High-order radiomics features based on T2 FLAIR MRI predict multiple glioma immunohistochemical features: A more precise and personalized gliomas management

Prediction of prognosis in glioblastoma with radiomics features extracted by synthetic MRI images using cycle-consistent GAN

To propose a style transfer model for multi-contrast magnetic resonance imaging (MRI) images with a cycle-consistent generative adversarial network (CycleGAN) and evaluate the image quality and prognosis prediction performance for glioblastoma (GBM) patients from the extracted radiomics features. Style transfer models of T1 weighted MRI image (T1w) to T2 weighted MRI image (T2w) and T2w to T1w with CycleGAN were constructed using the BraTS dataset. The style transfer model was validated with the Cancer Genome Atlas Glioblastoma Multiforme (TCGA-GBM) dataset. Moreover, imaging features were extracted from real and synthesized images. These features were transformed to rad-scores by the least absolute shrinkage and selection operator (LASSO)-Cox regression. The prognosis performance was estimated by the Kaplan-Meier method. For the accuracy of the image quality of the real and synthesized MRI images, the MI, RMSE, PSNR, and SSIM were 0.991 ± 2.10 × 10 - 4 , 2.79 ± 0.16, 40.16 ± 0.38, and 0.995 ± 2.11 × 10 - 4 , for T2w, and .992 ± 2.63 × 10 - 4 , 2.49 ± 6.89 × 10 - 2 , 40.51 ± 0.22, and 0.993 ± 3.40 × 10 - 4 for T1w, respectively. The survival time had a significant difference between good and poor prognosis groups for both real and synthesized T2w (p < 0.05). However, the survival time had no significant difference between good and poor prognosis groups for both real and synthesized T1w. On the other hand, there was no significant difference between the real and synthesized T2w in both good and poor prognoses. The results of T1w were similar in the point that there was no significant difference between the real and synthesized T1w. It was found that the synthesized image could be used for prognosis prediction. The proposed prognostic model using CycleGAN could reduce the cost and time of image scanning, leading to a promotion to build the patient's outcome prediction with multi-contrast images.

Identification of Multiclass Pediatric Low-Grade Neuroepithelial Tumor Molecular Subtype with ADC MR Imaging and Machine Learning

BACKGROUND AND PURPOSE

Molecular biomarker identification increasingly influences the treatment planning of pediatric low-grade neuroepithelial tumors (PLGNTs). We aimed to develop and validate a radiomics-based ADC signature predictive of the molecular status of PLGNTs.

MATERIALS AND METHODS

In this retrospective bi-institutional study, we searched the PACS for baseline brain MRIs from children with PLGNTs. Semiautomated tumor segmentation on ADC maps was performed using the semiautomated level tracing effect tool with 3D Slicer. Clinical variables, including age, sex, and tumor location, were collected from chart review. The molecular status of tumors was derived from biopsy. Multiclass random forests were used to predict the molecular status and fine-tuned using a grid search on the validation sets. Models were evaluated using independent and unseen test sets based on the combined data, and the area under the receiver operating characteristic curve (AUC) was calculated for the prediction of 3 classes: KIAA1549-BRAF fusion, BRAF V600E mutation, and non-BRAF cohorts. Experiments were repeated 100 times using different random data splits and model initializations to ensure reproducible results.

RESULTS

Two hundred ninety-nine children from the first institution and 23 children from the second institution were included (53.6% male; mean, age 8.01 years; 51.8% supratentorial; 52.2% with KIAA1549-BRAF fusion). For the 3-class prediction using radiomics features only, the average test AUC was 0.74 (95% CI, 0.73-0.75), and using clinical features only, the average test AUC was 0.67 (95% CI, 0.66-0.68). The combination of both radiomics and clinical features improved the AUC to 0.77 (95% CI, 0.75-0.77). The diagnostic performance of the per-class test AUC was higher in identifying KIAA1549-BRAF fusion tumors among the other subgroups (AUC = 0.81 for the combined radiomics and clinical features versus 0.75 and 0.74 for BRAF V600E mutation and non-BRAF, respectively).

CONCLUSIONS

ADC values of tumor segmentations have differentiative signals that can be used for training machine learning classifiers for molecular biomarker identification of PLGNTs. ADC-based pretherapeutic differentiation of the BRAF status of PLGNTs has the potential to avoid invasive tumor biopsy and enable earlier initiation of targeted therapy.

Comparison of MRI Sequences to Predict IDH Mutation Status in Gliomas Using Radiomics-Based Machine Learning

OBJECTIVES

Regarding the 2021 World Health Organization (WHO) classification of central nervous system (CNS) tumors, the isocitrate dehydrogenase (IDH) mutation status is one of the most important factors for CNS tumor classification. The aim of our study is to analyze which of the commonly used magnetic resonance imaging (MRI) sequences is best suited to obtain this information non-invasively using radiomics-based machine learning models. We developed machine learning models based on different MRI sequences and determined which of the MRI sequences analyzed yields the highest discriminatory power in predicting the IDH mutation status.

MATERIAL AND METHODS

In our retrospective IRB-approved study, we used the MRI images of 106 patients with histologically confirmed gliomas. The MRI images were acquired using the T1 sequence with and without administration of a contrast agent, the T2 sequence, and the Fluid-Attenuated Inversion Recovery (FLAIR) sequence. To objectively compare performance in predicting the IDH mutation status as a function of the MRI sequence used, we included only patients in our study cohort for whom MRI images of all four sequences were available. Seventy-one of the patients had an IDH mutation, and the remaining 35 patients did not have an IDH mutation (IDH wild-type). For each of the four MRI sequences used, 107 radiomic features were extracted from the corresponding MRI images by hand-delineated regions of interest. Data partitioning into training data and independent test data was repeated 100 times to avoid random effects associated with the data partitioning. Feature preselection and subsequent model development were performed using Random Forest, Lasso regression, LDA, and Naïve Bayes. The performance of all models was determined with independent test data.

RESULTS

Among the different approaches we examined, the T1-weighted contrast-enhanced sequence was found to be the most suitable for predicting IDH mutations status using radiomics-based machine learning models. Using contrast-enhanced T1-weighted MRI images, our seven-feature model developed with Lasso regression achieved a mean area under the curve (AUC) of 0.846, a mean accuracy of 0.792, a mean sensitivity of 0.847, and a mean specificity of 0.681. The administration of contrast agents resulted in a significant increase in the achieved discriminatory power.

CONCLUSIONS

Our analyses show that for the prediction of the IDH mutation status using radiomics-based machine learning models, among the MRI images acquired with the commonly used MRI sequences, the contrast-enhanced T1-weighted images are the most suitable.

The effect of data resampling methods in radiomics

Radiomic datasets can be class-imbalanced, for instance, when the prevalence of diseases varies notably, meaning that the number of positive samples is much smaller than that of negative samples. In these cases, the majority class may dominate the model's training and thus negatively affect the model's predictive performance, leading to bias. Therefore, resampling methods are often utilized to class-balance the data. However, several resampling methods exist, and neither their relative predictive performance nor their impact on feature selection has been systematically analyzed. In this study, we aimed to measure the impact of nine resampling methods on radiomic models utilizing a set of fifteen publicly available datasets regarding their predictive performance. Furthermore, we evaluated the agreement and similarity of the set of selected features. Our results show that applying resampling methods did not improve the predictive performance on average. On specific datasets, slight improvements in predictive performance (+ 0.015 in AUC) could be seen. A considerable disagreement on the set of selected features was seen (only 28.7% of features agreed), which strongly impedes feature interpretability. However, selected features are similar when considering their correlation (82.9% of features correlated on average).

Study on the Relationship Between MRI Functional Imaging and Multiple Immunohistochemical Features of Glioma: A Noninvasive and More Precise Glioma Management

Objective

To investigate the performance of diffusion-tensor imaging (DTI) and hydrogen proton magnetic resonance spectroscopy (1H-MRS) parameters in predicting the immunohistochemistry (IHC) biomarkers of glioma.

Methods

Patients with glioma confirmed by pathology from March 2015 to September 2019 were analyzed, the preoperative DTI and 1H-MRS images were collected, apparent diffusion coefficient (ADC) and fractional anisotropy (FA), in the lesion area were measured, the relative values relative ADC (rADC) and relative FA (rFA) were obtained by the ratio of them in the lesion area to the contralateral normal area. The peak of each metabolite in the lesion area of 1H-MRS image: N-acetylaspartate (NAA), choline (Cho), and creatine (Cr), and metabolite ratio: NAA/Cho, NAA/(Cho + Cr) were selected and calculated. The preoperative IHC data were collected including CD34, Ki-67, p53, S-100, syn, vimentin, NeuN, Nestin, and glial fibrillary acidic protein.

Results

One predicting parameter of DTI was screened, the rADC of the Ki-67 positive group was lower than that of the negative group. Two parameters of 1H-MRS were found to have significant reference values for glioma grades, the NAA and Cr decreased as the grade of glioma increased, moreover, Ki-67 Li was negatively correlated with NAA and Cr.

Conclusion

NAA and Cr have potential application value in predicting glioma grades and tumor proliferation activity. Only rADC has predictive value for Ki-67 expression among DTI parameters.

Radiomics features from 3D-MPRAGE imaging can differentiate temporal-plus epilepsy from temporal lobe epilepsy

OBJECTIVE

This study aimed to differentiate temporal-plus epilepsy (TPE) from temporal lobe epilepsy (TLE) using extraction of radiomics features from three-dimensional magnetization-prepared rapid acquisition gradient echo (3D-MPRAGE) imaging data.

METHODS

Data from patients with TLE or TPE who underwent epilepsy surgery between January 2019 and January 2021 were retrospectively analyzed. Thirty-three regions of interest in the affected hemisphere of each patient were defined on 3D-MPRAGE images. A total of 3531 image features were extracted from each patient. Four feature selection methods and 10 machine learning algorithms were used to build 40 differentiation models. Model performance was evaluated using receiver operating characteristic analysis.

RESULTS

Eighty-two patients were included for analysis, 47 with TLE and 35 with TPE. The model combining logistic regression and the relief selection method had the best performance (area under the receiver operating characteristic curve, .779; accuracy, .875; sensitivity, .800; specificity, .929; positive predictive value, .889; negative predictive value, .867).

SIGNIFICANCE

Radiomics analysis can differentiate TPE from TLE. The logistic regression classifier trained with radiomics features extracted from 3D-MPRAGE images had the highest accuracy and best performance.

Prediction of IDH1 gene mutation by a nomogram based on multiparametric and multiregional MR images

OBJECTIVE

To investigate the value of a nomogram based on multiparametric and multiregional MR images to predict Isocitrate Dehydrogenase-1 (IDH1) gene mutations in glioma.

DATA AND METHODS

The authors performed a retrospective analysis of 110 MR images of surgically confirmed pathological gliomas; 33 patients with IDH1 gene Mutation (IDH1-M) and 77 patients with Wild-type IDH1 (IDH1-W) were divided into training and validation sets in a 7:3 ratio. The clinical features were statistically analyzed using SPSS and R software. Three glioma regions (rCET, rE, rNEC) were outlined using ITK-SNAP software and projected to four conventional sequences (T1, T2, Flair, T1C) for feature extraction using AI-Kit software. The extracted features were screened using R software. A logistic regression model was established, and a nomogram was generated using the selected clinical features. Eight models were developed based on different sequences and ROIs, and Receiver Operating Characteristic (ROC) curves were used to evaluate the predictive efficacy. Decision curve analysis was performed to assess the clinical usefulness.

RESULTS

Age was selected with Radscore to construct the nomogram. The Model 1 AUC values based on four sequences and three ROIs were the highest in these models, at 0.93 and 0.89, respectively. Decision curve analysis indicated that the net benefit of model 1 was higher than that of the other models for most Pt-values.

CONCLUSION

A nomogram based on multiparametric and multiregional MR images can predict the mutation status of the IDH1 gene accurately.

Radiomic Features Based on MRI Predict Progression-Free Survival in Pediatric Diffuse Midline Glioma/Diffuse Intrinsic Pontine Glioma

Purpose: Biopsy-based assessment of H3 K27 M status helps in predicting survival, but biopsy is usually limited to unusual presentations and clinical trials. We aimed to evaluate whether radiomics can serve as prognostic marker to stratify diffuse intrinsic pontine glioma (DIPG) subsets. Methods: In this retrospective study, diagnostic brain MRIs of children with DIPG were analyzed. Radiomic features were extracted from tumor segmentations and data were split into training/testing sets (80:20). A conditional survival forest model was applied to predict progression-free survival (PFS) using training data. The trained model was validated on the test data, and concordances were calculated for PFS. Experiments were repeated 100 times using randomized versions of the respective percentage of the training/test data. Results: A total of 89 patients were identified (48 females, 53.9%). Median age at time of diagnosis was 6.64 years (range: 1-16.9 years) and median PFS was 8 months (range: 1-84 months). Molecular data were available for 26 patients (29.2%) (1 wild type, 3 K27M-H3.1, 22 K27M-H3.3). Radiomic features of FLAIR and nonenhanced T1-weighted sequences were predictive of PFS. The best FLAIR radiomics model yielded a concordance of .87 [95% CI: .86-.88] at 4 months PFS. The best T1-weighted radiomics model yielded a concordance of .82 [95% CI: .8-.84] at 4 months PFS. The best combined FLAIR + T1-weighted radiomics model yielded a concordance of .74 [95% CI: .71-.77] at 3 months PFS. The predominant predictive radiomic feature matrix was gray-level size-zone. Conclusion: MRI-based radiomics may predict progression-free survival in pediatric diffuse midline glioma/diffuse intrinsic pontine glioma.

An integrative non-invasive malignant brain tumors classification and Ki-67 labeling index prediction pipeline with radiomics approach

BACKGROUND

The histological sub-classes of brain tumors and the Ki-67 labeling index (LI) of tumor cells are major factors in the diagnosis, prognosis, and treatment management of patients. Many existing studies primarily focused on the classification of two classes of brain tumors and the Ki-67LI of gliomas. This study aimed to develop a preoperative non-invasive radiomics pipeline based on multiparametric-MRI to classify-three types of brain tumors, glioblastoma (GBM), metastasis (MET) and primary central nervous system lymphoma (PCNSL), and to predict their corresponding Ki-67LI.

METHODS

In this retrospective study, 153 patients with malignant brain tumors were involved. The radiomics features were extracted from three types of MRI (T1-weighted imaging (T1WI), fluid-attenuated inversion recovery (FLAIR), and contrast-enhanced T1-weighted imaging (CE-T1WI)) with three masks (tumor core, edema, and whole tumor masks) and selected by a combination of Pearson correlation coefficient (CORR), LASSO, and Max-Relevance and Min-Redundancy (mRMR) filters. The performance of six classifiers was compared and the top three performing classifiers were used to construct the ensemble learning model (ELM). The proposed ELM was evaluated in the training dataset (108 patients) by 5-fold cross-validation and in the test dataset (45 patients) by hold-out. The accuracy (ACC), sensitivity (SEN), specificity (SPE), F1-Score, and the area under the receiver operating characteristic curve (AUC) indicators evaluated the performance of the models.

RESULTS

The best feature sets and ELM with the optimal performance were selected to construct the tri-categorized brain tumor aided diagnosis model (training dataset AUC: 0.96 (95% CI: 0.93, 0.99); test dataset AUC: 0.93) and Ki-67LI prediction model (training dataset AUC: 0.96 (95% CI: 0.94, 0.98); test dataset AUC: 0.91). The CE-T1WI was the best single modality for all classifiers. Meanwhile, the whole tumor was the most vital mask for the tumor classification and the tumor core was the most vital mask for the Ki-67LI prediction.

CONCLUSION

The developed radiomics models led to the precise preoperative classification of GBM, MET, and PCNSL and the prediction of Ki-67LI, which could be utilized in clinical practice for the treatment planning for brain tumors.

Radiomics analysis of amide proton transfer-weighted and structural MR images for treatment response assessment in malignant gliomas

The purpose of this study was to evaluate the value of amide proton transfer-weighted (APTw) MRI radiomic features for the differentiation of tumor recurrence from treatment effect in malignant gliomas. Eighty-six patients who had suspected tumor recurrence after completion of chemoradiation or radiotherapy, and who had APTw-MRI data acquired at 3 T, were retrospectively analyzed. Using a fluid-attenuated inversion recovery (FLAIR) image-based mask, radiomics analysis was applied to the processed APTw and structural MR images. A chi-square automatic interaction detector decision tree was used for classification analysis. Models with and without APTw features were built using the same strategy. Tenfold cross-validation was applied to obtain the overall classification performance of each model. Sixty patients were confirmed as having tumor recurrence, and the remainder were confirmed as having treatment effect, at median time points of 190 and 171 days after therapy, respectively. There were 525 radiomic features extracted from each of the processed APTw and structural MR images. Based on these, the APTw-based model yielded the highest accuracy (86.0%) for the differentiation of tumor recurrence from treatment effect, compared with 74.4%, 76.7%, 83.7%, and 76.7% for T1 w, T2 w, FLAIR, and Gd-T1 w, respectively. Model classification accuracy was 82.6% when using the combined structural MR images (T1 w, T2 w, FLAIR, Gd-T1 w), and increased to 89.5% when using these structural plus APTw images. The corresponding sensitivity and specificity were 85.0% and 76.9% for the combination of structural MR images, and 85.0% and 100% after adding APTw image features. Adding APTw-based radiomic features increased MRI accuracy in the assessment of the treatment response in post-treatment malignant gliomas.

MRI based radiomics enhances prediction of neurodevelopmental outcome in very preterm neonates

To predict adverse neurodevelopmental outcome of very preterm neonates. A total of 166 preterm neonates born between 24–32 weeks’ gestation underwent brain MRI early in life. Radiomics features were extracted from T1- and T2- weighted images. Motor, cognitive, and language outcomes were assessed at a corrected age of 18 and 33 months and 4.5 years. Elastic Net was implemented to select the clinical and radiomic features that best predicted outcome. The area under the receiver operating characteristic (AUROC) curve was used to determine the predictive ability of each feature set. Clinical variables predicted cognitive outcome at 18 months with AUROC 0.76 and motor outcome at 4.5 years with AUROC 0.78. T1-radiomics features showed better prediction than T2-radiomics on the total motor outcome at 18 months and gross motor outcome at 33 months (AUROC: 0.81 vs 0.66 and 0.77 vs 0.7). T2-radiomics features were superior in two 4.5-year motor outcomes (AUROC: 0.78 vs 0.64 and 0.8 vs 0.57). Combining clinical parameters and radiomics features improved model performance in motor outcome at 4.5 years (AUROC: 0.84 vs 0.8). Radiomic features outperformed clinical variables for the prediction of adverse motor outcomes. Adding clinical variables to the radiomics model enhanced predictive performance.

Radiomics-Based Machine Learning Models for Predicting P504s/P63 Immunohistochemical Expression: A Noninvasive Diagnostic Tool for Prostate Cancer

Objective

To develop and validate a noninvasive radiomic-based machine learning (ML) model to identify P504s/P63 status and further achieve the diagnosis of prostate cancer (PCa).

Methods

A retrospective dataset of patients with preoperative prostate MRI examination and P504s/P63 pathological immunohistochemical results between June 2016 and February 2021 was conducted. As indicated by P504s/P63 expression, the patients were divided into label 0 (atypical prostatic hyperplasia), label 1 (benign prostatic hyperplasia, BPH) and label 2 (PCa) groups. This study employed T2WI, DWI and ADC sequences to assess prostate diseases and manually segmented regions of interest (ROIs) with Artificial Intelligence Kit software for radiomics feature acquisition. Feature dimensionality reduction and selection were performed by using a mutual information algorithm. Based on screened features, P504s/P63 prediction models were established by random forest (RF), gradient boosting decision tree (GBDT), logistic regression (LR), adaptive boosting (AdaBoost) and k-nearest neighbor (KNN) algorithms. The performance was evaluated by the area under the ROC curve (AUC) and accuracy.

Results

A total of 315 patients were enrolled. Among the 851 radiomic features, the 32 top features were derived from T2WI, in which the gray-level run length matrix (GLRLM) and gray-level cooccurrence matrix (GLCM) features accounted for the largest proportion. Among the five models, the RF algorithm performed best in general evaluations (microaverage AUC=0.920, macroaverage AUC=0.870) and provided the most accurate result in further sublabel prediction (the accuracies of label 0, 1, and 2 were 0.831, 0.831, and 0.932, respectively). In comparative sequence analyses, T2WI was the best single-sequence candidate (microaverage AUC=0.94 and macroaverage AUC=0.78). The merged datasets of T2WI, DWI, and ADC yielded optimal AUCs (microaverage AUC=0.930 and macroaverage AUC=0.900).

Conclusions

The radiomic-based RF classifier has the potential to be used to evaluate the presurgical P504s/P63 status and further diagnose PCa noninvasively and accurately.

Multi-parameter diffusion and perfusion magnetic resonance imaging and radiomics nomogram for preoperative evaluation of aquaporin-1 expression in rectal cancer

PURPOSE

The overexpression of aquaporin-1 (AQP1) is associated with poor prognosis in rectal cancer. This study aimed to explore the value of multi-parameter diffusion and perfusion MRI and radiomics models in predicting AQP1 high expression.

METHODS

This prospective study was performed from July 2019 to February 2021, which included rectal cancer participants after preoperative rectal MRI, with diffusion-weighted imaging, intravoxel incoherent motion (IVIM), diffusion kurtosis imaging (DKI), and dynamic contrast-enhanced (DCE) sequences. Radiomic features were extracted from MR images, and immunohistochemical tests assessed AQP1 expression. Selected quantitative MRI and radiomic features were analyzed. Receiver operating characteristic (ROC) curves evaluated the predictive performance. The nomogram performance was evaluated by its calibration, discrimen, and clinical utility. The intraclass correlation coefficient evaluated the interobserver agreement for the MRI features.

RESULTS

110 participants with the age of 60.7 ± 12.5 years been enrolled in this study. The apparent diffusion coefficient (ADC), IVIM_D, DKI_diffusivity, and DCE_Ktrans were significantly higher in participants with high AQP1 expression than in those with low expression (P < 0.05). ADC (b = 1000, 2000, and 3000 s/mm²), IVIM_D, DKI_diffusivity, and DCE_Ktrans were positively correlated (r = 0.205, 0.275, 0.37, 0.235, 0.229, and 0.227, respectively; P < 0.05), whereas DKI_Kurtosis was negatively correlated (r = - 0.22, P = 0.021) with AQP1 expression. ADC (b = 3000 s/mm²), IVIM_D, DKI_ diffusivity, DKI_Kurtosis, and DCE_Ktrans had moderate diagnostic efficiencies for high AQP1 expression (AUC = 0.715, 0.636, 0.627, 0.633, and 0.632, respectively; P < 0.05). The radiomic features had excellent predictive efficiency for high AQP1 expression (AUC = 0.967 and 0.917 for training and validation). The model-based nomogram had C-indexes of 0.932 and 0.851 for the training and validation cohorts, which indicated good fitting to the calibration curves (p > 0.05).

CONCLUSION

Diffusion and perfusion MRI can indicate the aquaporin-1 expression in rectal cancer, and radiomic features can enhance the predictive efficiency for high AQP1 expression. A nomogram for high aquaporin-1 expression will improve clinical decision-making.

Evaluation of the dependence of radiomic features on the machine learning model

Background

In radiomic studies, several models are often trained with different combinations of feature selection methods and classifiers. The features of the best model are usually considered relevant to the problem, and they represent potential biomarkers. Features selected from statistically similarly performing models are generally not studied. To understand the degree to which the selected features of these statistically similar models differ, 14 publicly available datasets, 8 feature selection methods, and 8 classifiers were used in this retrospective study. For each combination of feature selection and classifier, a model was trained, and its performance was measured with AUC-ROC. The best-performing model was compared to other models using a DeLong test. Models that were statistically similar were compared in terms of their selected features.

Results

Approximately 57% of all models analyzed were statistically similar to the best-performing model. Feature selection methods were, in general, relatively unstable (0.58; range 0.35–0.84). The features selected by different models varied largely (0.19; range 0.02–0.42), although the selected features themselves were highly correlated (0.71; range 0.4–0.92).

Conclusions

Feature relevance in radiomics strongly depends on the model used, and statistically similar models will generally identify different features as relevant. Considering features selected by a single model is misleading, and it is often not possible to directly determine whether such features are candidate biomarkers.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13244-022-01170-2.

Advanced Imaging Techniques for Newly Diagnosed and Recurrent Gliomas

Management of gliomas following initial diagnosis requires thoughtful presurgical planning followed by regular imaging to monitor treatment response and survey for new tumor growth. Traditional MR imaging modalities such as T1 post-contrast and T2-weighted sequences have long been a staple of tumor diagnosis, surgical planning, and post-treatment surveillance. While these sequences remain integral in the management of gliomas, advances in imaging techniques have allowed for a more detailed characterization of tumor characteristics. Advanced MR sequences such as perfusion, diffusion, and susceptibility weighted imaging, as well as PET scans have emerged as valuable tools to inform clinical decision making and provide a non-invasive way to help distinguish between tumor recurrence and pseudoprogression. Furthermore, these advances in imaging have extended to the operating room and assist in making surgical resections safer. Nevertheless, surgery, chemotherapy, and radiation treatment continue to make the interpretation of MR changes difficult for glioma patients. As analytics and machine learning techniques improve, radiomics offers the potential to be more quantitative and personalized in the interpretation of imaging data for gliomas. In this review, we describe the role of these newer imaging modalities during the different stages of management for patients with gliomas, focusing on the pre-operative, post-operative, and surveillance periods. Finally, we discuss radiomics as a means of promoting personalized patient care in the future.

Benchmarking Feature Selection Methods in Radiomics

OBJECTIVES

A critical problem in radiomic studies is the high dimensionality of the datasets, which stems from small sample sizes and many generic features extracted from the volume of interest. Therefore, feature selection methods are used, which aim to remove redundant as well as irrelevant features. Because there are many feature selection algorithms, it is key to understand their performance in the context of radiomics.

MATERIALS AND METHODS

A total of 29 feature selection algorithms and 10 classifiers were evaluated on 10 publicly available radiomic datasets. Feature selection methods were compared for training times, for the stability of the selected features, and for ranking, which measures the pairwise similarity of the methods. In addition, the predictive performance of the algorithms was measured by utilizing the area under the receiver operating characteristic curve of the best-performing classifier.

RESULTS

Feature selections differed largely in training times as well as stability and similarity. No single method was able to outperform another one consistently in predictive performance.

CONCLUSION

Our results indicated that simpler methods are more stable than complex ones and do not perform worse in terms of area under the receiver operating characteristic curve. Analysis of variance, least absolute shrinkage and selection operator, and minimum redundancy, maximum relevance ensemble appear to be good choices for radiomic studies in terms of predictive performance, as they outperformed most other feature selection methods.

Radiomic signatures based on multiparametric MR images for predicting Ki-67 index expression in medulloblastoma

BACKGROUND

Medulloblastoma (MB) is a common central nervous system tumor in children with extensive heterogeneity and different prognoses. This study aimed to classify the Ki-67 index in MB with radiomic characteristics based on multi-parametric magnetic resonance imaging to guide treatment and assess the prognosis of patients.

METHODS

Three sequences of T1W, CE-T1W, and T2W were used as test data. Two experienced radiologists manually segmented the tumors according to T2W images from 90 patients. The patients were divided into training and test sets at a ratio of 7:3, and 833 dimensional image features were extracted for each patient. Five models were trained using the feature set selected in three ways. Finally, the area under the curve (AUC) and accuracy (ACC) were used on the test set to evaluate the performance of the different models.

RESULTS

A random forest (RF) model combining three sequence features achieved the best performance (ACC: 0.771, 95% CI: 0.727 to 0.816; AUC: 0.697, 95% CI: 0.614 to 0.78). The voting model that combined a RF and a support vector machine (SVM) had higher performance than the other models (ACC: 0.796, 95% CI: 0.76 to 0.833; AUC: 0.689, 95% CI: 0.615 to 0.763). The best prediction model that used only one sequence feature was voting in the T2W sequence (ACC: 0.736, 95% CI: 0.705 to 0.766; AUC: 0.636, 95% CI: 0.585 to 0.688). The ensemble model was better than the single training model, and a multi-sequence combination was better than a single sequence prediction. The multiple feature selection methods were better than a combination of the two methods.

CONCLUSIONS

A model obtained by machine learning could help doctors predict the Ki-67 values of patients more efficiently to make targeted judgments for subsequent treatments.

Texture analysis imaging "what a clinical radiologist needs to know"

Texture analysis has arisen as a tool to explore the amount of data contained in images that cannot be explored by humans visually. Radiomics is a method that extracts a large number of features from radiographic medical images using data-characterisation algorithms. These features, termed radiomic features, have the potential to uncover disease characteristics. The goal of both radiomics and texture analysis is to go beyond size or human-eye based semantic descriptors, to enable the non-invasive extraction of quantitative radiological data to correlate them with clinical outcomes or pathological characteristics. In the latest years there has been a flourishing sub-field of radiology where texture analysis and radiomics have been used in many settings. It is difficult for the clinical radiologist to cope with such amount of data in all the different radiological sub-fields and to identify the most significant papers. The aim of this review is to provide a tool to better understand the basic principles underlining texture analysis and radiological data mining and a summary of the most significant papers of the latest years.

Measuring the bias of incorrect application of feature selection when using cross-validation in radiomics

BACKGROUND

Many studies in radiomics are using feature selection methods to identify the most predictive features. At the same time, they employ cross-validation to estimate the performance of the developed models. However, if the feature selection is performed before the cross-validation, data leakage can occur, and the results can be biased. To measure the extent of this bias, we collected ten publicly available radiomics datasets and conducted two experiments. First, the models were developed by incorrectly applying the feature selection prior to cross-validation. Then, the same experiment was conducted by applying feature selection correctly within cross-validation to each fold. The resulting models were then evaluated against each other in terms of AUC-ROC, AUC-F1, and Accuracy.

RESULTS

Applying the feature selection incorrectly prior to the cross-validation showed a bias of up to 0.15 in AUC-ROC, 0.29 in AUC-F1, and 0.17 in Accuracy.

CONCLUSIONS

Incorrect application of feature selection and cross-validation can lead to highly biased results for radiomic datasets.

Radiomics features based on T2-weighted fluid-attenuated inversion recovery MRI predict the expression levels of CD44 and CD133 in lower-grade gliomas

Objective: To verify the association between CD44 and CD133 expression levels and the prognosis of patients with lower-grade gliomas (LGGs) and constructing radiomic models to predict those two genes' expression levels before surgery. Materials & methods: Genomic data of patients with LGG and the corresponding T2-weighted fluid-attenuated inversion recovery images were downloaded from the Cancer Genome Atlas and the Cancer Imaging Archive, which were utilized for prognosis analysis, radiomic feature extraction and model construction, respectively. Results & conclusion: CD44 and CD133 expression levels in LGG can significantly affect the prognosis of patients with LGG. Based on the T2-weighted fluid-attenuated inversion recovery images, the radiomic features can effectively predict the expression levels of CD44 and CD133 before surgery.

Evolving Role and Translation of Radiomics and Radiogenomics in Adult and Pediatric Neuro-Oncology

Exponential technologic advancements in imaging, high-performance computing, and artificial intelligence, in addition to increasing access to vast amounts of diverse data, have revolutionized the role of imaging in medicine. Radiomics is defined as a high-throughput feature-extraction method that unlocks microscale quantitative data hidden within standard-of-care medical imaging. Radiogenomics is defined as the linkage between imaging and genomics information. Multiple radiomics and radiogenomics studies performed on conventional and advanced neuro-oncology image modalities show that they have the potential to differentiate pseudoprogression from true progression, classify tumor subgroups, and predict recurrence, survival, and mutation status with high accuracy. In this article, we outline the technical steps involved in radiomics and radiogenomics analyses with the use of artificial intelligence methods and review current applications in adult and pediatric neuro-oncology.

Systematic Review on the Association of Radiomics with Tumor Biological Endpoints

Radiomics supposes an alternative non-invasive tumor characterization tool, which has experienced increased interest with the advent of more powerful computers and more sophisticated machine learning algorithms. Nonetheless, the incorporation of radiomics in cancer clinical-decision support systems still necessitates a thorough analysis of its relationship with tumor biology. Herein, we present a systematic review focusing on the clinical evidence of radiomics as a surrogate method for tumor molecular profile characterization. An extensive literature review was conducted in PubMed, including papers on radiomics and a selected set of clinically relevant and commonly used tumor molecular markers. We summarized our findings based on different cancer entities, additionally evaluating the effect of different modalities for the prediction of biomarkers at each tumor site. Results suggest the existence of an association between the studied biomarkers and radiomics from different modalities and different tumor sites, even though a larger number of multi-center studies are required to further validate the reported outcomes.

Development of CT-Based Imaging Signature for Preoperative Prediction of Invasive Behavior in Pancreatic Solid Pseudopapillary Neoplasm

Purpose

It is challenging for traditional CT signs to predict invasiveness of pancreatic solid pseudopapillary neoplasm (pSPN). We aim to develop and evaluate CT-based radiomics signature to preoperatively predict invasive behavior in pSPN.

Methods

Eighty-five patients who had pathologically confirmed pSPN and preoperative contrasted-enhanced CT imaging in our hospital were retrospectively analyzed (invasive: 24; non-invasive: 61). 1316 radiomics features were separately extracted from delineated 2D or 3D ROIs in arterial and venous phases. 200% (SMOTE) was used to generate balanced dataset (invasive: 72, non-invasive: 96) for each phase, which was for feature selection and modeling. The model was internally validated in the original dataset. Inter-observer consistency analysis, spearman correlation, univariate analysis, LASSO regression and backward stepwise logical regression were mainly applied to screen the features, and 6 logistic regression models were established based on multi-phase features from 2D or 3D segmentations. The ROC analysis and Delong's test were mainly used for model assessment and AUC comparison.

Results

It retained 11, 8, 7 and 7 features to construct 3D-arterial, 3D-venous, 2D-arterial and 2D-venous model. Based on 3D ROIs, the arterial model (AUC: 0.914) performed better than venous (AUC: 0.815) and the arterial-venous combined model was slightly improved (AUC: 0.918). Based on 2D ROIs, the arterial model (AUC: 0.814) performed better than venous (AUC:0.768), while the arterial-venous combined model (AUC:0.893) performed better than any single-phase model. In addition, the 3D arterial model performed better than the best combined 2D model. The Delong's test showed that the significant difference of model AUC existed in arterial models in original dataset (p = 0.019) while not in arterial-venous combined model (p=0.49) as comparing 2D and 3D ROIs.

Conclusion

The arterial radiomics model constructed by 3D-ROI feature is potential to predict the invasiveness of pSPN preoperatively.

Multiparametric MRI texture analysis in prediction of glioma biomarker status: added value of MR diffusion

Background

Early identification of glioma molecular phenotypes can lead to understanding of patient prognosis and treatment guidance. We aimed to develop a multiparametric MRI texture analysis model using a combination of conventional and diffusion MRI to predict a wide range of biomarkers in patients with glioma.

Methods

In this retrospective study, patients were included if they (1) had diagnosis of gliomas with known IDH1, EGFR, MGMT, ATRX, TP53, and PTEN status from surgical pathology and (2) had preoperative MRI including FLAIR, T1c+ and diffusion for radiomic texture analysis. Statistical analysis included logistic regression and receiver-operating characteristic (ROC) curve analysis to determine the optimal model for predicting glioma biomarkers. A comparative analysis between ROCs (conventional only vs conventional + diffusion) was performed.

Results

From a total of 111 patients included, 91 (82%) were categorized to training and 20 (18%) to test datasets. Constructed cross-validated model using a combination of texture features from conventional and diffusion MRI resulted in overall AUC/accuracy of 1/79% for IDH1, 0.99/80% for ATRX, 0.79/67% for MGMT, and 0.77/66% for EGFR. The addition of diffusion data to conventional MRI features significantly (P < .05) increased predictive performance for IDH1, MGMT, and ATRX. The overall accuracy of the final model in predicting biomarkers in the test group was 80% (IDH1), 70% (ATRX), 70% (MGMT), and 75% (EGFR).

Conclusion

Addition of MR diffusion to conventional MRI features provides added diagnostic value in preoperative determination of IDH1, MGMT, and ATRX in patients with glioma.

MRI-Based Radiomics and Radiogenomics in the Management of Low-Grade Gliomas: Evaluating the Evidence for a Paradigm Shift

Low-grade gliomas (LGGs) are tumors that affect mostly adults. These neoplasms are comprised mainly of oligodendrogliomas and diffuse astrocytomas. LGGs remain vexing to current management and therapeutic modalities although they exhibit more favorable survival rates compared with high-grade gliomas (HGGs). The specific genetic subtypes that these tumors exhibit result in variable clinical courses and the need to involve multidisciplinary teams of neurologists, epileptologists, neurooncologists and neurosurgeons. Currently, the diagnosis of an LGG pivots mainly around the preliminary radiological findings and the subsequent definitive surgical diagnosis (via surgical sampling). The introduction of radiomics as a high throughput quantitative imaging technique that allows for improved diagnostic, prognostic and predictive indices has created more interest for such techniques in cancer research and especially in neurooncology (MRI-based classification of LGGs, predicting Isocitrate dehydrogenase (IDH) and Telomerase reverse transcriptase (TERT) promoter mutations and predicting LGG associated seizures). Radiogenomics refers to the linkage of imaging findings with the tumor/tissue genomics. Numerous applications of radiomics and radiogenomics have been described in the clinical context and management of LGGs. In this review, we describe the recently published studies discussing the potential application of radiomics and radiogenomics in LGGs. We also highlight the potential pitfalls of the above-mentioned high throughput computerized techniques and, most excitingly, explore the use of machine learning artificial intelligence technologies as standalone and adjunct imaging tools en route to enhance a personalized MRI-based tumor diagnosis and management plan design.

Employment of Artificial Intelligence Based on Routine Laboratory Results for the Early Diagnosis of Multiple Myeloma

Objective

In order to enhance the detection rate of multiple myeloma and execute an early and more precise disease management, an artificial intelligence assistant diagnosis system is developed.

Methods

4,187 routine blood and biochemical examination records were collected from Shengjing Hospital affiliated to China Medical University from January 2010 to January 2020, which include 1,741 records of multiple myeloma (MM) and 2,446 records of non-myeloma (infectious diseases, rheumatic immune system diseases, hepatic diseases and renal diseases). The data set was split into training and test subsets with the ratio of 4:1 while connecting hemoglobin, serum creatinine, serum calcium, immunoglobulin (A, G and M), albumin, total protein, and the ratio of albumin to globulin data. An early assistant diagnostic model of MM was established by Gradient Boosting Decision Tree (GBDT), Support Vector Machine (SVM), Deep Neural Networks (DNN), and Random Forest (RF). Out team calculated the precision and recall of the system. The performance of the diagnostic model was evaluated by using the receiver operating characteristic (ROC) curve.

Results

By designing the features properly, the typical machine learning algorithms SVM, DNN, RF and GBDT all performed well. GBDT had the highest precision (92.9%), recall (90.0%) and F1 score (0.915) for the myeloma group. The maximized area under the ROC (AUROC) was calculated, and the results of GBDT (AUC: 0.975; 95% confidence interval (CI): 0.963–0.986) outperformed that of SVM, DNN and RF.

Conclusion

The model established by artificial intelligence derived from routine laboratory results can accurately diagnose MM, which can boost the rate of early diagnosis.

Radiomics of Pediatric Low-Grade Gliomas: Toward a Pretherapeutic Differentiation of BRAF-Mutated and BRAF-Fused Tumors

BACKGROUND AND PURPOSE

B-Raf proto-oncogene, serine/threonine kinase (BRAF) status has important implications for prognosis and therapy of pediatric low-grade gliomas. Currently, BRAF status classification relies on biopsy. Our aim was to train and validate a radiomics approach to predict BRAF fusion and BRAF V600E mutation.

MATERIALS AND METHODS

In this bi-institutional retrospective study, FLAIR MR imaging datasets of 115 pediatric patients with low-grade gliomas from 2 children's hospitals acquired between January 2009 and January 2016 were included and analyzed. Radiomics features were extracted from tumor segmentations, and the predictive model was tested using independent training and testing datasets, with all available tumor types. The model was selected on the basis of a grid search on the number of trees, opting for the best split for a random forest. We used the area under the receiver operating characteristic curve to evaluate model performance.

RESULTS

The training cohort consisted of 94 pediatric patients with low-grade gliomas (mean age, 9.4 years; 45 boys), and the external validation cohort comprised 21 pediatric patients with low-grade gliomas (mean age, 8.37 years; 12 boys). A 4-fold cross-validation scheme predicted BRAF status with an area under the curve of 0.75 (SD, 0.12) (95% confidence interval, 0.62-0.89) on the internal validation cohort. By means of the optimal hyperparameters determined by 4-fold cross-validation, the area under the curve for the external validation was 0.85. Age and tumor location were significant predictors of BRAF status (P values = .04 and <.001, respectively). Sex was not a significant predictor (P value = .96).

CONCLUSIONS

Radiomics-based prediction of BRAF status in pediatric low-grade gliomas appears feasible in this bi-institutional exploratory study.

Unraveling response to temozolomide in preclinical GL261 glioblastoma with MRI/MRSI using radiomics and signal source extraction

Glioblastoma is the most frequent aggressive primary brain tumor amongst human adults. Its standard treatment involves chemotherapy, for which the drug temozolomide is a common choice. These are heterogeneous and variable tumors which might benefit from personalized, data-based therapy strategies, and for which there is room for improvement in therapy response follow-up, investigated with preclinical models. This study addresses a preclinical question that involves distinguishing between treated and control (untreated) mice bearing glioblastoma, using machine learning techniques, from magnetic resonance-based data in two modalities: MRI and MRSI. It aims to go beyond the comparison of methods for such discrimination to provide an analytical pipeline that could be used in subsequent human studies. This analytical pipeline is meant to be a usable and interpretable tool for the radiology expert in the hope that such interpretation helps revealing new insights about the problem itself. For that, we propose coupling source extraction-based and radiomics-based data transformations with feature selection. Special attention is paid to the generation of radiologist-friendly visual nosological representations of the analyzed tumors.