Support vector machines in DSC-based glioma imaging: suggestions for optimal characterization

Normalizing Flow-Based Distribution Estimation of Pharmacokinetic Parameters in Dynamic Contrast-Enhanced Magnetic Resonance Imaging

OBJECTIVE

The pharmacokinetic (PK) parameters estimated from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) provide valuable information for clinical research and diagnosis. However, these estimated PK parameters suffer from many sources of variability. Thus, the estimation of the posterior distributions of these PK parameters could provide a way to simultaneously quantify the values and uncertainties of the PK parameters. Our objective is to develop an efficient and flexible method to more closely approximate and estimate the underlying posterior distributions of the PK parameters.

METHODS

The normalizing flow model-based parameters distribution estimation neural network (FPDEN) is proposed to adaptively learn and estimate the posterior distributions of the PK parameters. The maximum likelihood estimation (MLE) loss is directly constructed based on the parameter distributions learned by the normalizing flow model, rather than pre-defined distributions.

RESULTS

Experimental analysis shows that the proposed method can improve parameter estimation accuracy. Moreover, the uncertainty derived from the parameter distribution constitutes an effective indicator to exclude unreliable parametric results. A successful demonstration is the improved classification performance of the glioma World Health Organization (WHO) grading task, specifically in terms of distinguishing between low and high grades, as well as between Grade III and Grade IV.

CONCLUSION

The FPDEN method offers improved accuracy for estimation of PK parameters and boosts the performance of the glioma grading task.

SIGNIFICANCE

By enhancing the precision and reliability of DCE-MRI, the proposed method promotes its further applications in clinical practice.

SASG-GCN: Self-Attention Similarity Guided Graph Convolutional Network for Multi-Type Lower-Grade Glioma Classification

Identifying the subtypes of low-grade glioma (LGG) can help prevent brain tumor progression and patient death. However, the complicated non-linear relationship and high dimensionality of 3D brain MRI limit the performance of machine learning methods. Therefore, it is important to develop a classification method that can overcome these limitations. This study proposes a self-attention similarity-guided graph convolutional network (SASG-GCN) that uses the constructed graphs to complete multi-classification (tumor-free (TF), WG, and TMG). In the pipeline of SASG-GCN, we use a convolutional deep belief network and a self-attention similarity-based method to construct the vertices and edges of the constructed graphs at 3D MRI level, respectively. The multi-classification experiment is performed in a two-layer GCN model. SASG-GCN is trained and evaluated on 402 3D MRI images which are produced from the TCGA-LGG dataset. Empirical tests demonstrate that SASG-GCN accurately classifies the subtypes of LGG. The accuracy of SASG-GCN achieves 93.62%, outperforming several other state-of-the-art classification methods. In-depth discussion and analysis reveal that the self-attention similarity-guided strategy improves the performance of SASG-GCN. The visualization revealed differences between different gliomas.

AGGN: Attention-based glioma grading network with multi-scale feature extraction and multi-modal information fusion

In this paper, a magnetic resonance imaging (MRI) oriented novel attention-based glioma grading network (AGGN) is proposed. By applying the dual-domain attention mechanism, both channel and spatial information can be considered to assign weights, which benefits highlighting the key modalities and locations in the feature maps. Multi-branch convolution and pooling operations are applied in a multi-scale feature extraction module to separately obtain shallow and deep features on each modality, and a multi-modal information fusion module is adopted to sufficiently merge low-level detailed and high-level semantic features, which promotes the synergistic interaction among different modality information. The proposed AGGN is comprehensively evaluated through extensive experiments, and the results have demonstrated the effectiveness and superiority of the proposed AGGN in comparison to other advanced models, which also presents high generalization ability and strong robustness. In addition, even without the manually labeled tumor masks, AGGN can present considerable performance as other state-of-the-art algorithms, which alleviates the excessive reliance on supervised information in the end-to-end learning paradigm.

Overview of Multi-Modal Brain Tumor MR Image Segmentation

The precise segmentation of brain tumor images is a vital step towards accurate diagnosis and effective treatment of brain tumors. Magnetic Resonance Imaging (MRI) can generate brain images without tissue damage or skull artifacts, providing important discriminant information for clinicians in the study of brain tumors and other brain diseases. In this paper, we survey the field of brain tumor MRI images segmentation. Firstly, we present the commonly used databases. Then, we summarize multi-modal brain tumor MRI image segmentation methods, which are divided into three categories: conventional segmentation methods, segmentation methods based on classical machine learning methods, and segmentation methods based on deep learning methods. The principles, structures, advantages and disadvantages of typical algorithms in each method are summarized. Finally, we analyze the challenges, and suggest a prospect for future development trends.

Machine learning applications to neuroimaging for glioma detection and classification: An artificial intelligence augmented systematic review

Glioma is the most common primary intraparenchymal tumor of the brain and the 5-year survival rate of high-grade glioma is poor. Magnetic resonance imaging (MRI) is essential for detecting, characterizing and monitoring brain tumors but definitive diagnosis still relies on surgical pathology. Machine learning has been applied to the analysis of MRI data in glioma research and has the potential to change clinical practice and improve patient outcomes. This systematic review synthesizes and analyzes the current state of machine learning applications to glioma MRI data and explores the use of machine learning for systematic review automation. Various datapoints were extracted from the 153 studies that met inclusion criteria and analyzed. Natural language processing (NLP) analysis involved keyword extraction, topic modeling and document classification. Machine learning has been applied to tumor grading and diagnosis, tumor segmentation, non-invasive genomic biomarker identification, detection of progression and patient survival prediction. Model performance was generally strong (AUC = 0.87 ± 0.09; sensitivity = 0.87 ± 0.10; specificity = 0.0.86 ± 0.10; precision = 0.88 ± 0.11). Convolutional neural network, support vector machine and random forest algorithms were top performers. Deep learning document classifiers yielded acceptable performance (mean 5-fold cross-validation AUC = 0.71). Machine learning tools and data resources were synthesized and summarized to facilitate future research. Machine learning has been widely applied to the processing of MRI data in glioma research and has demonstrated substantial utility. NLP and transfer learning resources enabled the successful development of a replicable method for automating the systematic review article screening process, which has potential for shortening the time from discovery to clinical application in medicine.

Machine learning assisted DSC-MRI radiomics as a tool for glioma classification by grade and mutation status

BACKGROUND

Combining MRI techniques with machine learning methodology is rapidly gaining attention as a promising method for staging of brain gliomas. This study assesses the diagnostic value of such a framework applied to dynamic susceptibility contrast (DSC)-MRI in classifying treatment-naïve gliomas from a multi-center patients into WHO grades II-IV and across their isocitrate dehydrogenase (IDH) mutation status.

METHODS

Three hundred thirty-three patients from 6 tertiary centres, diagnosed histologically and molecularly with primary gliomas (IDH-mutant = 151 or IDH-wildtype = 182) were retrospectively identified. Raw DSC-MRI data was post-processed for normalised leakage-corrected relative cerebral blood volume (rCBV) maps. Shape, intensity distribution (histogram) and rotational invariant Haralick texture features over the tumour mask were extracted. Differences in extracted features across glioma grades and mutation status were tested using the Wilcoxon two-sample test. A random-forest algorithm was employed (2-fold cross-validation, 250 repeats) to predict grades or mutation status using the extracted features.

RESULTS

Shape, distribution and texture features showed significant differences across mutation status. WHO grade II-III differentiation was mostly driven by shape features while texture and intensity feature were more relevant for the III-IV separation. Increased number of features became significant when differentiating grades further apart from one another. Gliomas were correctly stratified by mutation status in 71% and by grade in 53% of the cases (87% of the gliomas grades predicted with distance less than 1).

CONCLUSIONS

Despite large heterogeneity in the multi-center dataset, machine learning assisted DSC-MRI radiomics hold potential to address the inherent variability and presents a promising approach for non-invasive glioma molecular subtyping and grading.

Using an artificial neural network for fast mapping of the oxygen extraction fraction with combined QSM and quantitative BOLD

PURPOSE

To apply an artificial neural network (ANN) for fast and robust quantification of the oxygen extraction fraction (OEF) from a combined QSM and quantitative BOLD analysis of gradient echo data and to compare the ANN to a traditional quasi-Newton (QN) method for numerical optimization.

METHODS

Random combinations of OEF, deoxygenated blood volume ( ν ), R₂ , and nonblood magnetic susceptibility ( χ nb ) with each parameter following a Gaussian distribution that represented physiological gray matter and white matter values were used to simulate quantitative BOLD signals and QSM values. An ANN was trained with the simulated data with added Gaussian noise. The ANN was applied to multigradient echo brain data of 7 healthy subjects, and the reconstructed parameters and maps were compared to QN results using Student t test and Bland-Altman analysis.

RESULTS

Intersubject means and SDs of gray matter were OEF = 43.5 ± 0.8 %, R 2 = 13.5 ± 0.3 Hz, ν = 3.4 ± 0.1 %, χ nb = - 25 ± 5 ppb for ANN; and OEF = 43.8 ± 5.2 %, R 2 = 12.2 ± 0.8 Hz, ν = 4.2 ± 0.6 %, χ nb = - 39 ± 7 ppb for QN, with a significant difference ( P < 0.05 ) for R 2 , ν , and χ nb . For white matter, they were OEF = 47.5 ± 1.1 %, R 2 = 17.1 ± 0.4 Hz, ν = 2.5 ± 0.2 %, χ nb = - 38 ± 5 ppb for ANN; and OEF = 42.3 ± 5.6 %, R 2 = 16.7 ± 0.7 Hz, ν = 2.9 ± 0.3 %, χ nb = - 45 ± 9 ppb for QN, with a significant difference ( P < 0.05 ) for OEF and ν . ANN revealed more gray-white matter contrast but less intersubject variation in OEF than QN. In contrast to QN, the ANN reconstruction did not need an additional sequence for parameter initialization and took approximately 1 s rather than roughly 1 h.

CONCLUSION

ANNs allow faster and, with regard to initialization, more robust reconstruction of OEF maps with lower intersubject variation than QN approaches.

A quantitative SVM approach potentially improves the accuracy of magnetic resonance spectroscopy in the preoperative evaluation of the grades of diffuse gliomas

OBJECTIVES

To investigate the association between proton magnetic resonance spectroscopy (1H-MRS) metabolic features and the grade of gliomas, and to establish a machine-learning model to predict the glioma grade.

METHODS

This study included 112 glioma patients who were divided into the training (n = 74) and validation (n = 38) sets based on the time of hospitalization. Twenty-six metabolic features were extracted from the preoperative 1H-MRS image. The Student's t-test was conducted to screen for differentially expressed features between low- and high-grade gliomas (WHO grades II and III/IV, respectively). Next, the minimum Redundancy Maximum Relevance (mRMR) algorithm was performed to further select features for a support vector machine (SVM) classifier building. Performance of the predictive model was evaluated both in the training and validation sets using ROC curve analysis.

RESULTS

Among the extracted 1H-MRS metabolic features, thirteen features were differentially expressed. Four features were further selected as grade-predictive imaging signatures using the mRMR algorithm. The predictive performance of the machine-learning model measured by the AUC was 0.825 and 0.820 in the training and validation sets, respectively. This was better than the predictive performances of individual metabolic features, the best of which was 0.812.

CONCLUSIONS

1H-MRS metabolic features could help in predicting the grade of gliomas. The machine-learning model achieved a better prediction performance in grading gliomas than individual features, indicating that it could complement the traditionally used metabolic features.

Machine-learning in grading of gliomas based on multi-parametric magnetic resonance imaging at 3T

OBJECTIVE

The objective of this study was to assess the contribution of multi-parametric (mp) magnetic resonance imaging (MRI) quantitative features in the machine learning-based grading of gliomas with a multi-region-of-interests approach.

MATERIALS AND METHODS

Forty-three patients who were newly diagnosed as having a glioma were included in this study. The patients were scanned prior to any therapy using a standard brain tumor magnetic resonance (MR) imaging protocol that included T1 and T2-weighted, diffusion-weighted, diffusion tensor, MR perfusion and MR spectroscopic imaging. Three different regions-of-interest were drawn for each subject to encompass tumor, immediate tumor periphery, and distant peritumoral edema/normal. The normalized mp-MRI features were used to build machine-learning models for differentiating low-grade gliomas (WHO grades I and II) from high grades (WHO grades III and IV). In order to assess the contribution of regional mp-MRI quantitative features to the classification models, a support vector machine-based recursive feature elimination method was applied prior to classification.

RESULTS

A machine-learning model based on support vector machine algorithm with linear kernel achieved an accuracy of 93.0%, a specificity of 86.7%, and a sensitivity of 96.4% for the grading of gliomas using ten-fold cross validation based on the proposed subset of the mp-MRI features.

CONCLUSION

In this study, machine-learning based on multiregional and multi-parametric MRI data has proven to be an important tool in grading glial tumors accurately even in this limited patient population. Future studies are needed to investigate the use of machine learning algorithms for brain tumor classification in a larger patient cohort.

Diffusion parameter mapping with the combined intravoxel incoherent motion and kurtosis model using artificial neural networks at 3 T

Artificial neural networks (ANNs) were used for voxel-wise parameter estimation with the combined intravoxel incoherent motion (IVIM) and kurtosis model facilitating robust diffusion parameter mapping in the human brain. The proposed ANN approach was compared with conventional least-squares regression (LSR) and state-of-the-art multi-step fitting (LSR-MS) in Monte-Carlo simulations and in vivo in terms of estimation accuracy and precision, number of outliers and sensitivity in the distinction between grey (GM) and white (WM) matter. Both the proposed ANN approach and LSR-MS yielded visually increased parameter map quality. Estimations of all parameters (perfusion fraction f, diffusion coefficient D, pseudo-diffusion coefficient D*, kurtosis K) were in good agreement with the literature using ANN, whereas LSR-MS resulted in D* overestimation and LSR yielded increased values for f and D*, as well as decreased values for K. Using ANN, outliers were reduced for the parameters f (ANN, 1%; LSR-MS, 19%; LSR, 8%), D* (ANN, 21%; LSR-MS, 25%; LSR, 23%) and K (ANN, 0%; LSR-MS, 0%; LSR, 15%). Moreover, ANN enabled significant distinction between GM and WM based on all parameters, whereas LSR facilitated this distinction only based on D and LSR-MS on f, D and K. Overall, the proposed ANN approach was found to be superior to conventional LSR, posing a powerful alternative to the state-of-the-art method LSR-MS with several advantages in the estimation of IVIM-kurtosis parameters, which might facilitate increased applicability of enhanced diffusion models at clinical scan times.

Multi-class texture analysis in colorectal cancer histology

Automatic recognition of different tissue types in histological images is an essential part in the digital pathology toolbox. Texture analysis is commonly used to address this problem; mainly in the context of estimating the tumour/stroma ratio on histological samples. However, although histological images typically contain more than two tissue types, only few studies have addressed the multi-class problem. For colorectal cancer, one of the most prevalent tumour types, there are in fact no published results on multiclass texture separation. In this paper we present a new dataset of 5,000 histological images of human colorectal cancer including eight different types of tissue. We used this set to assess the classification performance of a wide range of texture descriptors and classifiers. As a result, we found an optimal classification strategy that markedly outperformed traditional methods, improving the state of the art for tumour-stroma separation from 96.9% to 98.6% accuracy and setting a new standard for multiclass tissue separation (87.4% accuracy for eight classes). We make our dataset of histological images publicly available under a Creative Commons license and encourage other researchers to use it as a benchmark for their studies.

Classification of tumor area using combined DCE and DSC MRI in patients with glioblastoma

This study proposes an automatic method for identification and quantification of different tissue components: the non-enhanced infiltrative tumor, vasogenic edema and enhanced tumor areas, at the subject level, in patients with glioblastoma (GB) based on dynamic contrast enhancement (DCE) and dynamic susceptibility contrast (DSC) MRI. Nineteen MR data sets, obtained from 12 patients with GB, were included. Seven patients were scanned before and 8 weeks following bevacizumab initiation. Segmentation of the tumor area was performed based on the temporal data of DCE and DSC at the group-level using k-means algorithm, and further at the subject-level using support vector machines algorithm. The obtained components were associated to different tissues types based on their temporal characteristics, calculated perfusion and permeability values and MR-spectroscopy. The method enabled the segmentation of the tumor area into the enhancing permeable component; the non-enhancing hypoperfused component, associated with vasogenic edema; and the non-enhancing hyperperfused component, associated with infiltrative tumor. Good agreement was obtained between the group-level, unsupervised and subject-level, supervised classification results, with significant correlation (r = 0.93, p < 0.001) and average symmetric root-mean-square surface distance of 2.5 ± 5.1 mm. Longitudinal changes in the volumes of the three components were assessed alongside therapy. Tumor area segmentation using DCE and DSC can be used to differentiate between vasogenic edema and infiltrative tumors in patients with GB, which is of major clinical importance in therapy response assessment.

Functional imaging of acute kidney injury at 3 Tesla: investigating multiple parameters using DCE-MRI and a two-compartment filtration model

OBJECT

To investigate how MR-based parameters reflect functional changes in kidneys with acute kidney injury (AKI) using dynamic contrast enhanced MRI and a two-compartment renal filtration model.

MATERIALS AND METHODS

MRI data of eight male Lewis rats were analyzed retrospectively. Five animals were subjected to AKI, three native rats served as control. All animals underwent perfusion imaging by dynamic contrast-enhanced MRI. Renal blood volume, glomerular filtration rate (GFR) as well as plasma and tubular mean transit times were estimated from regions-of-interest drawn in the renal cortex. Differences between healthy kidneys and kidneys subjected to AKI were analyzed using a paired t-test.

RESULTS

Significant differences between ischemic and healthy kidneys could only be detected for the glomerular filtration rate. For all other calculated parameters, differences were present, however not significant. In rats with AKI, average single kidney GFR was 0.66 ± 0.37 ml/min for contralateral and 0.26 ± 0.12 ml/ min for diseased kidneys (P = 0.0254). For the healthy control group, the average GFR was 0.39 ± 0.06 ml/min and 0.41 ± 0.11 ml/min, respectively. Differences between diseased kidneys of AKI rats and ipsilateral kidneys of the healthy control group were significant (P = 0.0381).

CONCLUSION

Significant differences of functional parameters reflecting damage of the renal tissue of kidneys with AKI compared to the contralateral, healthy kidneys could only be detected by GFR. GFR might be a useful parameter that allows for a spatially resolved detection of abnormal changes of renal tissue by AKI.

Molecular and metabolic pattern classification for detection of brain glioma progression

OBJECTIVES

The ability to differentiate between brain tumor progression and radiation therapy induced necrosis is critical for appropriate patient management. In order to improve the differential diagnosis, we combined fluorine-18 2-fluoro-deoxyglucose positron emission tomography ((18)F-FDG PET), proton magnetic resonance spectroscopy ((1)H MRS) and histological data to develop a multi-parametric machine-learning model.

METHODS

We enrolled twelve post-therapy patients with grade 2 and 3 gliomas that were suspicious of tumor progression. All patients underwent (18)F-FDG PET and (1)H MRS. Maximal standardized uptake value (SUVmax) of the tumors and reference regions were obtained. Multiple 2D maps of choline (Cho), creatine (Cr), and N-acetylaspartate (NAA) of the tumors were generated. A support vector machine (SVM) learning model was established to take imaging biomarkers and histological data as input vectors. A combination of clinical follow-up and multiple sequential MRI studies served as the basis for assessing the clinical outcome. All vector combinations were evaluated for diagnostic accuracy and cross validation. The optimal cutoff value of individual parameters was calculated using Receiver operating characteristic (ROC) plots.

RESULTS

The SVM and ROC analyses both demonstrated that SUVmax of the lesion was the most significant single diagnostic parameter (75% accuracy) followed by Cho concentration (67% accuracy). SVM analysis of all paired parameters showed SUVmax and Cho concentration in combination could achieve 83% accuracy. SUVmax of the lesion paired with SUVmax of the white matter as well as the tumor Cho paired with the tumor Cr both showed 83% accuracy. These were the most significant paired diagnostic parameters of either modality. Combining all four parameters did not improve the results. However, addition of two more parameters, Cho and Cr of brain parenchyma contralateral to the tumor, increased the accuracy to 92%.

CONCLUSION

This study suggests that SVM models may improve detection of glioma progression more accurately than single parametric imaging methods.

RESEARCH SUPPORT

National Cancer Institute, Cancer Center Support Grant Supplement Award, Imaging Response Assessment Teams.

A survey of MRI-based medical image analysis for brain tumor studies

MRI-based medical image analysis for brain tumor studies is gaining attention in recent times due to an increased need for efficient and objective evaluation of large amounts of data. While the pioneering approaches applying automated methods for the analysis of brain tumor images date back almost two decades, the current methods are becoming more mature and coming closer to routine clinical application. This review aims to provide a comprehensive overview by giving a brief introduction to brain tumors and imaging of brain tumors first. Then, we review the state of the art in segmentation, registration and modeling related to tumor-bearing brain images with a focus on gliomas. The objective in the segmentation is outlining the tumor including its sub-compartments and surrounding tissues, while the main challenge in registration and modeling is the handling of morphological changes caused by the tumor. The qualities of different approaches are discussed with a focus on methods that can be applied on standard clinical imaging protocols. Finally, a critical assessment of the current state is performed and future developments and trends are addressed, giving special attention to recent developments in radiological tumor assessment guidelines.

SVM-based glioma grading: Optimization by feature reduction analysis

We investigated the predictive power of feature reduction analysis approaches in support vector machine (SVM)-based classification of glioma grade. In 101 untreated glioma patients, three analytic approaches were evaluated to derive an optimal reduction in features; (i) Pearson's correlation coefficients (PCC), (ii) principal component analysis (PCA) and (iii) independent component analysis (ICA). Tumor grading was performed using a previously reported SVM approach including whole-tumor cerebral blood volume (CBV) histograms and patient age. Best classification accuracy was found using PCA at 85% (sensitivity=89%, specificity=84%) when reducing the feature vector from 101 (100-bins rCBV histogram+age) to 3 principal components. In comparison, classification accuracy by PCC was 82% (89%, 77%, 2 dimensions) and 79% by ICA (87%, 75%, 9 dimensions). For improved speed (up to 30%) and simplicity, feature reduction by all three methods provided similar classification accuracy to literature values (∼87%) while reducing the number of features by up to 98%.

Data mining methods in the prediction of Dementia: A real-data comparison of the accuracy, sensitivity and specificity of linear discriminant analysis, logistic regression, neural networks, support vector machines, classification trees and random forests

Background

Dementia and cognitive impairment associated with aging are a major medical and social concern. Neuropsychological testing is a key element in the diagnostic procedures of Mild Cognitive Impairment (MCI), but has presently a limited value in the prediction of progression to dementia. We advance the hypothesis that newer statistical classification methods derived from data mining and machine learning methods like Neural Networks, Support Vector Machines and Random Forests can improve accuracy, sensitivity and specificity of predictions obtained from neuropsychological testing. Seven non parametric classifiers derived from data mining methods (Multilayer Perceptrons Neural Networks, Radial Basis Function Neural Networks, Support Vector Machines, CART, CHAID and QUEST Classification Trees and Random Forests) were compared to three traditional classifiers (Linear Discriminant Analysis, Quadratic Discriminant Analysis and Logistic Regression) in terms of overall classification accuracy, specificity, sensitivity, Area under the ROC curve and Press'Q. Model predictors were 10 neuropsychological tests currently used in the diagnosis of dementia. Statistical distributions of classification parameters obtained from a 5-fold cross-validation were compared using the Friedman's nonparametric test.

Results

Press' Q test showed that all classifiers performed better than chance alone (p < 0.05). Support Vector Machines showed the larger overall classification accuracy (Median (Me) = 0.76) an area under the ROC (Me = 0.90). However this method showed high specificity (Me = 1.0) but low sensitivity (Me = 0.3). Random Forest ranked second in overall accuracy (Me = 0.73) with high area under the ROC (Me = 0.73) specificity (Me = 0.73) and sensitivity (Me = 0.64). Linear Discriminant Analysis also showed acceptable overall accuracy (Me = 0.66), with acceptable area under the ROC (Me = 0.72) specificity (Me = 0.66) and sensitivity (Me = 0.64). The remaining classifiers showed overall classification accuracy above a median value of 0.63, but for most sensitivity was around or even lower than a median value of 0.5.

Conclusions

When taking into account sensitivity, specificity and overall classification accuracy Random Forests and Linear Discriminant analysis rank first among all the classifiers tested in prediction of dementia using several neuropsychological tests. These methods may be used to improve accuracy, sensitivity and specificity of Dementia predictions from neuropsychological testing.