Individual variability of olfactory fMRI in normosmia and olfactory dysfunction

Thirty-eight subjects participated (mean age 46.6 years; 21 females, males) in this study. The cohort was divided into two groups according to their olfactory function. Of those, 22 were subjects (11 females) with diagnosed olfactory dysfunction, which was confirmed followed a standardized method [13] including standardized psychophysical testing of olfactory threshold, identification, and discrimination with the “Sniffin’ Sticks” test battery [14] patients’ characteristics were listed in Table 1. The rest of the participants consisted of 16 normosmic subjects (10 females). A detailed medical history was taken including conditions that potentially may have affected olfaction such as significant head trauma, chronic rhino sinusitis, neurological/endocrinological disorders, and previous nasal surgery. All included participants were in good health. Prior to inclusion in the study they received an otorhinolaryngological examination including nasal endoscopy. All participants were informed of the testing procedures and signed written informed consent before the experiment. The study design was in accordance with the Declaration of Helsinki and had been approved by the Ethics Committee of the Medical Faculty Carl Gustav Carus at the Technical University of Dresden.

All participants received birhinal olfactory testing using the validated and reliable "Sniffin' Sticks" battery (Burghart, Wedel, Germany). The test consists of measurements of odor sensitivity (phenyl ethyl alcohol odor detection thresholds), odor discrimination, and odor identification [14] First, odor threshold (T) is assessed using phenyl ethyl alcohol (a rose-like odor) presented by means of a single staircase, using stepwise dilutions in a row of 16 felt tip pens. Second, odor discrimination (D) is assessed by asking subjects to perform a three-alternative forced choice task using 16 pairs of odors. Third, odor identification (I) is assessed by asking subjects to identify 16 individual odors from a list of four verbal descriptors using a forced choice task. Scores from threshold (T), discrimination (D) and identification (I) were then added up to provide the TDI score The TDI score was the base for the diagnosis of olfactory dysfunction as well as its severity (e.g. hyposmia or anosmia). The TDI score ranges from 1 to 48, with a score above 30.5 being considered as normosmia, TDI score between 16 and 30.5 as hyposmia, and TDI score less than 16 as functional anosmia [15,16]

One odor was used as stimuli: “Coffee” odor (Frey und Lau, Henstedt-Ulzburg, Germany). Odorless air served as control stimulus. A block design was adopted for odor stimulation during fMRI measurements. During each functional run, each block lasted for 20 s, during one block, odorized air and was delivered to the participants’ bilateral nostril intermittently (8 s for odorized air, followed by 12 s of odorless air), in order to minimize adaptation to the odors which refers to a recent literature indicating larger BOLD signal in response to odor stimulation in shorter runs compared to longer but fewer runs [17], totally 12 blocks and lasting 240 s. Odorized and odorless air were delivered at a flow of 2l/min using a portable olfactometer.

A 3 T MRI scanner (Siemens Prisma, Erlangen, Germany) with a 32-channel head coil was used for image acquisition. For functional image collection, a spin echo/echo planar imaging (EPI) sequence was applied with echo time (TE) = 30 ms, repetition time (TR) = 2000 ms, flip angle = 90°, and in-plan resolution 2 mm. A T1-weighted structural scan was collected after functional runs with 121 slices, voxel size 1 × 1 × 1 mm³.

Imaging data were analyzed by means of SPM12 (Wellcome Trust Centre for Neuroimaging, London, UK) implemented in Matlab (version 2013a, Mathworks, Natick, MA, USA). Functional image volumes were pre-processed, starting with realignment and unwarp. In addition, the high-resolution T1 image was co-registered to the mean image of the EPI series for each participant. Co-registered T1-weighted MR images were segmented for gray and white matter to compute spatial transformation parameters for normalization. The registered functional scans were normalized to a standard Montreal Neurological Institute (MNI) template. Normalized images were spatially smoothed with an 8-mm full-width half-maximum Gaussian kernel. Finally, removal of head motion artifacts using ArtRepair (version 4, Stanford University) was applied to the preprocessed images based on the following rules: image to image motion less than 0.5 mm/TR and total images repaired less than 20%.

First, MRI scans were checked by an experienced radiologist for possible pathological brain anomalies. On the first level, the contrast “odor ON vs odor OFF” was modeled for each participant, in each block, we used the first 8 s coffee odor stimulation compared to the last 8 s odorless air (Fig. 1). The 1st level contrasts were submitted into second level analysis to show: (1) brain activation in patients and normosmic groups separately, and (2) different brain activation between the two groups. Analyses were focused on three regions of interest (ROIs): the primary olfactory cortex (POC), the olfactory orbitofrontal cortex and the insular cortex. The anatomical mask defined by Fjaeldstad et al.[8] including the piriform cortex and part of the amygdala, was used as the mask for POC. The olfactory OFC mask was built based on a 10-mm sphere centered on the right (x, y, z: 24, 36, −12) or left putative olfactory OFC (x, y, z: −22,32,−16) [17]. The insular cortex ROI was defined with the WFU_PickAtlas software (ANSIR, Wake Forest University, Winston-Salem, NC, USA)[18] . The template of the region for each ROI was showed in Fig. 2.

On an individual level, we conducted anatomical region-of-interest (ROI) analyses by extracting the mean % BOLD signal change (% BOLD signal change is often calculated using a baseline of the mean of the time series on a voxel, the MarsBaR software will calculate a baseline from the mean signal over a small ROI of many voxels) for each participant from all voxels in three ROIs [19] The anatomical ROI approach increases statistical power by averaging BOLD activation across multiple voxels within a specific brain region using the MarsBaR toolbox [20]. thereby reducing noise and the need to correct for multiple comparisons. Critically, anatomical ROIs are independent of the functional data analyzed, resulting in unbiased estimates of effect size [21].

Next, the spatial distribution of activation extent (defined as the number of significantly activated voxels) and intensity (defined as the cluster peak z value) were also extracted for individual participants from each ROI. An activation was considered significant when surviving to an uncorrected threshold of p < 0.05. The overall visualization and screenshots were performed using the software of MRIcroGL (https://​www.​cabiatl.​com/​mricrogl/​). The receiver operating characteristic (ROC) curve was constructed to test the ability of the % BOLD signal change within each of the three ROIs to discriminate between people with normal olfaction and patient with olfactory dysfunctions, defined based on their TDI scores. The classification accuracy was quantified using the area under the (AUC) value. All statistical analysis was performed by means of SPSS 23 (SPSS Inc., Chicago, IL, USA). Pearson correlation was performed between activations response of odor stimulation of selected ROIs of individuals and the Sniffin’ Sticks test scores. Chi-square analysis was used in calculating the proportion of effective activations response of odor stimulation of selected ROIs of individuals between normosmics and subjects with olfactory dysfunction.

Brain activations to odor stimulation in the normosmic group were shown in Fig. 3a, the local maxima of the main activated areas were found in the left POC (Montreal Neurological Institute [MNI] peak coordinates −26, −2, −18), insula (MNI peak coordinates −38, −8, 6 and 36, 0, 16), no obvious activations were found in the OFC (Fig. 3a). In subjects, the local maxima of the main activated areas were found in the left insula (MNI peak coordinates −36, −2, 18), no obvious activations were found in the POC and OFC regions (Fig. 3b). Between group comparison revealed stronger activation in the POC among normosmics as compared to subjects. Table 2 recaps maximum activation voxels and peak-z value responses to olfactory stimulation obtained from the the POC, OFC and insula regions in the group-level and found that the intensity of activation response to odor stimulation was much higher in the insular region than other ROIs.

Tables 3 and 4 show the % of BOLD signal change, cluster size (assessment of significant activation in fMRI) and peak-z value (normalized magnitude of peak voxel from statistical parametric map from publication) in responses to odor stimulation obtained from the three ROIs ROI for each participant. No significant correlation between TDI scores and the % of BOLD signal change, cluster size or peak-z value was found (p > 0.05).

Table 4 summarizes the cluster size and peak z-value of each ROIs for each participant. With the applied threshold (uncorrected p < 0.05), 93.8%, 62.5%, 62.5% of the normosmics showed detectable cluster (minimum 1 voxel) in the POC, OFC or insula, respectively. In the dysosmic group, the percentage of participants with observed activated clusters was 40.9%, 59.1% and 68.2% for the POC, OFC and insula, respectively. The proportion of activated clusters in POC in normosmic participants (93.8%) was significantly higher than that in subjects with olfactory dysfunction (40.9%) (X² = 14.9, P < 0.001) ROC analyses showed no significant AUC for % BOLD signal change for all three ROIs (POC: AUC = 0.53, p = 0.72; OFC: AUC = 0.58, p = 0.41; insula: AUC = 0.67, p = 0.07), meaning that it was not possible to distinguish between normosmics and subjects with olfactory dysfunction by means of the % BOLD signal change.