Neelam Sinha

Cognitive state classification using transformed fMRI data

One approach, for understanding human brain functioning, is to analyze the changes in the brain while performing cognitive tasks. Towards this, Functional Magnetic Resonance (fMR) images of subjects performing well-defined tasks are widely utilized for task-specific analyses. In this work, we propose a procedure to enable classification between two chosen cognitive tasks, using their respective fMR image sequences. The time series of expert-marked anatomically-mapped relevant voxels are processed and fed as input to the classical Naive Bayesian and SVM classifiers. The processing involves use of random sieve function, phase information in the data transformed using Fourier and Hilbert transformations. This processing results in improved classification, as against using the voxel intensities directly, as illustrated. The novelty of the proposed method lies in utilizing the phase information in the transformed domain, for classifying between the cognitive tasks along with random sieve function chosen with a particular probability distribution. The proposed classification procedure is applied on a publicly available dataset, StarPlus data, with 6 subjects performing the two distinct cognitive tasks of watching either a picture or a sentence. The classification accuracy stands at an average of 65.6%(using Naive Bayes classifier) and 76.4%(using SVM classifier) for raw data. The corresponding classification accuracy stands at 96.8% and 97.5% for Fourier transformed data. For Hilbert transformed data, it is 93.7% and 99%, for 6 subjects, on 2 cognitive tasks.

Adaptive k-space sampling design for edge-enhanced DCE-MRI using compressed sensing.

The critical challenge in dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is the trade-off between spatial and temporal resolution due to the limited availability of acquisition time. To address this, it is imperative to under-sample k-space and to develop specific reconstruction techniques. Our proposed method reconstructs high-quality images from under-sampled dynamic k-space data by proposing two main improvements; i) design of an adaptive k-space sampling lattice and ii) edge-enhanced reconstruction technique. A high-resolution data set obtained before the start of the dynamic phase is utilized. The sampling pattern is designed to adapt to the nature of k-space energy distribution obtained from the static high-resolution data. For image reconstruction, the well-known compressed sensing-based total variation (TV) minimization constrained reconstruction scheme is utilized by incorporating the gradient information obtained from the static high-resolution data. The proposed method is tested on seven real dynamic time series consisting of 2 breast data sets and 5 abdomen data sets spanning 1196 images in all. For data availability of only 10%, performance improvement is seen across various quality metrics. Average improvements in Universal Image Quality Index and Structural Similarity Index Metric of up to 28% and 24% on breast data and about 17% and 9% on abdomen data, respectively, are obtained for the proposed method as against the baseline TV reconstruction with variable density random sampling pattern.

Characterization of White and Gray Matters in healthy brain: An in-vivo Diffusion Kurtosis Imaging study

The extensions of diffusion weighted magnetic resonance imaging (DW-MRI) technique such as Diffusion Tensor Imaging (DTI) and Diffusion Kurtosis Imaging (DKI) are useful in understanding the complex cellular micro-structures non-invasively. DTI is used to quantify the three-dimensional water diffusion in biological tissues by assuming the diffusion to be Gaussian, whereas DKI which is an extension of DTI, quantifies the deviation of water diffusion from Gaussian model. In this study, we focus on finding the correlation between DTI and DKI parameters for characterizing the brain regions such as White Matter (WM) and Gray Matter (GM) of healthy human brain. The data obtained from DKI scans on nine healthy volunteers are used in this study. Parameters such as Mean Kurtosis (MK) and Fractional Anisotropy (FA) were estimated for the two brain regions, WM and GM. Correlation analysis was performed between MK and FA in WM and GM. Significant differences in correlation results were observed for WM and GM. These preliminary findings indicate that this correlation measure could be used to characterize WM and GM regions in normal human brain. This could be used in applications such as segmentation of WM and GM and additionally bear potential for discriminating pathologies.

An improved approach of high graded glioma segmentation using sparse autoencoder and fuzzy c-means clustering from multi-modal MR images

Accurate and automated brain tumor segmentation using multi modal MR images is essential for the evaluation of the disease progression in order to improve disease diagnosis and treatment planning. We present a new fully automated method for high graded brain tumor segmentation combining sparse autoencoder and multimodal Fuzzy C-means clustering. The approach utilizes multimodal MRI contrast: T1, T2, FLAIR and T1c (contrast-enhanced) for 15 high graded glioma (HGG) subjects. The objective of the proposed study is to segment tumor tissues from HGG including edema and tumor core within edema. The segmentation was performed on the training data of the multimodal brain tumor image segmentation benchmark 2015. Sparse autoencoder, which is an unsupervised learning algorithm, was used to automatically learn features from unlabeled dataset of tumor in order to segment edema. Followed by edema segmentation, tumor core was segmented from edema using multimodal FCM clustering. Evaluating the performance of the segmentation results with the ground truth yields high dice score (DS) of 0.9866±0.01 and 0.9843±0.01 for edema and tumor core respectively and high Jaccard similarity (JS) of 0.9738±0.02 and 0.9692±0.02 for edema and tumor core respectively; showed high accuracy in segmenting the complex tumor structures from multi-contrast MR scans of HGG patients. We also compared our methodology in terms of segmentation efficiency with some recent techniques reported in proceedings of MICCAI-BRATS challenge 2015.

Quantitative analysis of structural variations in corpus callosum in adults with multiple system atrophy (MSA)

Multiple system atrophy (MSA) is a rare, non-curable, progressive neurodegenerative disorder that affects nervous system and movement, poses a considerable diagnostic challenge to medical researchers. Corpus callosum (CC) being the largest white matter structure in brain, enabling inter-hemispheric communication, quantification of callosal atrophy may provide vital information at the earliest possible stages. The main objective is to identify the differences in CC structure for this disease, based on quantitative analysis on the pattern of callosal atrophy. We report results of quantification of structural changes in regional anatomical thickness, area and length of CC between patient-groups with MSA with respect to healthy controls. The method utilizes isolating and parcellating the mid-sagittal CC into 100 segments along the length - measuring the width of each segment. It also measures areas within geometrically defined five callosal compartments of the well-known Witelson, and Hofer-Frahma schemes. For quantification, statistical tests are performed on these different callosal measurements. From the statistical analysis, it is concluded that compared to healthy controls, width is reduced drastically throughout CC for MSA group and as well as changes in area and length are also significant for MSA. The study is further extended to check if any significant difference in thickness is found between the two variations of MSA, Parkinsonian MSA and Cerebellar MSA group, using the same methodology. However area and length of this two sub-MSA group, no substantial difference is obtained. The study is performed on twenty subjects for each control and MSA group, who had T1-weighted MRI.

Customizing CNNs for blood vessel segmentation from fundus images

For automatic screening of eye diseases, it is very important to segment regions corresponding to the different eye-parts from the fundal images. A challenging task, in this context, is to segment the network of blood vessels. The blood vessel network runs all along the fundal image, varying in density and fineness of structure. Besides, changes in illumination, color and pathology also add to the difficulties in blood vessel segmentation. In this paper, we propose segmentation of blood vessels from fundal images in the deep learning framework, without any pre-processing. A deep convolutional network, consisting of 8 convolutional layers and 3 pooling layers in between, is used to achieve the segmentation. In this work, a Convolutional Neural Network currently in use for semantic image segmentation is customized for blood vessel segmentation by replacing the output layer with a convolutional layer of kernel size 1 × 1 which generates the final segmented image. The output of CNN is a gray scale image and is binarized by thresholding. The proposed method is applied on 2 publicly available databases DRIVE and HRF (capturing diversity in image resolution), consisting of healthy and diseased fundal images boosted by mirror versions of the originals. The method results in an accuracy of 93.94% and yields 0.894 as area under the ROC curve on the test data comprising of randomly selected 23 images from HRF dataset. The promising results illustrate generalizability of the proposed approach.

Distribution Based EEG Baseline Classification

Electrical signals generated in the brain, known as Electroencephalographic (EEG) signals, form a non-invasive measure of brain functioning. Baseline states of EEG are Eyes Open (EO) and Eyes Closed (EC) relaxed states. The choice of baseline used in an experiment is of critical importance since they form a reference with which other states are measured. In Brain Machine Interface, it is imperative that the system should be able to distinguish between these states and hence the need for automated classification of EEG baselines. In the proposed method, Statistical Moments are utilized. The Moment Generating Functions (MGFs) obtained using these moments are given as features to SVM and k-NN classifiers resulting in mean accuracies of 86.71% and 86.54%. The fact that MGF is able to differentiate between these states indicate that the two states have different source distribution parameters. A Smirnov test verified that the data of two classes indeed come from different distributions.

Statistical feature analysis for EEG baseline classification: Eyes Open vs Eyes Closed

Electroencephalographic (EEG) patterns are electrical signals generated in the brain indicating brain functioning. Due to its non-invasive nature, it has been used in applications ranging from disorder detection, sleep analysis to Brain Machine Interface. A baseline state is required in all these applications to compare the required state with a reference state. In EEG analysis, Eyes Open (EO) and Eyes Closed (EC) relaxed states are the baselines used. The choice of baseline is important especially in Brain Machine Interface. Thus the system should be able to distinguish between these two states and hence the need for automated classification of EEG Baseline States. In the proposed approach, statistical features are used in classification of these two states along with Support Vector Machine (SVM) and k-Nearest Neighbour(k-NN) classifiers. Thirteen different statistical features are considered and it was found that the combination of kurtosis, IQR and MAD with k-NN classifier (k=7) gave the mean accuracy of 77.92%. The fact that kurtosis, IQR and MAD perform better implies that the underlying distributions of the two classes have significant difference.

Automation of cross-sectional analysis of neuroimages using diffusion kurtosis imaging

Recent advances in magnetic resonance imaging allow quantification of diffusion of water molecules, through diffusion weighted magnetic resonance images(DW-MRI)techniques called diffusion tensor imaging(DTI) and diffusion kurtosis imaging( DKI). These techniques are being extensively used to study finer details of the micro-structure in the human brain. Inspite of availability of several software packages for DW-MRI analysis, not many are available for estimating DKI. An easy to use and fully automated processing of DW-MRI for estimation and analysis of DKI is still lacking. In this paper, we describe the development of a MATLAB toolbox which facilitates estimation, display and cross-sectional analysis of brain images using DKI by fully automating the DKI analysis pipeline, right from initial data loading till performing statistical analysis with an user friendly graphical user interface(GUI). The major functionalities which are integrated in this toolbox includes diffusion data preprocessing, estimation of diffusion tensor and diffusion kurtosis metrics, displaying parametric maps and statistical analysis of diffusion metrics for cross sectional brain studies. The functionalities and performance of the toolbox has been extensively evaluated with multiple DW-MRI datasets acquired from normal subjects achieving very promising results. The capabilities of the toolbox are demonstrated by conducting in-vivo DKI studies for analysing the aging effects.

Extended neighbourhood based linear reconstruction of Diffusion Kurtosis Imaging

Diffusion weighted magnetic resonance imaging(DW-MRI) is used for the quantification of water diffusion with the availability of various tensor based models such as Diffusion Tensor Imaging (DTI) and Diffusion Kurtosis Imaging (DKI). The increased significance of DKI parameters for the assessment of neurologic diseases as compared to DTI parameters has been shown in several recent studies. Kurtosis tensors were reconstructed using either linear or non-linear least squares approaches including several variants of these approaches. In this work, we proposed an extended linear least squares(LLS) reconstruction of DKI parameters which makes use of the correlation existing in the DW-MRI data in order to have a robust and accurate estimation of the kurtosis parameters. All the available methods of DKI reconstruction uses an independent voxel-wise estimation of the kurtosis parameters. The proposed method attempts to make use of the spatial correlation in DW-MRI by including the neighbourhood voxels for the estimation of kurtosis parameters voxel-wise. Our study includes simulation and real data experiments for validation of the proposed method. The estimation from the proposed method revealed better details and accuracy as compared to LLS and weighted LLS approaches. The proposed method is also robust to noise which is illustrated by using noise corrupted data for different levels of added rician noise.