Model-free functional MRI analysis using Kohonen clustering neural network and fuzzy C-means

A novel biomarker selection method using multimodal neuroimaging data

Identifying biomarkers is essential to obtain the optimal therapeutic benefit while treating patients with late-life depression (LLD). We compare LLD patients with healthy controls (HC) using resting-state functional magnetic resonance and diffusion tensor imaging data to identify neuroimaging biomarkers that may be potentially associated with the underlying pathophysiology of LLD. We implement a Bayesian multimodal local false discovery rate approach for functional connectivity, borrowing strength from structural connectivity to identify disrupted functional connectivity of LLD compared to HC. In the Bayesian framework, we develop an algorithm to control the overall false discovery rate of our findings. We compare our findings with the literature and show that our approach can better detect some regions never discovered before for LLD patients. The Hub of our discovery related to various neurobehavioral disorders can be used to develop behavioral interventions to treat LLD patients who do not respond to antidepressants.

From single-neuron dynamics to higher-order circuit motifs in control and pathological brain networks

The convergence of advanced single-cell in vivo functional imaging techniques, computational modelling tools and graph-based network analytics has heralded new opportunities to study single-cell dynamics across large-scale networks, providing novel insights into principles of brain communication and pointing towards potential new strategies for treating neurological disorders. A major recent finding has been the identification of unusually richly connected hub cells that have capacity to synchronize networks and may also be critical in network dysfunction. While hub neurons are traditionally defined by measures that consider solely the number and strength of connections, novel higher-order graph analytics now enables the mining of massive networks for repeating subgraph patterns called motifs. As an illustration of the power offered by higher-order analysis of neuronal networks, we highlight how recent methodological advances uncovered a new functional cell type, the superhub, that is predicted to play a major role in regulating network dynamics. Finally, we discuss open questions that will be critical for assessing the importance of higher-order cellular-scale network analytics in understanding brain function in health and disease.

Model-Based and Model-Free Techniques for Amyotrophic Lateral Sclerosis Diagnostic Prediction and Patient Clustering

Amyotrophic lateral sclerosis (ALS) is a complex progressive neurodegenerative disorder with an estimated prevalence of about 5 per 100,000 people in the United States. In this study, the ALS disease progression is measured by the change of Amyotrophic Lateral Sclerosis Functional Rating Scale (ALSFRS) score over time. The study aims to provide clinical decision support for timely forecasting of the ALS trajectory as well as accurate and reproducible computable phenotypic clustering of participants. Patient data are extracted from DREAM-Phil Bowen ALS Prediction Prize4Life Challenge data, most of which are from the Pooled Resource Open-Access ALS Clinical Trials Database (PRO-ACT) archive. We employed model-based and model-free machine-learning methods to predict the change of the ALSFRS score over time. Using training and testing data we quantified and compared the performance of different techniques. We also used unsupervised machine learning methods to cluster the patients into separate computable phenotypes and interpret the derived subcohorts. Direct prediction of univariate clinical outcomes based on model-based (linear models) or model-free (machine learning based techniques - random forest and Bayesian adaptive regression trees) was only moderately successful. The correlation coefficients between clinically observed changes in ALSFRS scores relative to the model-based/model-free predicted counterparts were 0.427 (random forest) and 0.545(BART). The reliability of these results were assessed using internal statistical cross validation and well as external data validation. Unsupervised clustering generated very reliable and consistent partitions of the patient cohort into four computable phenotypic subgroups. These clusters were explicated by identifying specific salient clinical features included in the PRO-ACT archive that discriminate between the derived subcohorts. There are differences between alternative analytical methods in forecasting specific clinical phenotypes. Although predicting univariate clinical outcomes may be challenging, our results suggest that modern data science strategies are useful in clustering patients and generating evidence-based ALS hypotheses about complex interactions of multivariate factors. Predicting univariate clinical outcomes using the PRO-ACT data yields only marginal accuracy (about 70%). However, unsupervised clustering of participants into sub-groups generates stable, reliable and consistent (exceeding 95%) computable phenotypes whose explication requires interpretation of multivariate sets of features. HIGHLIGHTS: • Used a large ALS data archive of 8,000 patients consisting of 3 million records, including 200 clinical features tracked over 12 months. • Employed model-based and model-free methods to predict ALSFRS changes over time, cluster patients into cohorts, and derive computable phenotypes. • Research findings include stable, reliable, and consistent (95%) patient stratification into computable phenotypes. However, clinical explication of the results requires interpretation of multivariate information. Graphical Abstract ᅟ.

Application of Graph Theory for Identifying Connectivity Patterns in Human Brain Networks: A Systematic Review

Background: Analysis of the human connectome using functional magnetic resonance imaging (fMRI) started in the mid-1990s and attracted increasing attention in attempts to discover the neural underpinnings of human cognition and neurological disorders. In general, brain connectivity patterns from fMRI data are classified as statistical dependencies (functional connectivity) or causal interactions (effective connectivity) among various neural units. Computational methods, especially graph theory-based methods, have recently played a significant role in understanding brain connectivity architecture. Objectives: Thanks to the emergence of graph theoretical analysis, the main purpose of the current paper is to systematically review how brain properties can emerge through the interactions of distinct neuronal units in various cognitive and neurological applications using fMRI. Moreover, this article provides an overview of the existing functional and effective connectivity methods used to construct the brain network, along with their advantages and pitfalls. Methods: In this systematic review, the databases Science Direct, Scopus, arXiv, Google Scholar, IEEE Xplore, PsycINFO, PubMed, and SpringerLink are employed for exploring the evolution of computational methods in human brain connectivity from 1990 to the present, focusing on graph theory. The Cochrane Collaboration's tool was used to assess the risk of bias in individual studies. Results: Our results show that graph theory and its implications in cognitive neuroscience have attracted the attention of researchers since 2009 (as the Human Connectome Project launched), because of their prominent capability in characterizing the behavior of complex brain systems. Although graph theoretical approach can be generally applied to either functional or effective connectivity patterns during rest or task performance, to date, most articles have focused on the resting-state functional connectivity. Conclusions: This review provides an insight into how to utilize graph theoretical measures to make neurobiological inferences regarding the mechanisms underlying human cognition and behavior as well as different brain disorders.

Application of Graph Theory for Identifying Connectivity Patterns in Human Brain Networks: A Systematic Review

Background: Analysis of the human connectome using functional magnetic resonance imaging (fMRI) started in the mid-1990s and attracted increasing attention in attempts to discover the neural underpinnings of human cognition and neurological disorders. In general, brain connectivity patterns from fMRI data are classified as statistical dependencies (functional connectivity) or causal interactions (effective connectivity) among various neural units. Computational methods, especially graph theory-based methods, have recently played a significant role in understanding brain connectivity architecture. Objectives: Thanks to the emergence of graph theoretical analysis, the main purpose of the current paper is to systematically review how brain properties can emerge through the interactions of distinct neuronal units in various cognitive and neurological applications using fMRI. Moreover, this article provides an overview of the existing functional and effective connectivity methods used to construct the brain network, along with their advantages and pitfalls. Methods: In this systematic review, the databases Science Direct, Scopus, arXiv, Google Scholar, IEEE Xplore, PsycINFO, PubMed, and SpringerLink are employed for exploring the evolution of computational methods in human brain connectivity from 1990 to the present, focusing on graph theory. The Cochrane Collaboration's tool was used to assess the risk of bias in individual studies. Results: Our results show that graph theory and its implications in cognitive neuroscience have attracted the attention of researchers since 2009 (as the Human Connectome Project launched), because of their prominent capability in characterizing the behavior of complex brain systems. Although graph theoretical approach can be generally applied to either functional or effective connectivity patterns during rest or task performance, to date, most articles have focused on the resting-state functional connectivity. Conclusions: This review provides an insight into how to utilize graph theoretical measures to make neurobiological inferences regarding the mechanisms underlying human cognition and behavior as well as different brain disorders.

Iterative spatial fuzzy clustering for 3D brain magnetic resonance image supervoxel segmentation

BACKGROUND

Although supervoxel segmentation methods have been employed for brain Magnetic Resonance Image (MRI) processing and analysis, due to the specific features of the brain, including complex-shaped internal structures and partial volume effect, their performance remains unsatisfactory.

NEW METHODS

To address these issues, this paper presents a novel iterative spatial fuzzy clustering (ISFC) algorithm to generate 3D supervoxels for brain MRI volume based on prior knowledge. This work makes use of the common topology among the human brains to obtain a set of seed templates from a population-based brain template MRI image. After selecting the number of supervoxels, the corresponding seed template is projected onto the considered individual brain for generating reliable seeds. Then, to deal with the influence of partial volume effect, an efficient iterative spatial fuzzy clustering algorithm is proposed to allocate voxels to the seeds and to generate the supervoxels for the overall brain MRI volume.

RESULTS

The performance of the proposed algorithm is evaluated on two widely used public brain MRI datasets and compared with three other up-to-date methods.

CONCLUSIONS

The proposed algorithm can be utilized for several brain MRI processing and analysis, including tissue segmentation, tumor detection and segmentation, functional parcellation and registration.

Clustering of fMRI data: the elusive optimal number of clusters

Model-free methods are widely used for the processing of brain fMRI data collected under natural stimulations, sleep, or rest. Among them is the popular fuzzy c-mean algorithm, commonly combined with cluster validity (CV) indices to identify the ‘true’ number of clusters (components), in an unsupervised way. CV indices may however reveal different optimal c-partitions for the same fMRI data, and their effectiveness can be hindered by the high data dimensionality, the limited signal-to-noise ratio, the small proportion of relevant voxels, and the presence of artefacts or outliers. Here, the author investigated the behaviour of seven robust CV indices. A new CV index that incorporates both compactness and separation measures is also introduced. Using both artificial and real fMRI data, the findings highlight the importance of looking at the behavior of different compactness and separation measures, defined here as building blocks of CV indices, to depict a full description of the data structure, in particular when no agreement is found between CV indices. Overall, for fMRI, it makes sense to relax the assumption that only one unique c-partition exists, and appreciate that different c-partitions (with different optimal numbers of clusters) can be useful explanations of the data, given the hierarchical organization of many brain networks.

A predictive approach to benthic marine habitat mapping: Efficacy and management implications

The availability of marine habitats maps remains limited due to difficulty and cost of working at sea. Reduced light penetration in the water hampers the use of optical imagery, and acoustic methods require extensive sea-truth activities. Predictive spatial modelling may offer an alternative to produce benthic habitat maps based on complete acoustic coverage of the seafloor together with a comparatively low number of sea truths. This approach was applied to the coralligenous reefs of the Marine Protected Area of Tavolara - Punta Coda Cavallo (NE Sardinia, Italy). Fuzzy clustering, applied to a set of observations made by scuba diving and used as sea truth, allowed recognising five coralligenous habitats, all but one existing within EUNIS (European Nature Information System) types. Variable importance plots showed that the distribution of habitats was driven by distance from coast, depth, and lithotype, and allowed mapping their distribution over the MPA. Congruence between observed and predicted distributions and accuracy of the classification was high. Results allowed calculating the occurrence of the distinct coralligenous habitats in zones with different protection level. The five habitats are unequally protected since the protection regime was established when detailed marine habitat maps were not available. A SWOT (Strengths-Weaknesses-Opportunities-Threats) analysis was performed to identify critical points and potentialities of the method. The method developed proved to be reliable and the results obtained will be useful when modulating on-going and future management actions in the studied area and in other Mediterranean MPAs to develop conservation efforts at basin scale.

A decomposition model and voxel selection framework for fMRI analysis to predict neural response of visual stimuli

This paper presents an fMRI signal analysis methodology using geometric mean curve decomposition (GMCD) and mutual information-based voxel selection framework. Previously, the fMRI signal analysis has been conducted using empirical mean curve decomposition (EMCD) model and voxel selection on raw fMRI signal. The erstwhile methodology loses frequency component, while the latter methodology suffers from signal redundancy. Both challenges are addressed by our methodology in which the frequency component is considered by decomposing the raw fMRI signal using geometric mean rather than arithmetic mean and the voxels are selected from EMCD signal using GMCD components, rather than raw fMRI signal. The proposed methodologies are adopted for predicting the neural response. Experimentations are conducted in the openly available fMRI data of six subjects, and comparisons are made with existing decomposition models and voxel selection frameworks. Subsequently, the effect of degree of selected voxels and the selection constraints are analyzed. The comparative results and the analysis demonstrate the superiority and the reliability of the proposed methodology.

Effect of Explicit Evaluation on Neural Connectivity Related to Listening to Unfamiliar Music

People can experience different emotions when listening to music. A growing number of studies have investigated the brain structures and neural connectivities associated with perceived emotions. However, very little is known about the effect of an explicit act of judgment on the neural processing of emotionally-valenced music. In this study, we adopted the novel consensus clustering paradigm, called binarisation of consensus partition matrices (Bi-CoPaM), to study whether and how the conscious aesthetic evaluation of the music would modulate brain connectivity networks related to emotion and reward processing. Participants listened to music under three conditions - one involving a non-evaluative judgment, one involving an explicit evaluative aesthetic judgment, and one involving no judgment at all (passive listening only). During non-evaluative attentive listening we obtained auditory-limbic connectivity whereas when participants were asked to decide explicitly whether they liked or disliked the music excerpt, only two clusters of intercommunicating brain regions were found: one including areas related to auditory processing and action observation, and the other comprising higher-order structures involved with visual processing. Results indicate that explicit evaluative judgment has an impact on the neural auditory-limbic connectivity during affective processing of music.

Comparison of fMRI analysis methods for heterogeneous BOLD responses in block design studies

A large number of fMRI studies have shown that the temporal dynamics of evoked BOLD responses can be highly heterogeneous. Failing to model heterogeneous responses in statistical analysis can lead to significant errors in signal detection and characterization and alter the neurobiological interpretation. However, to date it is not clear that, out of a large number of options, which methods are robust against variability in the temporal dynamics of BOLD responses in block-design studies. Here, we used rodent optogenetic fMRI data with heterogeneous BOLD responses and simulations guided by experimental data as a means to investigate different analysis methods' performance against heterogeneous BOLD responses. Evaluations are carried out within the general linear model (GLM) framework and consist of standard basis sets as well as independent component analysis (ICA). Analyses show that, in the presence of heterogeneous BOLD responses, conventionally used GLM with a canonical basis set leads to considerable errors in the detection and characterization of BOLD responses. Our results suggest that the 3rd and 4th order gamma basis sets, the 7th to 9th order finite impulse response (FIR) basis sets, the 5th to 9th order B-spline basis sets, and the 2nd to 5th order Fourier basis sets are optimal for good balance between detection and characterization, while the 1st order Fourier basis set (coherence analysis) used in our earlier studies show good detection capability. ICA has mostly good detection and characterization capabilities, but detects a large volume of spurious activation with the control fMRI data.

SOM and MCODE methods of defining functional clusters in MRI of the brain

Recent advances in analysis of fMRI have established the existence of functional sub-networks in the human brain that are active during the performance of visual, motor, language, and other tasks. We describe two computational methods of delineating functional sub-networks that are active when an individual performs an approach-avoidance paradigm. The paradigm consisted of presentation of images of pleasant and unpleasant faces that were shown to nine volunteers for 10 seconds after a preceding rest period of 50 seconds during which a green computer screen was displayed. The subjects were instructed to squeeze a ball with their right hand if they judged the face to be unpleasant, in which case the unpleasant face would disappear. An fMRI BOLD activation was created and used as input for two different kinds of clustering method: The MCODE algorithm based on graph-theoretical analysis and a Conscious Self-Organizing Map (CSOM). Clustering obtained with both methods was based on the temporal variations of the fMRI BOLD signal activity. Both methods identified distinct regions in the brain which were separated by long-range connections. The MCODE algorithm was supplied with time-courses for activated voxels when performing the paradigm, while the CSOM clustering used all voxels in the brain. Both yielded similar clusters for activated voxels. The combination of MCODE and CSOM presents a new approach in identifying functional subunits in the human brain and warrants further investigation into the subject.

Exploiting Complexity Information for Brain Activation Detection

We present a complexity-based approach for the analysis of fMRI time series, in which sample entropy (SampEn) is introduced as a quantification of the voxel complexity. Under this hypothesis the voxel complexity could be modulated in pertinent cognitive tasks, and it changes through experimental paradigms. We calculate the complexity of sequential fMRI data for each voxel in two distinct experimental paradigms and use a nonparametric statistical strategy, the Wilcoxon signed rank test, to evaluate the difference in complexity between them. The results are compared with the well known general linear model based Statistical Parametric Mapping package (SPM12), where a decided difference has been observed. This is because SampEn method detects brain complexity changes in two experiments of different conditions and the data-driven method SampEn evaluates just the complexity of specific sequential fMRI data. Also, the larger and smaller SampEn values correspond to different meanings, and the neutral-blank design produces higher predictability than threat-neutral. Complexity information can be considered as a complementary method to the existing fMRI analysis strategies, and it may help improving the understanding of human brain functions from a different perspective.

Functional Magnetic Resonance Imaging Methods

Since its inception in 1992, Functional Magnetic Resonance Imaging (fMRI) has become an indispensible tool for studying cognition in both the healthy and dysfunctional brain. FMRI monitors changes in the oxygenation of brain tissue resulting from altered metabolism consequent to a task-based evoked neural response or from spontaneous fluctuations in neural activity in the absence of conscious mentation (the "resting state"). Task-based studies have revealed neural correlates of a large number of important cognitive processes, while fMRI studies performed in the resting state have demonstrated brain-wide networks that result from brain regions with synchronized, apparently spontaneous activity. In this article, we review the methods used to acquire and analyze fMRI signals.

Combining self-organizing mapping and supervised affinity propagation clustering approach to investigate functional brain networks involved in motor imagery and execution with fMRI measurements

Clustering analysis methods have been widely applied to identifying the functional brain networks of a multitask paradigm. However, the previously used clustering analysis techniques are computationally expensive and thus impractical for clinical applications. In this study a novel method, called SOM-SAPC that combines self-organizing mapping (SOM) and supervised affinity propagation clustering (SAPC), is proposed and implemented to identify the motor execution (ME) and motor imagery (MI) networks. In SOM-SAPC, SOM was first performed to process fMRI data and SAPC is further utilized for clustering the patterns of functional networks. As a result, SOM-SAPC is able to significantly reduce the computational cost for brain network analysis. Simulation and clinical tests involving ME and MI were conducted based on SOM-SAPC, and the analysis results indicated that functional brain networks were clearly identified with different response patterns and reduced computational cost. In particular, three activation clusters were clearly revealed, which include parts of the visual, ME and MI functional networks. These findings validated that SOM-SAPC is an effective and robust method to analyze the fMRI data with multitasks.

Overcoming the effects of false positives and threshold bias in graph theoretical analyses of neuroimaging data

Graph theory (GT) is a powerful framework for quantifying topological features of neuroimaging-derived functional and structural networks. However, false positive (FP) connections arise frequently and influence the inferred topology of networks. Thresholding is often used to overcome this problem, but an appropriate threshold often relies on a priori assumptions, which will alter inferred network topologies. Four common network metrics (global efficiency, mean clustering coefficient, mean betweenness and smallworldness) were tested using a model tractography dataset. It was found that all four network metrics were significantly affected even by just one FP. Results also show that thresholding effectively dampens the impact of FPs, but at the expense of adding significant bias to network metrics. In a larger number (n=248) of tractography datasets, statistics were computed across random group permutations for a range of thresholds, revealing that statistics for network metrics varied significantly more than for non-network metrics (i.e., number of streamlines and number of edges). Varying degrees of network atrophy were introduced artificially to half the datasets, to test sensitivity to genuine group differences. For some network metrics, this atrophy was detected as significant (p<0.05, determined using permutation testing) only across a limited range of thresholds. We propose a multi-threshold permutation correction (MTPC) method, based on the cluster-enhanced permutation correction approach, to identify sustained significant effects across clusters of thresholds. This approach minimises requirements to determine a single threshold a priori. We demonstrate improved sensitivity of MTPC-corrected metrics to genuine group effects compared to an existing approach and demonstrate the use of MTPC on a previously published network analysis of tractography data derived from a clinical population. In conclusion, we show that there are large biases and instability induced by thresholding, making statistical comparisons of network metrics difficult. However, by testing for effects across multiple thresholds using MTPC, true group differences can be robustly identified.

Voxel-based clustered imaging by multiparameter diffusion tensor images for glioma grading

Gliomas are the most common intra-axial primary brain tumour; therefore, predicting glioma grade would influence therapeutic strategies. Although several methods based on single or multiple parameters from diagnostic images exist, a definitive method for pre-operatively determining glioma grade remains unknown. We aimed to develop an unsupervised method using multiple parameters from pre-operative diffusion tensor images for obtaining a clustered image that could enable visual grading of gliomas. Fourteen patients with low-grade gliomas and 19 with high-grade gliomas underwent diffusion tensor imaging and three-dimensional T1-weighted magnetic resonance imaging before tumour resection. Seven features including diffusion-weighted imaging, fractional anisotropy, first eigenvalue, second eigenvalue, third eigenvalue, mean diffusivity and raw T2 signal with no diffusion weighting, were extracted as multiple parameters from diffusion tensor imaging. We developed a two-level clustering approach for a self-organizing map followed by the K-means algorithm to enable unsupervised clustering of a large number of input vectors with the seven features for the whole brain. The vectors were grouped by the self-organizing map as protoclusters, which were classified into the smaller number of clusters by K-means to make a voxel-based diffusion tensor-based clustered image. Furthermore, we also determined if the diffusion tensor-based clustered image was really helpful for predicting pre-operative glioma grade in a supervised manner. The ratio of each class in the diffusion tensor-based clustered images was calculated from the regions of interest manually traced on the diffusion tensor imaging space, and the common logarithmic ratio scales were calculated. We then applied support vector machine as a classifier for distinguishing between low- and high-grade gliomas. Consequently, the sensitivity, specificity, accuracy and area under the curve of receiver operating characteristic curves from the 16-class diffusion tensor-based clustered images that showed the best performance for differentiating high- and low-grade gliomas were 0.848, 0.745, 0.804 and 0.912, respectively. Furthermore, the log-ratio value of each class of the 16-class diffusion tensor-based clustered images was compared between low- and high-grade gliomas, and the log-ratio values of classes 14, 15 and 16 in the high-grade gliomas were significantly higher than those in the low-grade gliomas (p < 0.005, p < 0.001 and p < 0.001, respectively). These classes comprised different patterns of the seven diffusion tensor imaging-based parameters. The results suggest that the multiple diffusion tensor imaging-based parameters from the voxel-based diffusion tensor-based clustered images can help differentiate between low- and high-grade gliomas.

•

We have developed a novel unsupervised method for voxel-based clustered imaging.

•

Each class ratio in clustered images differentiated high from low-grade gliomas.

•

The 16-class clustered images showed the best performance for the differentiation.

•

Each class comprised different patterns of the seven diffusion tensor-based features.

•

Multiple parameters from diffusion tensor images are useful for glioma grading.

Unsupervised spatiotemporal analysis of fMRI data using graph-based visualizations of self-organizing maps

We present novel graph-based visualizations of self-organizing maps for unsupervised functional magnetic resonance imaging (fMRI) analysis. A self-organizing map is an artificial neural network model that transforms high-dimensional data into a low-dimensional (often a 2-D) map using unsupervised learning. However, a postprocessing scheme is necessary to correctly interpret similarity between neighboring node prototypes (feature vectors) on the output map and delineate clusters and features of interest in the data. In this paper, we used graph-based visualizations to capture fMRI data features based upon 1) the distribution of data across the receptive fields of the prototypes (density-based connectivity); and 2) temporal similarities (correlations) between the prototypes (correlation-based connectivity). We applied this approach to identify task-related brain areas in an fMRI reaction time experiment involving a visuo-manual response task, and we correlated the time-to-peak of the fMRI responses in these areas with reaction time. Visualization of self-organizing maps outperformed independent component analysis and voxelwise univariate linear regression analysis in identifying and classifying relevant brain regions. We conclude that the graph-based visualizations of self-organizing maps help in advanced visualization of cluster boundaries in fMRI data enabling the separation of regions with small differences in the timings of their brain responses.

Machine learning and radiology

In this paper, we give a short introduction to machine learning and survey its applications in radiology. We focused on six categories of applications in radiology: medical image segmentation, registration, computer aided detection and diagnosis, brain function or activity analysis and neurological disease diagnosis from fMR images, content-based image retrieval systems for CT or MRI images, and text analysis of radiology reports using natural language processing (NLP) and natural language understanding (NLU). This survey shows that machine learning plays a key role in many radiology applications. Machine learning identifies complex patterns automatically and helps radiologists make intelligent decisions on radiology data such as conventional radiographs, CT, MRI, and PET images and radiology reports. In many applications, the performance of machine learning-based automatic detection and diagnosis systems has shown to be comparable to that of a well-trained and experienced radiologist. Technology development in machine learning and radiology will benefit from each other in the long run. Key contributions and common characteristics of machine learning techniques in radiology are discussed. We also discuss the problem of translating machine learning applications to the radiology clinical setting, including advantages and potential barriers.

REGION SIMILARITY RELATIONSHIP BETWEEN WATERSHED AND PENALIZED FUZZY HOPFIELD NEURAL NETWORK ALGORITHMS FOR BRAIN IMAGE SEGMENTATION

A robust image segmentation method that combines the watershed segmentation and penalized fuzzy Hopfield neural network (PFHNN) algorithms to minimize undesirable over-segmentation is described in this paper. This method incorporates spatial graph representation derived from the watershed segmented regions and cluster analysis information obtained from the PFHNN algorithm to achieve efficient image segmentation. The proposed scheme employs the Markov random field (MRF) model on the region adjacency graph (RAG) to improve the quality of watershed segmentation. Here, the fusion criterion is according to the correlation coefficient, which uses inter-region similarities to determine the merging of regions. Analysis of the performance of the proposed technique is presented through quantitative and qualitative validation experiments on benchmark images, and significant and promising segmentation results are presented using brain phantom simulated data.

Search for patterns of functional specificity in the brain: a nonparametric hierarchical Bayesian model for group fMRI data

Functional MRI studies have uncovered a number of brain areas that demonstrate highly specific functional patterns. In the case of visual object recognition, small, focal regions have been characterized with selectivity for visual categories such as human faces. In this paper, we develop an algorithm that automatically learns patterns of functional specificity from fMRI data in a group of subjects. The method does not require spatial alignment of functional images from different subjects. The algorithm is based on a generative model that comprises two main layers. At the lower level, we express the functional brain response to each stimulus as a binary activation variable. At the next level, we define a prior over sets of activation variables in all subjects. We use a Hierarchical Dirichlet Process as the prior in order to learn the patterns of functional specificity shared across the group, which we call functional systems, and estimate the number of these systems. Inference based on our model enables automatic discovery and characterization of dominant and consistent functional systems. We apply the method to data from a visual fMRI study comprised of 69 distinct stimulus images. The discovered system activation profiles correspond to selectivity for a number of image categories such as faces, bodies, and scenes. Among systems found by our method, we identify new areas that are deactivated by face stimuli. In empirical comparisons with previously proposed exploratory methods, our results appear superior in capturing the structure in the space of visual categories of stimuli.

Analysis of FMRI data using an integrated principal component analysis and supervised affinity propagation clustering approach

Clustering analysis is a promising data-driven method for analyzing functional magnetic resonance imaging (fMRI) time series data. The huge computational load, however, creates practical difficulties for this technique. We present a novel approach, integrating principal component analysis (PCA) and supervised affinity propagation clustering (SAPC). In this method, fMRI data are initially processed by PCA to obtain a preliminary image of brain activation. SAPC is then used to detect different brain functional activation patterns. We used a supervised Silhouette index to optimize clustering quality and automatically search for the optimal parameter p in SAPC, so that the basic affinity propagation clustering is improved by applying SAPC. Four simulation studies and tests with three in vivo fMRI datasets containing data from both block-design and event-related experiments revealed that functional brain activation was effectively detected and different response patterns were distinguished using our integrated method. In addition, the improved SAPC method was superior to the k -centers clustering and hierarchical clustering methods in both block-design and event-related fMRI data, as measured by the average squared error. These results suggest that our proposed novel integrated approach will be useful for detecting brain functional activation in both block-design and event-related experimental fMRI data.

Feature-space-based FMRI analysis using the optimal linear transformation

The optimal linear transformation (OLT), an image analysis technique of feature space, was first presented in the field of MRI. This paper proposes a method of extending OLT from MRI to functional MRI (fMRI) to improve the activation-detection performance over conventional approaches of fMRI analysis. In this method, first, ideal hemodynamic response time series for different stimuli were generated by convolving the theoretical hemodynamic response model with the stimulus timing. Second, constructing hypothetical signature vectors for different activity patterns of interest by virtue of the ideal hemodynamic responses, OLT was used to extract features of fMRI data. The resultant feature space had particular geometric clustering properties. It was then classified into different groups, each pertaining to an activity pattern of interest; the applied signature vector for each group was obtained by averaging. Third, using the applied signature vectors, OLT was applied again to generate fMRI composite images with high SNRs for the desired activity patterns. Simulations and a blocked fMRI experiment were employed for the method to be verified and compared with the general linear model (GLM)-based analysis. The simulation studies and the experimental results indicated the superiority of the proposed method over the GLM-based analysis in detecting brain activities.

A novel model-free data analysis technique based on clustering in a mutual information space: application to resting-state FMRI

Non-parametric data-driven analysis techniques can be used to study datasets with few assumptions about the data and underlying experiment. Variations of independent component analysis (ICA) have been the methods mostly used on fMRI data, e.g., in finding resting-state networks thought to reflect the connectivity of the brain. Here we present a novel data analysis technique and demonstrate it on resting-state fMRI data. It is a generic method with few underlying assumptions about the data. The results are built from the statistical relations between all input voxels, resulting in a whole-brain analysis on a voxel level. It has good scalability properties and the parallel implementation is capable of handling large datasets and databases. From the mutual information between the activities of the voxels over time, a distance matrix is created for all voxels in the input space. Multidimensional scaling is used to put the voxels in a lower-dimensional space reflecting the dependency relations based on the distance matrix. By performing clustering in this space we can find the strong statistical regularities in the data, which for the resting-state data turns out to be the resting-state networks. The decomposition is performed in the last step of the algorithm and is computationally simple. This opens up for rapid analysis and visualization of the data on different spatial levels, as well as automatically finding a suitable number of decomposition components.

Automated classification of multi-spectral MR images using Linear Discriminant Analysis

Magnetic resonance imaging (MRI) is a valuable instrument in medical science owing to its capabilities in soft tissue characterization and 3D visualization. A potential application of MRI in clinical practice is brain parenchyma classification. This work proposes a novel approach called "Unsupervised Linear Discriminant Analysis (ULDA)" to classify and segment the three major tissues, i.e. gray matter (GM), white matter (WM) and cerebral spinal fluid (CSF), from a multi-spectral MR image of the human brain. The ULDA comprises two processes, namely Target Generation Process (TGP) and Linear Discriminant Analysis (LDA) classification. TGP is a fuzzy-set process that generates a set of potential targets from unknown information, and applies these targets to train the optimal division boundary by LDA, such that three tissues GM, WM and CSF are separated. Finally, two sets of images, namely computer-generated phantom images and real MR images are used in the experiments to evaluate the effectiveness of ULDA. Experiment results reveal that UDLA segments a multi-spectral MR image much more effectively than either FMRIB's Automated Segmentation Tool (FAST) or Fuzzy C-means (FC).

Convolution power spectrum analysis for FMRI data based on prior image signal

Functional MRI (fMRI) data-processing methods based on changes in the time domain involve, among other things, correlation analysis and use of the general linear model with statistical parametric mapping (SPM). Unlike conventional fMRI data analysis methods, which aim to model the blood-oxygen-level-dependent (BOLD) response of voxels as a function of time, the theory of power spectrum (PS) analysis focuses completely on understanding the dynamic energy change of interacting systems. We propose a new convolution PS (CPS) analysis of fMRI data, based on the theory of matched filtering, to detect brain functional activation for fMRI data. First, convolution signals are computed between the measured fMRI signals and the image signal of prior experimental pattern to suppress noise in the fMRI data. Then, the PS density analysis of the convolution signal is specified as the quantitative analysis energy index of BOLD signal change. The data from simulation studies and in vivo fMRI studies, including block-design experiments, reveal that the CPS method enables a more effective detection of some aspects of brain functional activation, as compared with the canonical PS SPM and the support vector machine methods. Our results demonstrate that the CPS method is useful as a complementary analysis in revealing brain functional information regarding the complex nature of fMRI time series.

Unsupervised spatiotemporal fMRI data analysis using support vector machines

In this work we present a new support vector machine (SVM)-based method for fMRI data analysis. SVM has been shown to be a powerful, efficient data-driven tool in pattern recognition, and has been applied to the supervised classification of brain cognitive states in fMRI experiments. We examine the unsupervised mapping of activated brain regions using SVM. Specifically, the mapping process is formulated as an outlier detection problem of one-class SVM (OCSVM) that provides initial mapping results. These results are further refined by applying prototype selection and SVM reclassification. Multiple spatial and temporal features are extracted and selected to facilitate SVM learning. The proposed method was compared with correlation analysis (CA), t-test (TT), and spatial independent component analysis (SICA) methods using synthetic and experimental data. Our results show that the proposed method can provide more accurate and robust activation mapping than CA, TT and SICA, and is computationally more efficient than SICA. Besides its applicability to typical fMRI experiments, the proposed method is also a powerful tool in fMRI studies where a reliable quantification of activated brain regions is required.

The optimal linear transformation-based fMRI feature space analysis

This paper proposes a method of extending the optimal linear transformation (OLT), an image analysis technique of feature space, from magnetic resonance imaging (MRI) to functional magnetic resonance imaging (fMRI) so as to improve the activation detection performance over conventional approaches of fMRI analysis. The method was: (1) ideal hemodynamic responses for different stimuli were generated by convolving the theoretical hemodynamic response model with the stimulus timing, (2) considering the ideal hemodynamic responses as hypothetical signature vectors for different activity patterns of interest, OLT was used to extract the features of fMRI data. The resultant feature space had particular geometric clustering properties. It was then classified into different groups, each pertaining to an activity pattern of interest; the applied signature vector for each group was obtained by averaging, (3) using the applied signature vectors, OLT was applied again to generate fMRI composite images with high SNRs for the desired activity patterns. Simulations and a blocked fMRI experiment were employed to validate the proposed method. The simulation and the experiment results indicated the proposed method was capable of improving some conventional methods to be more sensitive to activations, having strong contrast between activations and inactivations, and being more valid for complex activity patterns.

Dissociating functional brain networks by decoding the between-subject variability

In this study we illustrate how the functional networks involved in a single task (e.g. the sensory, cognitive and motor components) can be segregated without cognitive subtractions at the second-level. The method used is based on meaningful variability in the patterns of activation between subjects with the assumption that regions belonging to the same network will have comparable variations from subject to subject. fMRI data were collected from thirty nine healthy volunteers who were asked to indicate with a button press if visually presented words were semantically related or not. Voxels were classified according to the similarity in their patterns of between-subject variance using a second-level unsupervised fuzzy clustering algorithm. The results were compared to those identified by cognitive subtractions of multiple conditions tested in the same set of subjects. This illustrated that the second-level clustering approach (on activation for a single task) was able to identify the functional networks observed using cognitive subtractions (e.g. those associated with vision, semantic associations or motor processing). In addition the fuzzy clustering approach revealed other networks that were not dissociated by the cognitive subtraction approach (e.g. those associated with high- and low-level visual processing and oculomotor movements). We discuss the potential applications of our method which include the identification of “hidden” or unpredicted networks as well as the identification of systems level signatures for different subgroupings of clinical and healthy populations.

Review of methods for functional brain connectivity detection using fMRI

Since the mid of 1990s, functional connectivity study using fMRI (fcMRI) has drawn increasing attention of neuroscientists and computer scientists, since it opens a new window to explore functional network of human brain with relatively high resolution. A variety of methods for fcMRI study have been proposed. This paper intends to provide a technical review on computational methodologies developed for fcMRI analysis. From our perspective, these computational methods are classified into two general categories: model-driven methods and data-driven methods. Data-driven methods are a large family, and thus are further sub-classified into decomposition-based methods and clustering analysis methods. For each type of methods, principles, main contributors, and their advantages and drawbacks are discussed. Finally, potential applications of fcMRI are overviewed.

Selection criteria for the analysis of data-driven clusters in cerebral FMRI

Functional MRI (fMRI) may be possible without a priori models of the cerebral hemodynamic response. First, such data-driven fMRI requires that all cerebral territories with distinct patterns be identified. Second, a systematic selection method is necessary to prevent the subjective interpretation of the identified territories. This paper addresses the second point by proposing a novel method for the automated interpretation of identified territories in data-driven fMRI. Selection criteria are formulated using: 1) the temporal cross-correlation between each identified territory and the paradigm and 2) the spatial contiguity of the corresponding voxel map. Ten event-design fMRI data sets are analyzed with one prominent algorithm, fuzzy c-means clustering, before applying the selection criteria. For comparison, these data are also analyzed with an established, model-based method: statistical parametric mapping. Both methods produced similar results and identified potential activation in the expected territory of the sensorimotor cortex in all ten data sets. Moreover, the proposed method classified distinct territories in separate clusters. Selected clusters have a mean temporal correlation coefficient of 0.39+/-0.07 (n=19) with a mean 2.7+/-1.4 second response delay. At most, four separate contiguous territories were observed in 87% of these clusters. These results suggest that the proposed method may be effective for exploratory fMRI studies where the hemodynamic response is perturbed during cerebrovascular disease.

Analysis of fMRI data using improved self-organizing mapping and spatio-temporal metric hierarchical clustering

The self-organizing mapping (SOM) and hierarchical clustering (HC) methods are integrated to detect brain functional activation; functional magnetic resonance imaging (fMRI) data are first processed by SOM to obtain a primary merged neural nodes image, and then by HC to obtain further brain activation patterns. The conventional Euclidean distance metric was replaced by the correlation distance metric in SOM to improve clustering and merging of neural nodes. To improve the use of spatial and temporal information in fMRI data, a new spatial distance (node coordinates in the 2-D lattice) and temporal correlation (correlation degree of each time course in the exemplar matrix) are introduced in HC to merge the primary SOM results. Two simulation studies and two in vivo fMRI data that both contained block-design and event-related experiments revealed that brain functional activation can be effectively detected and that different response patterns can be distinguished using these methods. Our results demonstrate that the improved SOM and HC methods are clearly superior to the statistical parametric mapping (SPM), independent component analysis (ICA), and conventional SOM methods in the block-design, especially in the event-related experiment, as revealed by their performance measured by receiver operating characteristic (ROC) analysis. Our results also suggest that the proposed new integrated approach could be useful in detecting block-design and event-related fMRI data.

Contribution of exploratory methods to the investigation of extended large-scale brain networks in functional MRI: methodologies, results, and challenges

A large-scale brain network can be defined as a set of segregated and integrated regions, that is, distant regions that share strong anatomical connections and functional interactions. Data-driven investigation of such networks has recently received a great deal of attention in blood-oxygen-level-dependent (BOLD) functional magnetic resonance imaging (fMRI). We here review the rationale for such an investigation, the methods used, the results obtained, and also discuss some issues that have to be faced for an efficient exploration.

Analysis of dynamic susceptibility contrast MRI time series based on unsupervised clustering methods

We compare five different unsupervised clustering techniques as tools for the analysis of dynamic susceptibility contrast MRI time series. The study included four subjects: two subjects with stroke and two subjects without focal neurological deficit. The goal was to determine the robustness and reliability of clustering methods in providing a self-organized segmentation of perfusion MRI data sharing common properties of signal dynamics. For this purpose, the relative signal reduction time series was computed for each pixel. Clustering of the resulting high-dimensional feature vectors was performed by minimal free-energy deterministic annealing, self-organizing maps, two variants of fuzzy c-means clustering (FVQ and FSM), and the neural gas algorithm. Clustering results were evaluated by visual assessment of cluster assignment maps and corresponding signal time curves as well as by quantitative comparison of cluster assignment maps with conventional pixel-specific perfusion parameter maps based on quantitative receiver operating characteristic (ROC) curve analysis. Clustering methods provided a functional segmentation with respect to vessel size, detected side asymmetries of contrast-agent first pass, and identified regions of perfusion deficits in subjects with stroke. As confirmed by quantitative ROC analysis, the clustering approach can detect regions of reduced brain perfusion with high accuracy when compared to conventional analysis by pixel-specific cerebral blood volume and mean transit time maps. We conclude that by unveiling differences of signal dynamics and amplitude, clustering is a useful tool to analyze and visualize regional properties of brain perfusion. Thus, it may contribute to the computer-aided diagnosis of cerebral circulation deficits by noninvasive neuroimaging.

Dense mode clustering in brain maps

A mode-based clustering method is developed for identifying spatially dense clusters in brain maps. This type of clustering focuses on identifying clusters in brain maps independent of their shape or overall variance. This can be useful for both localization in terms of interpretation and for subsequent graphical analysis that might require more coherent or dense regions of interest as starting points. The method automatically does signal/noise sharpening through density mode seeking. We also discuss the problem of parameter selection with this method and propose a new method involving 2-parameter control surface, in which we show that the same cluster solution results from tradeoff of these 2 parameters (the local density k and the radius r of the spherical kernel). We benchmark the new dense mode clustering by using several artificially created data sets and brain imaging data sets from an event perception task by perturbing the data set with noise and measuring three kinds of deviation from the original cluster solution. We present benchmark results that demonstrate that the mode clustering method consistently outperforms the commonly used single-linkage clustering, k means method (centroid method) and Ward's method (variance method).

A spectral clustering approach to fMRI activation detection

Conventional clustering methods for fMRI activation detection implicitly assume that data scatter in clusters with certain shapes. But this assumption is inconsistent with the general reality in fMRI data, and will consequently achieve detection results with higher false alarm rate. To solve this problem, we propose an alternative clustering method, namely spectral cluster analysis (SCA), which uses eigenvectors of a matrix derived from the dataset to cluster the wavelet coefficients extracted from the fMRI time series. Experimental results demonstrate reliability and flexibility of this new fMRI clustering approach.

Studies of chinese original quiet sitting by using functional magnetic resonance imaging

Since different meditations may activate different regions in brain, we can use functional magnetic resonance imaging (fMRI) to investigate it. Chinese original quiet sitting is mainly one kind of traditional Chinese meditation. It contains two different parts: a short period of keeping phrase and intake spiritual energy, and a long period of relaxation with no further action. In this paper, both those two stages were studied by fMRI. We performed two different paradigms and found the accurate positions in the brain. The pineal gland and the hypothalamus showed positive activation during the first and second stages of this meditation. The BOLD (Blood Oxygenation Level Dependent) signal changes had also been found.

Magnetic resonance imaging segmentation techniques using batch-type learning vector quantization algorithms

In this article, we propose batch-type learning vector quantization (LVQ) segmentation techniques for the magnetic resonance (MR) images. Magnetic resonance imaging (MRI) segmentation is an important technique to differentiate abnormal and normal tissues in MR image data. The proposed LVQ segmentation techniques are compared with the generalized Kohonen's competitive learning (GKCL) methods, which were proposed by Lin et al. [Magn Reson Imaging 21 (2003) 863-870]. Three MRI data sets of real cases are used in this article. The first case is from a 2-year-old girl who was diagnosed with retinoblastoma in her left eye. The second case is from a 55-year-old woman who developed complete left side oculomotor palsy immediately after a motor vehicle accident. The third case is from an 84-year-old man who was diagnosed with Alzheimer disease (AD). Our comparisons are based on sensitivity of algorithm parameters, the quality of MRI segmentation with the contrast-to-noise ratio and the accuracy of the region of interest tissue. Overall, the segmentation results from batch-type LVQ algorithms present good accuracy and quality of the segmentation images, and also flexibility of algorithm parameters in all the comparison consequences. The results support that the proposed batch-type LVQ algorithms are better than the previous GKCL algorithms. Specifically, the proposed fuzzy-soft LVQ algorithm works well in segmenting AD MRI data set to accurately measure the hippocampus volume in AD MR images.