Evaluating spatial normalization for SVM-based EEG decoding: A within- and between-subjects perspective

ObjectiveWe developed a new computed tomography (CT)-based spatial normalization method and CT template to demonstrate its usefulness in spatial normalization of positron emission tomography (PET) images with [18F] fluorodeoxyglucose (FDG) PET studies in healthy controls.Materials and MethodsSeventy healthy controls underwent brain CT scan (120 KeV, 180 mAs, and 3 mm of thickness) and [18F] FDG PET scans using a PET/CT scanner. T1-weighted magnetic resonance (MR) images were acquired for all subjects. By averaging skull-stripped and spatially-normalized MR and CT images, we created skull-stripped MR and CT templates for spatial normalization. The skull-stripped MR and CT images were spatially normalized to each structural template. PET images were spatially normalized by applying spatial transformation parameters to normalize skull-stripped MR and CT images. A conventional perfusion PET template was used for PET-based spatial normalization. Regional standardized uptake values (SUV) measured by overlaying the template volume of interest (VOI) were compared to those measured with FreeSurfer-generated VOI (FSVOI).ResultsAll three spatial normalization methods underestimated regional SUV values by 0.3-20% compared to those measured with FSVOI. The CT-based method showed slightly greater underestimation bias. Regional SUV values derived from all three spatial normalization methods were correlated significantly (p < 0.0001) with those measured with FSVOI.ConclusionCT-based spatial normalization may be an alternative method for structure-based spatial normalization of [18F] FDG PET when MR imaging is unavailable. Therefore, it is useful for PET/CT studies with various radiotracers whose uptake is expected to be limited to specific brain regions or highly variable within study population.

BackgroundSpatial normalization is a prerequisite step for analyzing positron emission tomography (PET) images both by using volume-of-interest (VOI) template and voxel-based analysis. Magnetic resonance (MR) or ligand-specific PET templates are currently used for spatial normalization of PET images. We used computed tomography (CT) images acquired with PET/CT scanner for the spatial normalization for [18F]-N-3-fluoropropyl-2-betacarboxymethoxy-3-beta-(4-iodophenyl) nortropane (FP-CIT) PET images and compared target-to-cerebellar standardized uptake value ratio (SUVR) values with those obtained from MR- or PET-guided spatial normalization method in healthy controls and patients with Parkinson’s disease (PD).MethodsWe included 71 healthy controls and 56 patients with PD who underwent [18F]-FP-CIT PET scans with a PET/CT scanner and T1-weighted MR scans. Spatial normalization of MR images was done with a conventional spatial normalization tool (cvMR) and with DARTEL toolbox (dtMR) in statistical parametric mapping software. The CT images were modified in two ways, skull-stripping (ssCT) and intensity transformation (itCT). We normalized PET images with cvMR-, dtMR-, ssCT-, itCT-, and PET-guided methods by using specific templates for each modality and measured striatal SUVR with a VOI template. The SUVR values measured with FreeSurfer-generated VOIs (FSVOI) overlaid on original PET images were also used as a gold standard for comparison.ResultsThe SUVR values derived from all four structure-guided spatial normalization methods were highly correlated with those measured with FSVOI (P < 0.0001). Putaminal SUVR values were highly effective for discriminating PD patients from controls. However, the PET-guided method excessively overestimated striatal SUVR values in the PD patients by more than 30% in caudate and putamen, and thereby spoiled the linearity between the striatal SUVR values in all subjects and showed lower disease discrimination ability. Two CT-guided methods showed comparable capability with the MR-guided methods in separating PD patients from controls and showed better correlation between putaminal SUVR values and the parkinsonian motor severity than the PET-guided method.ConclusionCT-guided spatial normalization methods provided reliable striatal SUVR values comparable to those obtained with MR-guided methods. CT-guided methods can be useful for analyzing dopamine transporter PET images when MR images are unavailable.

Chapter 15 - Surface-Based Spatial Normalization Using Convex Hulls

P300 is a typical event related potential (ERP), which has been applied in the brain-computer interface (BCI) and attracted the attention of researchers for nearly thirty years. Until now, the most challenge of P300-based BCI is to detect P300 in a minimum of repeats, which is a balance between recognition accuracy and a number of repeats. Previous studies showed that the shallow learning model such as support vector machine (SVM), relevance vector machine (RVM) and linear discriminant analysis (LDA) had been successfully used for P300 classification. In this study, we proposed a P300 recognition algorithm based on ensemble of SVMs. Firstly, we intercept 600 ms data after the visual stimulation. Secondly, the extracted segment electroencephalogram (EEG) was averaged by iteration to enhance the signal-to-noise ratio. Thirdly, these obtained signals were fed to the ensemble of SVMs for target character is recognition. Finally, an output regulation mechanism was proposed for adjusting the outputs of ensemble of SVMs. The experimental results show that the highest accuracy (80%) and the information transmission rate (17.21 bits/min) are observed in the repeat number of only five. As an application and verification, this algorithm won the third prize in the BCI Controlled Robot Contest of 2019 World Robot Conference.

Brain-computer interfaces (BCI) based on P300 event-related potentials (ERP) could help to select characters from a visually presented character-matrix. They provide a communication channel for users with neurodegenerative disease. Associated to these kinds of BCI systems, there is the problem to determine whether or not a P300 was actually produced in response to the stimuli. The design of this classification step involves the choice of one or several classification algorithms from many alternatives. Support Vector Machines (SVM) and Linear Discriminant Analysis (LDA) have been used to achieve acceptable results in numerous P300 BCI applications. However, both of them suffers from the high dimensional problem which leads to deterioration of their performance. In this paper, we introduce a novel and combined approach of LDA and SVM to reduce the negative effect of high dimensional data on SVM and LDA, and investigate the performance of our method. The results shows that the new approach achieves similar or slightly better performance than the state-of-art method.

We applied event-related potential (ERP) to reinforcement signals that are equivalent to reward and punishment signals. We conducted an experiment using an electroencephalogram (EEG) in which volunteers identified the success or failure of an inverted pendulum task. We confirmed that there were differences in the EEG signal depending on whether the task was successful or not and that ERP might be used as a punishment of reinforcement learning. We used a support vector machine (SVM) for recognizing the ERP. We selected the feature vector in SVM that was composed of averages of each 35 msec for each of three channels (F3,Fz,F4) on the frontal area, for a total of 700 msec. Our experimental results suggest that reinforcement learning using ERP can be performed accurately. Finally, we suggest the possibility of developing an adaptive brain-computer interface (BCI) by ERP.

Multi-variable time series (MTS) information is a typical type of data inference in the real world. Every instance of MTS is produced via a hybrid dynamical scheme, the dynamics of which are often unknown. The hybrid species of this dynamical service are the outcome of high-frequency and low-frequency external impacts, as well as global and local spatial impacts. These influences impact MTS’s future growth; hence, they must be incorporated into time series forecasts. Two types of normalization modules, temporal and spatial normalization, are recommended to accomplish this. Each boosts the original data’s local and high-frequency processes distinctly. In addition, all components are easily incorporated into well-known deep learning techniques, such as Wavenet and Transformer. However, existing methodologies have inherent limitations when it comes to isolating the variables produced by each sort of influence from the real data. Consequently, the study encompasses conventional neural networks, such as the multi-layer perceptron (MLP), complex deep learning methods such as LSTM, two recurrent neural networks, support vector machines (SVM), and their application for regression, XGBoost, and others. Extensive experimental work on three datasets shows that the effectiveness of canonical frameworks could be greatly improved by adding more normalization components to how the MTS is used. This would make it as effective as the best MTS designs are currently available. Recurrent models, such as LSTM and RNN, attempt to recognize the temporal variability in the data; however, as a result, their effectiveness might soon decline. Last but not least, it is claimed that training a temporal framework that utilizes recurrence-based methods such as RNN and LSTM approaches is challenging and expensive, while the MLP network structure outperformed other models in terms of time series predictive performance.

Brain-computer interfaces (BCI) provide direct and non-muscular communication methods for the people with severe motor impairments. Event-related potentials (ERPs) as efficient modals are commonly used in some of the BCI systems, including visual stimulus, auditory stimulus as well as tactile stimulus. In this experiment, the corresponding Chinese pronunciations were inserted into the visual Oddball series of 1–9 numbers to carry out the cross-sense stimuli of BCI. The experimental data analysis result proves that the P300 components produced by visual-auditory associate stimulation have higher amplitudes and shorter latencies than those produced by visual-only stimulus. For further analysis the constrained independent component analysis (cICA) method was applied when extracting the signal features of ERP and the support vector machine (SVM) method was used to BCI classification. Result proves that the ERPs produced by visual-auditory associate stimulation modal have better recognition efficiency than those in visual-only stimulation. It can relevant the capacity of information alteration in BCI and is worth to do more studies.

IntroductionSpatial normalization is a prerequisite step for the quantitative analysis of SPECT or PET brain images using volume-of-interest (VOI) template or voxel-based analysis. MRI-guided spatial normalization is the gold standard, but the wide use of PET/CT or SPECT/CT in routine clinical practice makes CT-guided spatial normalization a necessary alternative. Ventricular enlargement is observed with aging, and it hampers the spatial normalization of the lateral ventricles and striatal regions, limiting their analysis. The aim of the present study was to propose a robust spatial normalization method based on CT scans that takes into account features of the aging brain to reduce bias in the CT-guided striatal analysis of SPECT images. MethodsWe propose an enhanced CT-guided spatial normalization pipeline based on SPM12. Performance of the proposed pipeline was assessed on visually normal [123I]-FP-CIT SPECT/CT images. SPM12 default CT-guided spatial normalization was used as reference method. The metrics assessed were the overlap between the spatially normalized lateral ventricles and caudate/putamen VOIs, and the computation of caudate and putamen specific binding ratios (SBR). ResultsIn total 231 subjects (mean age ± SD = 61.9 ± 15.5 years) were included in the statistical analysis. The mean overlap between the spatially normalized lateral ventricles of subjects and the caudate VOI and the mean SBR of caudate were respectively 38.40 % (± SD = 19.48 %) of the VOI and 1.77 (± 0.79) when performing SPM12 default spatial normalization. The mean overlap decreased to 9.13 % (± SD = 1.41 %, P < 0.001) of the VOI and the SBR of caudate increased to 2.38 (± 0.51, P < 0.0001) when performing the proposed pipeline. Spatially normalized lateral ventricles did not overlap with putamen VOI using either method. The mean putamen SBR value derived from the proposed spatial normalization (2.75 ± 0.54) was not significantly different from that derived from the default SPM12 spatial normalization (2.83 ± 0.52, P > 0.05). ConclusionThe automatic CT-guided spatial normalization used herein led to a less biased spatial normalization of SPECT images, hence an improved semi-quantitative analysis. The proposed pipeline could be implemented in clinical routine to perform a more robust SBR computation using hybrid imaging.

One of the most interesting clinical applications of 18F-FDG PET imaging in neurodegenerative pathologies is that of establishing the prognosis of patients with mild cognitive impairment (MCI), some of whom have a high risk of progressing to Alzheimer's disease (AD). One method of analyzing these images is to perform statistical parametric mapping (SPM) analysis. Spatial normalization is a critical step in such an analysis. The purpose of this study was to assess the effect of using different methods of spatial normalization on the results of SPM analysis of 18F-FDG PET images by comparing patients with MCI and controls. We evaluated the results of three spatial normalization methods in an SPM analysis by comparing patients diagnosed with MCI with a group of control subjects. We tested three methods of spatial normalization: MRI-DARTEL and MRI-SPM8, which combine structural and functional images, and FDG-SPM8, which is based on the functional images only. The results obtained with the three methods were consistent in terms of the main pattern of functional alterations detected; namely, a bilateral reduction in glucose metabolism in the frontal and parietal cortices in the patient group. However, MRI-SPM8 also revealed differences in the left temporal cortex, and MRI-DARTEL revealed further differences in the left temporal cortex, precuneus, and left posterior cingulate. The results obtained with MRI-DARTEL were the most consistent with the pattern of changes in AD. When we compared our observations with those of previous reports, MRI-SPM8 and FDG-SPM8 seemed to show an incomplete pattern. Our results suggest that basing the spatial normalization method on functional images only can considerably impair the results of SPM analysis of 18F-FDG PET studies.

As the size of the neuroimaging cohorts being increased to address key questions in the field of cognitive neuroscience, cognitive aging, and neurodegenerative diseases, the accuracy of the spatial normalization as an essential preprocessing step becomes extremely important. Existing spatial normalization methods have poor accuracy particularly when dealing with the highly convoluted human cerebral cortex and when brain morphology is severely altered (e.g., aging populations). To address this shortcoming, we propose a novel spatial normalization technique that takes advantage of the existing surface‐based human brain parcellation to automatically identify and match regional landmarks. To simplify the nonlinear whole brain registration, the identified landmarks of each region and its counterpart are registered independently with topology‐preserving deformation. Next, the regional warping fields are combined by an inverse distance weighted interpolation technique to have a global warping field for the whole brain. To ensure that the final warping field is topology‐preserving, we used simultaneously forward and reverse maps with certain symmetric constraints to yield bijectivity. We have evaluated our proposed solution using both simulated and real (structural and functional) human brain images. Our evaluation shows that our solution can enhance structural correspondence compared to the existing methods. Such improvement also increases the sensitivity and specificity of the functional imaging studies, reducing the required number of subjects and subsequent study costs. We conclude that our proposed solution can effectively substitute existing substandard spatial normalization methods to deal with the demand of large cohorts which is now common in clinical and aging studies.

Evaluation of MR-driven versus MR-free PET/CT spatial normalization approaches for quantification of regional uptake in Dementia.

Spatial normalization is a preliminary step in any PET image analysis based on statistical parametric mapping (SPM). This step consists in applying a spatial deformation to match each PET scan to an anatomical reference template The purpose of this study was to evaluate the effect of using different methods of spatial normalization on the results of SPM analysis of 18F-FDG PET images by comparing controls and patients diagnosed with mild cognitive impairment (MCI) that converted to probable Alzheimer’s disease (AD) after two years of follow-up. We performed an SPM analysis between the two groups using three spatial normalization methods: 1) MRI-DARTEL 2) MRI-SPM8 and 3) FDG-SPM8. MRI-DARTEL and MRI-SPM8 combine structural and functional images, while FDG-SPM8 is based only on functional images. The results obtained with the three methods were consistent in terms of the pattern of hypometabolism detected in the patient group. However, MRI-DARTEL was the method more consistent with the patterns previously reported in the literature. These results suggest that MRI-DARTEL is the most accurate and powerful method for spatial normalization in SPM analysis of 18F-FDG PET images. Normalization based solely on functional imaging shows less sensitivity to detect significant differences.

Spatial normalization is frequently used to map data to a standard coordinate system by removing intersubject morphological differences, thereby allowing for group analysis to be carried out. The work presented in this paper is motivated by the need for an automated cortical surface normalization technique that will automatically identify homologous cortical landmarks and map them to the same coordinates on a standard manifold. The geometry of a cortical surface is analyzed using two shape measures that distinguish the sulcal and gyral regions in a multiscale framework. A multichannel optical flow warping procedure aligns these shape measures between a reference brain and a subject brain, creating the desired normalization. The partial differential equation that carries out the warping is implemented in a Euclidean framework in order to facilitate a multiresolution strategy, thereby permitting large deformations between the two surfaces. The technique is demonstrated by aligning 33 normal cortical surfaces and showing both improved structural alignment in manually labeled sulci and improved functional alignment in positron emission tomography data mapped to the surfaces. A quantitative comparison between our proposed surface-based spatial normalization method and a leading volumetric spatial normalization method is included to show that the surface-based spatial normalization performs better in matching homologous cortical anatomies.

Background:&#13;\nThe event related potential (ERP) is an important electrophysiological response to an internal or external stimulus on human body. In some studies, the ERP-based brain computer interface (BCI) systems were created in visual or auditory modality.&#13;\n&#13;\nHowever, in these type of BCIs, either the eyes or ears of the users are occupied when they are making a choice. It is not convenient to communicate with others. Thus, a somatosensory ERP based BCI can be developed to overcome this issue. According to this, the analysis of somatosensory ERP features is necessary to evaluate if somatosensory ERP is eligible for BCIs as an input.&#13;\n&#13;\nObjective:&#13;\n1. To study ERP features and design of P300 experiment.&#13;\n2. To compare three types of P300 features elicited by three modalities.&#13;\n3. To produce ERP response by electrical stimuli delivered to different position, and analyze ERP features.&#13;\n&#13;\nMethods:&#13;\nTwo experiments were conducted. In experiment 1, three modalities, including visual, auditory and electrical modality, were used to produce P300 response. Experiment 2 only presented electrical stimuli. In experiment 1 two electrical stimuli were presented with different intensities at one location, whereas four electrical stimuli were showed at different location with the same intensity. The amplitude and latency were compared among three modalities, and the ERP topography of experiment 2 was also analyzed.&#13;\n&#13;\nResult and conclusion:&#13;\nFourteen subjects’ data were analyzed in our study. The amplitude and latency of electrical P300 were similar to auditory ERP. But the ERP of visual modality had the largest amplitude and shortest latency. This result shows that electrical P300 can work as well as auditory P300 in BCIs, but not as good as visual P300. In experiment 2, the latency of electrical ERP occurred around 280 ms, and the amplitude and the topography showed that the largest amplitude was located around Cz electrode. This type of ERP in experiment 2 was considered as P3a, which also can be used in BCI systems.