Nirmal Keshava

Neuro magnetic resonance spectroscopy using wavelet decomposition and statistical testing identifies biochemical changes in people with spinal cord injury and pain.

Spinal cord injury (SCI) can be accompanied by chronic pain, the mechanisms for which are poorly understood. Here we report that magnetic resonance spectroscopy measurements from the brain, collected at 3T, and processed using wavelet-based feature extraction and classification algorithms, can identify biochemical changes that distinguish control subjects from subjects with SCI as well as subdividing the SCI group into those with and without chronic pain. The results from control subjects (n=10) were compared to those with SCI (n=10). The SCI cohort was made up of subjects with chronic neuropathic pain (n=5) and those without chronic pain (n=5). The wavelet-based decomposition of frequency domain MRS signals employs statistical significance testing to identify features best suited to discriminate different classes. Moreover, the features benefit from careful attention to the post-processing of the spectroscopy data prior to the comparison of the three cohorts. The spectroscopy data, from the thalamus, best distinguished control subjects without SCI from those with SCI with a sensitivity and specificity of 0.9 (Percentage of Correct Classification). The spectroscopy data obtained from the prefrontal cortex and anterior cingulate cortex both distinguished between SCI subjects with chronic neuropathic pain and those without pain with a sensitivity and specificity of 1.0. In this study, where two underlying mechanisms co-exist (i.e. SCI and pain), the thalamic changes appear to be linked more strongly to SCI, while the anterior cingulate cortex and prefrontal cortex changes appear to be specifically linked to the presence of pain.

Algorithms for characterizing brain metabolites in two-dimensional in vivo MR correlation spectroscopy

Traditional analyses of in vivo 1D MR spectroscopy of brain metabolites have been limited to the inspection of one-dimensional free induction decay (FID) signals from which only a limited number of metabolites are clearly observable. In this article we introduce a novel set of algorithms to process and characterize two-dimensional in vivo MR correlation spectroscopy (2D COSY) signals. 2D COSY data was collected from phantom solutions of topical metabolites found in the brain, namely glutamine, glutamate, and creatine. A statistical peak-detection and object segmentation algorithm is adapted for 2D COSY signals and applied to phantom solutions containing varied concentrations of glutamine and glutamate. Additionally, quantitative features are derived from peak and object structures, and we show that these measures are correlated with known phantom metabolite concentrations. These results are encouraging for future studies focusing on neurological disorders that induce subtle changes in brain metabolite concentrations and for which accurate quantitation is important.

Joint analysis of differential mobility spectrometer and mass spectrometer features for tuberculosis biomarkers

In this article, we present results of recent efforts to identify biomarkers for tuberculosis using a differential mobility spectrometer (DMS). We focus specifically on the capability of exploiting a data collection system that employs a DMS in parallel with a mass spectrometer. This system permits previously developed algorithms for DMS to be used in conjunction with a device considered a gold-standard for chemical identification, making it a unique discovery tool for the determination of biomarkers.

In Vivo Brain Magnetic Resonance Spectroscopy: A Measurement of Biomarker Sensitivity to Post-Processing Algorithms

Clinical translation of reported biomarkers requires reliable and consistent algorithms to derive biomarkers. However, the literature reports statistically significant differences between 1-D MRS measurements from control groups and subjects with disease states but frequently provides little information on the algorithms and parameters used to process the data. The sensitivity of in vivo brain magnetic resonance spectroscopy biomarkers is investigated with respect to parameter values for two key stages of post-acquisitional processing. Our effort is specifically motivated by the lack of consensus on approaches and parameter values for the two critical operations, water resonance removal, and baseline correction. The different stages of data processing also introduce varying levels of uncertainty and arbitrary selection of parameter values can significantly underutilize the intrinsic differences between two classes of signals. The sensitivity of biomarkers points to the need for a better understanding of how all stages of post-acquisitional processing affect biomarker discovery and ultimately, clinical translation. Our results also highlight the possibility of optimizing biomarker discovery by the careful selection of parameters that best reveal class differences. Using previously reported data and biomarkers, our results demonstrate that small changes in parameter values affect the statistical significance and corresponding effect size of biomarkers. Consequently, it is possible to increase the strength of biomarkers by selecting optimal parameter values in different spectral intervals. Our analyses with a previously reported data set demonstrate an increase in effect sizes for wavelet-based biomarkers of up to 36%, with increases in classification performance of up to 12%.

Detection of Deception in Structured Interviews Using Sensors and Algorithms

Draper Laboratory and MRAC have recently completed a comprehensive study to quantitatively evaluate deception detection performance under different interviewing styles. The interviews were performed while multiple physiological waveforms were collected from participants to determine how well automated algorithms can detect deception based upon changes in physiology. We report the results of a multi-factorial experiment with 77 human participants who were deceptive on specific topics during interviews conducted with one of two styles: a forcing style which relies on more coercive or confrontational techniques, or a fostering approach, which relies on open-ended interviewing and elements of a cognitive interview. The interviews were performed in a state-of-the-art facility where multiple sensors simultaneously collect synchronized physiological measurements, including electrodermal response, relative blood pressure, respiration, pupil diameter, and ECG. Features extracted from these waveforms during honest and deceptive intervals were then submitted to a hypothesis test to evaluate their statistical significance. A univariate statistical detection algorithm then assessed the ability to detect deception for different interview configurations. Our paper will explain the protocol and experimental design for this study. Our results will be in terms of statistical significances, effect sizes, and ROC curves and will identify how promising features performed in different interview scenarios.

Analysis of biomarker features from a differential mobility spectrometer for the detection of tuberculosis

In this article, we present an investigation of an approach to extract discriminating features from differential mobility spectrometer (DMS) signals generated from two sets of in vitro samples of headspace that contain volatile organic compounds. The two classes of signals we analyze are a strain of tuberculosis grown in media and the media alone. Our approach first preprocesses the DMS signals to recover a baselined signal and then applies a wavelet transform to obtain localized measures of chemical activity in the detector output. The approach then ranks the wavelet coefficients using a common measure of class separability to identify distinguishing wavelet coefficients. Our analysis indicates that the subsequent ranking can often identify areas of signal devoid of chemical structures and that when discriminating chemical features are identified, the constraints of the wavelet transform as a decompositional tool can result in mismatches between the main lobe of the wavelet basis function and the chemical peak. Techniques to mitigate these effects are also discussed, and considerations are made for how to track features across multiple experiments.

Sensitivity of biomarkers to post-acquisitional processing parameters for in vivo brain MR spectroscopy signals

We present the results of a study to determine the sensitivity of biomarkers in in vivo brain MRS signals to post-acquisitional processing algorithms and parameters. Using a comprehensive integrated suite of post-processing and inference algorithms (BIDASCA) we examine the impact of different parameter values for model-based water suppression on the identification of statistically significant wavelet-based and model-based features. We observe that the number, location, and effect-sizes of significant features vary as a function of the resonance components estimated and removed during model-based water suppression, as well as their spectral proximity to the dominant water resonance itself. Moreover, the ordering of SVD modes in different signals is not uniform, making consistent suppression of water-based resonances problematic. Less than half of all significant features remained significant when water-suppression parameters were varied, which indicates that different parameter values can lead to unique markers and effect sizes. This study demonstrates the need for an end-to-end understanding of the sensitivity of biomarkers to processes that introduce uncertainty within the data, from acquisition to post-processing algorithms.