Classification methods for the differentiation of atypical meningiomas using diffusion and perfusion techniques at 3-T MRI

Sample strategies for the assessment of the apparent diffusion coefficient in single large intracranial space-occupying lesions of dogs and cats

Diffusion-weighted imaging is increasingly available for brain investigation. Image interpretation of intracranial space-occupying lesions often includes the derived apparent diffusion coefficient (ADC) analysis. In human medicine, ADC can help discriminate between benign and malignant lesions in intracranial tumors. This study investigates the difference in ADC values depending on the sample strategies of image analysis. MRI examination, including diffusion-weighted images of canine and feline patients presented between 2015 and 2020, were reviewed retrospectively. Patients with single, large intracranial space-occupying lesions were included. Lesions homogeneity was subjectively scored. ADC values were calculated using six different methods of sampling (M1-M6) on the ADC map. M1 included as much as possible of the lesion on a maximum of five consecutive slices; M2 included five central and five peripheral ROIs; M3 included a single ROI on the solid part of the lesion; M4 included three central ROIs on one slice; M5 included three central ROIs on different slices; and M6 included one large ROI on the entire lesion. A total of 201 animals of various breeds, genders, and ages were analyzed. ADC values differed significantly between M5 against M2 (peripheral) (p < 0.001), M5 against M6 (p = 0.009), and M4 against M2 (peripheral) (p = 0.005). When lesions scored as homogeneous in all sequences were excluded, an additional significant difference in three further sampling methods was present (p < 0.005). ADC of single, large, intracranial space-occupying lesions differed significantly in half of the tested methods of sampling. Excluding homogeneous lesions, additional significant differences among the sampling methods were present. The obtained results should increase awareness of the variability of the ADC, depending on the sample strategies used.

Bayesian Networks in Radiology

A Bayesian network is a graphical model that uses probability theory to represent relationships among its variables. The model is a directed acyclic graph whose nodes represent variables, such as the presence of a disease or an imaging finding. Connections between nodes express causal influences between variables as probability values. Bayesian networks can learn their structure (nodes and connections) and/or conditional probability values from data. Bayesian networks offer several advantages: (a) they can efficiently perform complex inferences, (b) reason from cause to effect or vice versa, (c) assess counterfactual data, (d) integrate observations with canonical ("textbook") knowledge, and (e) explain their reasoning. Bayesian networks have been employed in a wide variety of applications in radiology, including diagnosis and treatment planning. Unlike deep learning approaches, Bayesian networks have not been applied to computer vision. However, hybrid artificial intelligence systems have combined deep learning models with Bayesian networks, where the deep learning model identifies findings in medical images and the Bayesian network formulates and explains a diagnosis from those findings. One can apply a Bayesian network's probabilistic knowledge to integrate clinical and imaging findings to support diagnosis, treatment planning, and clinical decision-making. This article reviews the fundamental principles of Bayesian networks and summarizes their applications in radiology. Keywords: Bayesian Network, Machine Learning, Abdominal Imaging, Musculoskeletal Imaging, Breast Imaging, Neurologic Imaging, Radiology Education Supplemental material is available for this article. © RSNA, 2023.

Use of advanced neuroimaging and artificial intelligence in meningiomas

Anatomical cross‐sectional imaging methods such as contrast‐enhanced MRI and CT are the standard for the delineation, treatment planning, and follow‐up of patients with meningioma. Besides, advanced neuroimaging is increasingly used to non‐invasively provide detailed insights into the molecular and metabolic features of meningiomas. These techniques are usually based on MRI, e.g., perfusion‐weighted imaging, diffusion‐weighted imaging, MR spectroscopy, and positron emission tomography. Furthermore, artificial intelligence methods such as radiomics offer the potential to extract quantitative imaging features from routinely acquired anatomical MRI and CT scans and advanced imaging techniques. This allows the linking of imaging phenotypes to meningioma characteristics, e.g., the molecular‐genetic profile. Here, we review several diagnostic applications and future directions of these advanced neuroimaging techniques, including radiomics in preclinical models and patients with meningioma.

Comparison of Canine and Feline Meningiomas Using the Apparent Diffusion Coefficient and Fractional Anisotropy

Meningiomas are the most common intracranial tumor in dogs and cats, and their surgical resection is often performed because they are present on the brain surface. Typical meningiomas show comparatively characteristic magnetic resonance imaging findings that lead to clinical diagnosis; however, it is necessary to capture not only macroscopic changes but also microstructural changes to devise a strategy for surgical resection and/or quality of removal. To visualize such microstructural changes, diffusion-weighted imaging (DWI) and diffusion tensor imaging (DTI) have been used in human medicine. The aim of this retrospective study was to investigate the different characteristics of the apparent diffusion coefficient (ADC) from DWI and fractional anisotropy (FA) from DTI of meningioma between dogs and cats. Statistical analyses were performed to compare ADC and FA values between the intratumoral or peritumoral regions and normal-appearing white matter (NAWM) among 13 dogs (13 lesions, but 12 each in ADC and FA analysis) and six cats (seven lesions). The NAWM of cats had a significantly lower ADC and higher FA compared to dogs. Therefore, for a comparison between dogs and cats, we used ADC and FA ratios that were calculated by dividing the subject (intra- or peritumoral) ADC and FA values by those of NAWM on the contralateral side. Regarding the intratumoral region, feline meningiomas showed a significantly lower ADC ratio and higher FA ratio than canine meningiomas. This study suggested that ADC and FA may be able to distinguish a meningioma that is solid and easy to detach, like as typical feline meningiomas.

Presurgical detection of brain invasion status in meningiomas based on first-order histogram based texture analysis of contrast enhanced imaging

OBJECTIVE

Invasion of brain parenchyma by meningioma can be a critical factor in surgical planning. The aim of this study was to determine the diagnostic utility of first-order texture parameters derived from both whole tumor and single largest slice of T1-contrast enhanced (T1-CE) images in differentiating meningiomas with and without brain invasion based on histopathology demonstration.

METHODS

T1-CE images of a total of 56 cases of grade II meningiomas with brain invasion (BI) and 52 meningiomas (37 grade I and 15 grade II) with no brain invasion (NBI) were analyzed. Filtration-based first-order histogram derived texture parameters were calculated both for whole tumor volume and largest axial cross-section. Random forest models were constructed both for whole tumor volume and largest axial cross-section individually and were assessed using a 5-fold cross validation with 100 repeats.

RESULTS

In detection of brain invasion, random forest model based on whole tumor segmentation had an AUC of 0.988 (95 % CI 0.976-1.00) with a cross validated value of 0.74 (95 % CI 0.45-0.96). For differentiation of grade I meningiomas from grade II meningiomas with brain invasion, the AUC was 0.999 (95 % CI 0.995-1.00) and 0.81 (95 % CI 0.61-0.99) in the training and validation cohorts, respectively. Similarly, when using only the single largest slice, the cross-validated AUC to distinguish BI versus NBI and BI versus grade I meningiomas was 0.67 (95 % CI 0.47, 0.92 and 0.78 (95 % CI 0.52, 0.95) respectively.

CONCLUSION

Radiomics based feature analysis applied on routine MRI post-contrast images may be helpful to predict presence of brain invasion in meningioma, possibly with better performance when comparing BI versus grade I meningiomas.

The Diagnostic Value of Radiomics-Based Machine Learning in Predicting the Grade of Meningiomas Using Conventional Magnetic Resonance Imaging: A Preliminary Study

Objective: The purpose of the current study is to investigate whether texture analysis-based machine learning algorithms could help devise a non-invasive imaging biomarker for accurate classification of meningiomas using machine learning algorithms.

Method: The study cohort was established from the hospital database by reviewing the medical records. Patients were selected if they underwent meningioma resection in the neurosurgery department between January 2015 and December 2018. A total number of 40 texture parameters were extracted from pretreatment postcontrast T1-weighted (T1C) images based on six matrixes. Three feature selection methods were adopted, namely, distance correlation, least absolute shrinkage and selection operator (LASSO), and gradient boosting decision tree (GBDT). Multiclass classification methods of linear discriminant analysis (LDA) and support vector machine (SVM) algorithms were employed to establish the classification models. The diagnostic performances of models were evaluated with confusion matrix based on which the areas under the curve, accuracy, and Kappa value of models were calculated.

Result: Confusion matrix showed that the LDA-based models represented better diagnostic performances than SVM-based models. The highest accuracy among LDA-based models was 75.6%, shown in the combination of Lasso + LDA. The optimal models for SVM-based models was Lasso+SVM, with accuracy of 59.0% in the testing group. One of the SVM-based models, GBDT+SVM, was overfitting, suggesting that this model was not suitable for application.

Conclusion: Machine learning algorithms with texture features extracted from T1C images could potentially serve as the assistant imaging biomarkers for presurgically grading meningiomas.

Preoperative and postoperative prediction of long-term meningioma outcomes

Background

Meningiomas are stratified according to tumor grade and extent of resection, often in isolation of other clinical variables. Here, we use machine learning (ML) to integrate demographic, clinical, radiographic and pathologic data to develop predictive models for meningioma outcomes.

Methods and findings

We developed a comprehensive database containing information from 235 patients who underwent surgery for 257 meningiomas at a single institution from 1990 to 2015. The median follow-up was 4.3 years, and resection specimens were re-evaluated according to current diagnostic criteria, revealing 128 WHO grade I, 104 grade II and 25 grade III meningiomas. A series of ML algorithms were trained and tuned by nested resampling to create models based on preoperative features, conventional postoperative features, or both. We compared different algorithms’ accuracy as well as the unique insights they offered into the data. Machine learning models restricted to preoperative information, such as patient demographics and radiographic features, had similar accuracy for predicting local failure (AUC = 0.74) or overall survival (AUC = 0.68) as models based on meningioma grade and extent of resection (AUC = 0.73 and AUC = 0.72, respectively). Integrated models incorporating all available demographic, clinical, radiographic and pathologic data provided the most accurate estimates (AUC = 0.78 and AUC = 0.74, respectively). From these models, we developed decision trees and nomograms to estimate the risks of local failure or overall survival for meningioma patients.

Conclusions

Clinical information has been historically underutilized in the prediction of meningioma outcomes. Predictive models trained on preoperative clinical data perform comparably to conventional models trained on meningioma grade and extent of resection. Combination of all available information can help stratify meningioma patients more accurately.

The diagnostic value of texture analysis in predicting WHO grades of meningiomas based on ADC maps: an attempt using decision tree and decision forest

OBJECTIVES

The preoperative prediction of the WHO grade of a meningioma is important for further treatment plans. This study aimed to assess whether texture analysis (TA) based on apparent diffusion coefficient (ADC) maps could non-invasively classify meningiomas accurately using tree classifiers.

METHODS

A pathology database was reviewed to identify meningioma patients who underwent tumour resection in our hospital with preoperative routine MRI scanning and diffusion-weighted imaging (DWI) between January 2011 and August 2017. A total of 152 meningioma patients with 421 preoperative ADC maps were included. Four categories of features, namely, clinical features, morphological features, average ADC values and texture features, were extracted. Three machine learning classifiers, namely, classic decision tree, conditional inference tree and decision forest, were built on these features from the training dataset. Then the performance of each classifier was evaluated and compared with the diagnosis made by two neuro-radiologists.

RESULTS

The ADC value alone was unable to distinguish three WHO grades of meningiomas. The machine learning classifiers based on clinical, morphological features and ADC value could achieve equivalent diagnostic performance (accuracy = 62.96%) compared to two experienced neuro-radiologists (accuracy = 61.11% and 62.04%). Upon analysis, the decision forest that was built with 23 selected texture features and the ADC value from the training dataset achieved the best diagnostic performance in the testing dataset (kappa = 0.64, accuracy = 79.51%).

CONCLUSIONS

Decision forest with the ADC value and ADC map-based texture features is a promising multiclass classifier that could potentially provide more precise diagnosis and aid diagnosis in the near future.

KEY POINTS

• A precise preoperative prediction of the WHO grade of a meningioma brings benefits to further treatment plans. • Machine learning models based on clinical, morphological features and ADC value could achieve equivalent diagnostic performance compared to experienced neuroradiologists. • The decision forest model built with 23 selected texture features and the ADC value achieved the best diagnostic performance (kappa = 0.64, accuracy = 79.51%).

The diagnostic value of using combined MR diffusion tensor imaging parameters to differentiate between low- and high-grade meningioma

OBJECTIVE

The purpose of this study was to examine whether the combined use of MR diffusion tensor imaging (DTI) parameters [DTI-apparent diffusion coefficient (ADC), fractional anisotropy (FA), axial diffusivity (AD), and radial diffusivity (RD)] could provide a more accurate diagnosis in differentiating between low-grade and atypical/anaplastic (high-grade) meningioma.

METHODS

Pathologically proven 45 meningioma patients [32 low-grade, 13 high-grade (11 atypical and 2 anaplastic)] who had received DTI before surgery were assessed retrospectively by 2 independent observers. For each lesion, MR DTI parameters (ADCmin, ADCmax, ADCmean, FA, AD, and RD) and ratios (rADCmin, rADCmax, rADCmean, rFA, rAD, and rRD) were calculated. When differentiating between low- and high-grade meningioma, the optimum cutoff values of all MR DTI parameters were determined by using receiver operating characteristic (ROC) analysis. Area under the curve (AUC) was measured with combined ROC analysis for different combinations of MR DTI parameters in order to identify the model combination with the best diagnostic accuracy in differentiation between low and high-grade meningioma.

RESULTS

Although the ADCmin, ADCmax, ADCmean, AD, RD, rADCmin, rADCmax, rADCmean, rAD, and rRD values of high-grade meningioma were significantly low (p = 0.007, p = 0.045, p = 0.035, p = 0.045, p = 0.003, p = 0.02, p = 0.03, p = 0.03, p = 0.045, and p = 0.01, respectively), when compared with low-grade meningioma, their FA and rFA values were significantly high (p = 0.007 and p = 0.01, respectively). For all MR DTI parameters, the highest individual distinctive power was RD with AUC of 0.778. The best diagnostic accuracy in differentiating between low- and high-grade meningioma was obtained by combining the ADCmin, RD, and FA parameters with 0.962 AUC.

CONCLUSION

This study shows that combined MR DTI parameters consisting of ADCmin, RD, and FA can differentiate high-grade from low-grade meningioma with a diagnostic accuracy of 96.2%. Advances in knowledge: To the best of our knowledge, this is the first study reporting that a combined use of all MR DTI parameters provides higher diagnostic accuracy for the differentiation of low- from high-grade meningioma. Our study shows that any of the model combinations was superior to use of any individual MR DTI parameters for differentiation between low and high-grade meningioma. A combination of ADCmin, RD, and FA was found to be the best model for differentiating low-grade from high-grade meningioma and sensitivity, specificity, and AUC values were found to be 92.3%, 100%, and 0.96, respectively. Thus, a combination of MR DTI parameters can provide more accurate diagnostic information when differentiation between low and high-grade meningioma.

Comparisons among MRI signs, apparent diffusion coefficient, and fractional anisotropy in dogs with a solitary intracranial meningioma or histiocytic sarcoma

Although MRI has become widely used in small animal practice, little is known about the validity of advanced MRI techniques such as diffusion-weighted imaging and diffusion tensor imaging. The aim of this retrospective analytical observational study was to investigate the characteristics of diffusion parameters, that is the apparent diffusion coefficient and fractional anisotropy, in dogs with a solitary intracranial meningioma or histiocytic sarcoma. Dogs were included based on the performance of diffusion MRI and histological confirmation. Statistical analyses were performed to compare apparent diffusion coefficient and fractional anisotropy for the two types of tumor in the intra- and peritumoral regions. Eleven cases with meningioma and six with histiocytic sarcoma satisfied the inclusion criteria. Significant differences in apparent diffusion coefficient value (× 10^-3 mm² /s) between meningioma vs. histiocytic sarcoma were recognized in intratumoral small (1.07 vs. 0.76) and large (1.04 vs. 0.77) regions of interest, in the peritumoral margin (0.93 vs. 1.08), and in the T2 high region (1.21 vs. 1.41). Significant differences in fractional anisotropy values were found in the peritumoral margin (0.29 vs. 0.24) and the T2 high region (0.24 vs. 0.17). The current study identified differences in measurements of apparent diffusion coefficient and fractional anisotropy for meningioma and histiocytic sarcoma in a small sample of dogs. In addition, we observed that all cases of intracranial histiocytic sarcoma showed leptomeningeal enhancement and/or mass formation invading into the sulci in the contrast study. Future studies are needed to determine the sensitivity of these imaging characteristics for differentiating between these tumor types.

Bayesian networks in neuroscience: a survey

Bayesian networks are a type of probabilistic graphical models lie at the intersection between statistics and machine learning. They have been shown to be powerful tools to encode dependence relationships among the variables of a domain under uncertainty. Thanks to their generality, Bayesian networks can accommodate continuous and discrete variables, as well as temporal processes. In this paper we review Bayesian networks and how they can be learned automatically from data by means of structure learning algorithms. Also, we examine how a user can take advantage of these networks for reasoning by exact or approximate inference algorithms that propagate the given evidence through the graphical structure. Despite their applicability in many fields, they have been little used in neuroscience, where they have focused on specific problems, like functional connectivity analysis from neuroimaging data. Here we survey key research in neuroscience where Bayesian networks have been used with different aims: discover associations between variables, perform probabilistic reasoning over the model, and classify new observations with and without supervision. The networks are learned from data of any kind-morphological, electrophysiological, -omics and neuroimaging-, thereby broadening the scope-molecular, cellular, structural, functional, cognitive and medical- of the brain aspects to be studied.

Resting state fMRI feature-based cerebral glioma grading by support vector machine

PURPOSE : Tumor grading plays an essential role in the optimal selection of solid tumor treatment. Noninvasive methods are needed for clinical grading of tumors. This study aimed to extract parameters of resting state blood oxygenation level-dependent functional magnetic resonance imaging (RS-fMRI) in the region of glioma and use the extracted features for tumor grading. METHODS : Tumor segmentation was performed with both conventional MRI and RS-fMRI. Four typical parameters, signal intensity difference ratio, signal intensity correlation (SIC), fractional amplitude of low-frequency fluctuation (fALFF) and regional homogeneity (ReHo), were defined to analyze tumor regions. Mann-Whitney [Formula: see text] test was employed to identify statistical difference of these four parameters between low-grade glioma (LGG) and high-grade glioma (HGG), respectively. Support vector machine (SVM) was employed to assess the diagnostic contributions of these parameters. RESULTS : Compared with LGG, HGG had more complex anatomical morphology and BOLD-fMRI features in the tumor region. SIC [Formula: see text], fALFF ([Formula: see text]) and ReHo ([Formula: see text]) were selected as features for classification according to the test [Formula: see text] value. The accuracy, sensitivity and specificity of SVM classification were better than 80, where SIC had the best classification accuracy (89). CONCLUSION : Parameters of RS-fMRI are effective to classify the tumor grade in glioma cases. The results indicate that this technique has clinical potential to serve as a complementary diagnostic tool.

Fast spectroscopic multiple analysis (FASMA) for brain tumor classification: a clinical decision support system utilizing multi-parametric 3T MR data

INTRODUCTION

A clinical decision support system (CDSS) for brain tumor classification can be used to assist in the diagnosis and grading of brain tumors. A Fast Spectroscopic Multiple Analysis (FASMA) system that uses combinations of multiparametric MRI data sets was developed as a CDSS for brain tumor classification.

METHODS

MRI metabolic ratios and spectra, from long and short TE, respectively, as well as diffusion and perfusion data were acquired from the intratumoral and peritumoral area of 126 patients with untreated intracranial tumors. These data were categorized based on the pathology, and different machine learning methods were evaluated regarding their classification performance for glioma grading and differentiation of infiltrating versus non-infiltrating lesions. Additional databases were embedded to the system, including updated literature values of the related MR parameters and typical tumor characteristics (imaging and histological), for further comparisons. Custom Graphical User Interface (GUI) layouts were developed to facilitate classification of the unknown cases based on the user's available MR data.

RESULTS

The highest classification performance was achieved with a support vector machine (SVM) using the combination of all MR features. FASMA correctly classified 89 and 79% in the intratumoral and peritumoral area, respectively, for cases from an independent test set. FASMA produced the correct diagnosis, even in the misclassified cases, since discrimination between infiltrative versus non-infiltrative cases was possible.

CONCLUSIONS

FASMA is a prototype CDSS, which integrates complex quantitative MR data for brain tumor characterization. FASMA was developed as a diagnostic assistant that provides fast analysis, representation and classification for a set of MR parameters. This software may serve as a teaching tool on advanced MRI techniques, as it incorporates additional information regarding typical tumor characteristics derived from the literature.