machine learning algorithm for both DLR models. Across the training set, internal validation set, external validation set, and prospective validation set, the macro-average Area Under the Curve (AUC) based on the RadImageNet dataset surpassed those based on the ImageNet dataset, with statistically significant differences observed (P<0.05). Utilizing the binary "One-vs-Rest" strategy, the model based on the RadImageNet dataset demonstrated superior efficacy in predicting Class 0, achieving an AUC of 0.969 and accuracy of 0.863. Predicting Class 1 yielded an AUC of 0.945 and accuracy of 0.875, while for Class 2, the AUC and accuracy were 0.809 and 0.692, respectively.

Conclusion

The DLR model, based on the RadImageNet dataset, outperformed the ImageNet model in predicting the classification of OVFs, with generalizability confirmed in the prospective validation set.

Multi-classification model incorporating radiomics and clinic-radiological features for predicting invasiveness and differentiation of pulmonary adenocarcinoma nodules

PURPOSE

To develop a comprehensive multi-classification model that combines radiomics and clinic-radiological features to accurately predict the invasiveness and differentiation of pulmonary adenocarcinoma nodules.

METHODS

A retrospective analysis was conducted on a cohort comprising 500 patients diagnosed with lung adenocarcinoma between January 2020 and December 2022. The dataset included preoperative CT images and histological reports of adenocarcinoma in situ (AIS, n = 97), minimally invasive adenocarcinoma (MIA, n = 139), and invasive adenocarcinoma (IAC, n = 264) with well-differentiated (WIAC, n = 99), moderately differentiated (MIAC, n = 84), and poorly differentiated IAC (PIAC, n = 81). The patients were classified into two groups (IAC and non-IAC) for binary classification and further divided into three and five groups for multi-classification. Feature selection was performed using the least absolute shrinkage and selection operator (LASSO) algorithm to identify the most informative radiomics and clinic-radiological features. Eight machine learning (ML) models were developed using these features, and their performance was evaluated using accuracy (ACC) and the area under the receiver-operating characteristic curve (AUC).

RESULTS

The combined model, utilizing the support vector machine (SVM) algorithm, demonstrated improved performance in the testing cohort, achieving an AUC of 0.942 and an ACC of 0.894 for the two-classification task. For the three- and five-classification tasks, the combined model employing the one versus one strategy of SVM (SVM-OVO) outperformed other models, with ACC values of 0.767 and 0.607, respectively. The AUC values for histological subtypes ranged from 0.787 to 0.929 in the testing cohort, while the Macro-AUC and Micro-AUC of the multi-classification models ranged from 0.858 to 0.896.

CONCLUSIONS

A multi-classification radiomics model combined with clinic-radiological features, using the SVM-OVO algorithm, holds promise for accurately predicting the histological characteristics of pulmonary adenocarcinoma nodules, which contributes to personalized treatment strategies for patients with lung adenocarcinoma.

Risk stratification for 1- to 2-cm gastric gastrointestinal stromal tumors: visual assessment of CT and EUS high-risk features versus CT radiomics analysis

OBJECTIVES

To investigate the ability of CT and endoscopic sonography (EUS) in predicting the malignant risk of 1-2-cm gastric gastrointestinal stromal tumors (gGISTs) and to clarify whether radiomics could be applied for risk stratification.

METHODS

A total of 151 pathologically confirmed 1-2-cm gGISTs from seven institutions were identified by contrast-enhanced CT scans between January 2010 and March 2021. A detailed description of EUS morphological features was available for 73 gGISTs. The association between EUS or CT high-risk features and pathological malignant potential was evaluated. gGISTs were randomly divided into three groups to build the radiomics model, including 74 in the training cohort, 37 in validation cohort, and 40 in testing cohort. The ROIs covering the whole tumor volume were delineated on the CT images of the portal venous phase. The Pearson test and least absolute shrinkage and selection operator (LASSO) algorithm were used for feature selection, and the ROC curves were used to evaluate the model performance.

RESULTS

The presence of EUS- and CT-based morphological high-risk features, including calcification, necrosis, intratumoral heterogeneity, irregular border, or surface ulceration, did not differ between very-low and intermediate risk 1-2-cm gGISTs (p > 0.05). The radiomics model consisting of five radiomics features showed favorable performance in discrimination of malignant 1-2-cm gGISTs, with the AUC of the training, validation, and testing cohort as 0.866, 0.812, and 0.766, respectively.

CONCLUSIONS

Instead of CT- and EUS-based morphological high-risk features, the CT radiomics model could potentially be applied for preoperative risk stratification of 1-2-cm gGISTs.

KEY POINTS

• The presence of EUS- and CT-based morphological high-risk factors, including calcification, necrosis, intratumoral heterogeneity, irregular border, or surface ulceration, did not correlate with the pathological malignant potential of 1-2-cm gGISTs. • The CT radiomics model could potentially be applied for preoperative risk stratification of 1-2-cm gGISTs.

Malignancy risk of gastrointestinal stromal tumors evaluated with noninvasive radiomics: A multi-center study

Purpose

This study was to investigate the diagnostic efficacy of radiomics models based on the enhanced CT images in differentiating the malignant risk of gastrointestinal stromal tumors (GIST) in comparison with the clinical indicators model and traditional CT diagnostic criteria.

Materials and methods

A total of 342 patients with GISTs confirmed histopathologically were enrolled from five medical centers. Data of patients wrom two centers comprised the training group (n=196), and data from the remaining three centers constituted the validation group (n=146). After CT image segmentation and feature extraction and selection, the arterial phase model and venous phase model were established. The maximum diameter of the tumor and internal necrosis were used to establish a clinical indicators model. The traditional CT diagnostic criteria were established for the classification of malignant potential of tumor. The performance of the four models was assessed using the receiver operating characteristics curve.

Reuslts

In the training group, the area under the curves(AUCs) of the arterial phase model, venous phase model, clinical indicators model, and traditional CT diagnostic criteria were 0.930 [95% confidence interval (CI): 0.895-0.965), 0.933 (95%CI 0.898-0.967), 0.917 (95%CI 0.872-0.961) and 0.782 (95%CI 0.717-0.848), respectively. In the validation group, the AUCs of the models were 0.960 (95%CI 0.930-0.990), 0.961 (95% CI 0.930-0.992), 0.922 (95%CI 0.884-0.960) and 0.768 (95%CI 0.692-0.844), respectively. No significant difference was detected in the AUC between the arterial phase model, venous phase model, and clinical indicators model by the DeLong test, whereas a significant difference was observed between the traditional CT diagnostic criteria and the other three models.

Conclusion

The radiomics model using the morphological features of GISTs play a significant role in tumor risk stratification and can provide a reference for clinical diagnosis and treatment plan.

Building Radiomics Models Based on Triple-Phase CT Images Combining Clinical Features for Discriminating the Risk Rating in Gastrointestinal Stromal Tumors

We aimed to build radiomics models based on triple-phase CT images combining clinical features to predict the risk rating of gastrointestinal stromal tumors (GISTs). A total of 231 patients with pathologically diagnosed GISTs from July 2012 to July 2020 were categorized into a training data set (82 patients with high risk, 80 patients with low risk) and a validation data set (35 patients with high risk, 34 patients with low risk) with a ratio of 7:3. Four diagnostic models were constructed by assessing 20 clinical characteristics and 18 radiomic features that were extracted from a lesion mask based on triple-phase CT images. The receiver operating characteristic (ROC) curves were applied to calculate the diagnostic performance of these models, and ROC curves of these models were compared using Delong test in different data sets. The results of ROC analyses showed that areas under ROC curves (AUC) of model 4 [Clinic + CT value of unenhanced (CTU) + CT value of arterial phase (CTA) + value of venous phase (CTV)], model 1 (Clinic + CTU), model 2 (Clinic + CTA), and model 3 (Clinic + CTV) were 0.925, 0.894, 0.909, and 0.914 in the training set and 0.897, 0.866, 0,892, and 0.892 in the validation set, respectively. Model 4, model 1, model 2, and model 3 yielded an accuracy of 88.3%, 85.8%, 86.4%, and 84.6%, a sensitivity of 85.4%, 84.2%, 76.8%, and 78.0%, and a specificity of 91.2%, 87.5%, 96.2%, and 91.2% in the training set and an accuracy of 88.4%, 84.1%, 82.6%, and 82.6%, a sensitivity of 88.6%, 77.1%, 74.3%, and 85.7%, and a specificity of 88.2%, 91.2%, 91.2%, and 79.4% in the validation set, respectively. There was a significant difference between model 4 and model 1 in discriminating the risk rating in gastrointestinal stromal tumors in the training data set (Delong test, p < 0.05). The radiomic models based on clinical features and triple-phase CT images manifested excellent accuracy for the discrimination of risk rating of GISTs.

Preoperative CT-Based Deep Learning Model for Predicting Risk Stratification in Patients With Gastrointestinal Stromal Tumors

Objective

To develop and evaluate a deep learning model (DLM) for predicting the risk stratification of gastrointestinal stromal tumors (GISTs).

Methods

Preoperative contrast-enhanced CT images of 733 patients with GISTs were retrospectively obtained from two centers between January 2011 and June 2020. The datasets were split into training (n = 241), testing (n = 104), and external validation cohorts (n = 388). A DLM for predicting the risk stratification of GISTs was developed using a convolutional neural network and evaluated in the testing and external validation cohorts. The performance of the DLM was compared with that of radiomics model by using the area under the receiver operating characteristic curves (AUROCs) and the Obuchowski index. The attention area of the DLM was visualized as a heatmap by gradient-weighted class activation mapping.

Results

In the testing cohort, the DLM had AUROCs of 0.90 (95% confidence interval [CI]: 0.84, 0.96), 0.80 (95% CI: 0.72, 0.88), and 0.89 (95% CI: 0.83, 0.95) for low-malignant, intermediate-malignant, and high-malignant GISTs, respectively. In the external validation cohort, the AUROCs of the DLM were 0.87 (95% CI: 0.83, 0.91), 0.64 (95% CI: 0.60, 0.68), and 0.85 (95% CI: 0.81, 0.89) for low-malignant, intermediate-malignant, and high-malignant GISTs, respectively. The DLM (Obuchowski index: training, 0.84; external validation, 0.79) outperformed the radiomics model (Obuchowski index: training, 0.77; external validation, 0.77) for predicting risk stratification of GISTs. The relevant subregions were successfully highlighted with attention heatmap on the CT images for further clinical review.

Conclusion

The DLM showed good performance for predicting the risk stratification of GISTs using CT images and achieved better performance than that of radiomics model.