Twenty-two patients with CO intoxication (7 men and 15 women; mean age, 41.09 ± 9.25 years; range, 28–56 years) in the chronic phase (> 6 months)  and 15 age-, sex-, and education-matched healthy subjects (7 men and 8 women; mean age, 36.27 ± 10.41 years; range, 24–58 years) were recruited. Subjects received MRI studies and NP tests in the chronic stage in the CO intoxication group and at the time of enrollment in the control group. The mean duration between acute exposure and the acquisition time of MRIs for all patients was 25.05 ± 15.20 months.
Patients with a CO intoxication history were included in the study. A clear history of acute CO intoxication was defined as an episode of past exposure to burning charcoal or gas in an enclosed space and/or an elevated COHb level . These patients either sought first aid at the emergency room of Chang Gung Memorial Hospital during acute CO intoxication and follow-up at the out-patient clinic (n = 17) or developed new symptoms after acute CO intoxication at the out-patient clinic (n = 5). The average initial COHb level of the patients (n = 17) was 15.95 ± 15.03% (range: 0.9-54.3%, mean of DE group: 22.8%, mean of non-DE group: 12.5%). All CO intoxicated patients awoke within 24 h and underwent hyperbaric oxygen therapy for several days. During the acute stage, 17 out of 22 patients underwent conventional MRI study. The exclusion criteria for this study included a history of neurologic or psychiatric illness, the presence of developmental disorders, the use of medication for unrelated conditions, and head injuries, which can affect the results of the neuropsychiatric or neuroimaging surveys .
The 22 CO intoxicated patients were further divided into two subgroups, based on the presence or absence of DE (11 in the non-DE group; 4 men and 7 women; mean age, 40.55 ± 9.45 years and 11 in the DE group; 3 men and 8 women; mean age, 41.64 ± 9.47 years), which was defined as a combination of events such as an initial change in consciousness due to CO exposure, recovery from the acute stage, lack of symptoms for periods of days to weeks, and exacerbation with neurologic and/or psychiatric symptoms .
The Chang Gung Memorial Hospital Ethics Committee approved the study and all patients and participants in the control group provided written informed consent.
Neuropsychological (NP) tests
Patients and healthy subjects were administered subtests of the Wisconsin card sorting test (WCST) and the Wechsler Adult Intelligence Scale (WAIS).
Wisconsin card sorting test (WCST)
The WCST is commonly used to evaluate frontal executive function, such as concept formation, set shifting, and flexibility . In this study, the WCST-128 card computerized version was administered by trained research assistants to decrease the complexity of the administered WCST and to increase the efficiency of data collection. The six WCST indices used for the analysis were perseverative response (PR), perseverative error (PE), non-perseverative error (NPE), percent conceptual level response, completed categories, and failure to maintain set .
Wechsler Adult Intelligence Scale (WAIS)
The WAIS, a family of tests created by David Wechsler to measure cognitive domains that contribute to intelligence (Wechsler, 1955, 1981, 1997), is used to assess a wide range of cognitive abilities and impairments. In this study, we used the full scale intelligence quotient measure from the Taiwanese version of the WAIS-III [21, 22], which is based on the combined Verbal Comprehension Index (VCI), Perceptual Organization Index (POI), Working Memory Index (WMI), and Processing Speed Index (PSI) scores . All of the participants finished picture completion and matrix reasoning, the subtests that comprised the POI, and the digit symbol and symbol search, which comprised the PSI.
Magnetic Resonance Imaging (MRI) data acquisition
Magnetic resonance scanning was performed on a 3 T MRI system (Excite; GE Medical System) equipped with an 8-channel head coil. High resolution T1-weighted images were acquired parallel to the anterior commissure-posterior commissure line (AC-PC line), using 3-dimensional fluid-attenuated inversion-recovery fast spoiled gradient echo (3D FLAIR-FSPGR) sequences. The parameters were TR 9.492 ms, TE 3.888 ms, TI 450 ms, flip angle 20°, field of view (FOV) 24 × 24 cm, matrix size 512 × 512, 110 continuous slices with a slice thickness of 1.3 mm, and an in-plane spatial resolution of 0.47 × 0.47 mm.
Imaging data processing
A T1 VBM approach based on Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra (DARTEL) was used for preprocessing and subsequent analyses of whole brain T1-weighted volumetric images [24, 25]. Individual T1-weighted volumetric images were analyzed using the Gaser’s VBM8 toolbox (http://dbm.neuro.uni-jena.de/vbm/) with SPM8 (Statistical Parametric Mapping. Wellcome Department of Imaging Neuroscience, London, UK; available online at http://www.fil.ion.ucl.ac.uk/spm), implemented in Matlab 7.3 (MathWorks, MA, USA). DARTEL is a novel image registration method that uses large deformation in an inverse-consistent framework for spatial normalization in the SPM toolbox.
Briefly, all images were carefully checked by an experienced neuroradiologist to ensure that no scanner artifacts, motion problems, or gross anatomic abnormalities existed for each participant. The semiautomatic approach  was used to evaluate the quality of structural images, specifically for motion problems. Motion artifact was evaluated on individual tissue segment images and assigned a rating of none, mild, moderate, or severe. To reduce bias during VBM processing, images were excluded if their tissue segment images were given a rating of moderate or severe. The anterior commissure was set as the origin of imaging for each participant.
Whole brain native space T1-weighted images were normalized and the bias field corrected and segmented into GM, WM, and cerebro-spinal fluid (CSF) partitions, based on the same generative model . Unified segmentation involved alternating between segmentation, bias field correction, and normalization to obtain local optimal solutions for each process. This procedure was further refined by applying an iterative hidden Markov field (HMRF) model  to improve the quality of tissue segmentation and minimize the influence of noise level.
To ensure accuracy of registration across subjects, the native space GM, WM, and CSF segments were imported into a rigidly aligned space and iteratively registered to group-specific templates generated from all images in this study through nonlinear warping using the DARTEL toolbox . Recent studies have indicated that the DARTEL algorithm can improve intersubject registration and be useful for population-based research [29, 30].
The deformation parameters obtained in the spatial normalization step were applied to individual tissue segments in a rigidly aligned space. The Jacobian determinants derived from nonlinear deformation for the correction of volume changes were also applied during the nonlinear spatial transformation to preserve the overall amount of each tissue segment after normalization. Since DARTEL worked on images with average brain size of total participants in this study, additional affine transformation between average group space and Montreal Neurological Institute ( MNI ) standard space was needed. Because the MNI standard space was constructed by affine registration of a number of subjects to a common standard coordinate system, it was reasonable to use only affine transformation to achieve a suitable alignment between these two spaces. The optimal-normalized tissue segments of each individual had an identical voxel size of 1.5 × 1.5 × 1.5 mm.
All normalized, segmented, and modulated MNI standard-space images were smoothed with an 8-mm Gaussian kernel prior to tissue volume calculation and voxel-wise group comparisons. Overall tissue volumes (i.e., GM, WM, and CSF) were estimated in mm3 by counting the voxels representing GM, WM, and CSF in standard space. The total intracranial volume (TIV) was determined as the sum of the three volumes.
Analysis between groups
Statistical analysis was performed using the statistics computer software SPSS 12 (SPSS Inc, Chicago, IL). All data were given as the mean ± standard deviation (SD). The demographic and clinical characteristics of those in the patient and control groups were compared by analysis of variance (ANOVA) (for age and education). One-way analysis of covariance (ANCOVA) was used to compare TIV, GM volume, and WM volume, with age and sex as added covariates. Statistical differences in NP data, including the WCST and WAIS results between the two groups, were estimated by ANCOVA, with age, sex, and education as covariates. The threshold for statistical significance was p < 0.05.
Smoothed, modulated gray matter segments were analyzed with SPM8 within the framework of a General Linear Model (GLM). ANCOVA was performed with the covariation of age, sex, and TIV to investigate differences in regional GM volume between the two groups. All voxels with a GM probability value < 0.2 (range, 0–1) were eliminated to avoid possible partial volume effects around the margin between the GW and the WM. Nonstationary correction (part of the VBM toolbox) for correcting non-isotropic smoothness of the data was used to investigate group differences . The differences in GM volume were compared between the following groups: 1) all patients vs. control group, 2) non-DE group vs. control group, 3) DE group vs. control group, and 4) non-DE group vs. DE group.
Because of the exploratory design of this study, strict criteria were used to obtain the findings. Low voxel-level thresholds (uncorrected p < 0.05) might sensitize the cluster inference for spatially extended and lower spatial resolutions. In contrast, high (uncorrected p < 0.001) contiguous voxel thresholds with a cluster size >50 might generate a higher spatial cluster resolution but result in the loss of spatial extent. In this study, the voxel-level threshold was set to an uncorrected p < 0.001 and a nonstationary cluster extent threshold of p < 0.05 corrected for multiple comparisons with family-wise error (FWE)  correction in to obtain precise findings with higher spatial cluster resolutions.
To minimize coordinate transformation discrepancies between the MNI and Talairach space, GingerALE provided by BrainMap (The BrainMap Development Team; available online at http://brainmap.org/ale/index.html) was used to transform MNI coordinates into Talairach coordinates. Anatomic structures of the coordinates representing significant clusters were identified based on the Talairach and Tournoux atlas .
Partial correlation analysis adjusted for age, sex, education, and TIV was performed to assess the correlation between NP test scores and areas with smaller volumes in all CO intoxicated patients compared to those in the control group. Regional GMV was extracted from the peak coordinate and correlated with NP test scores, with significance at p < 0.05 .