This study was approved by the local institutional review board of The University of Akron. All authors declare that they have no competing interest.

MR images from pre-surgery CMI adult females (n = 236) and healthy controls (n = 140) were evaluated. Adult women were the primary focus of this analysis because a large majority of participants of the Chiari1000 were adult female. RCs within our sample consisted of syringomyelia (N = 38), pseudotumor cerebri (N = 20), EDS (N = 21), scoliosis (N = 45), and CCI (N = 17). APCs include CFS (N = 31), fibromyalgia (N = 33), Raynaud phenomenon (N = 18), migraine headaches (N = 147), spinal dysraphism (N = 29), other endocrine diseases (N = 20), and sleep apnea (N = 16). Identification of the presence of PCs (RCs and APCs) was obtained through self-reported clinical diagnosis from the Chiari1000 project.

Prevalence and demographic information for each RC and APC within our dataset are shown in Table 2. PCs with less than a 7% prevalence are not listed in Table 2. These PCs include: attention deficit, seizures, brain tumor, empty sella syndrome, stroke, basilar invagination, hydrocephalus, systemic lupus erythrematosis, tethered cord syndrome, Meniere's disease, atlantoaxial assimilation, multiple sclerosis, neurofibromatosis (excluding type I), connective tissue disorder (excluding EDS, Marfan, and MASS syndrome), growth hormone deficiency, abnormal pituitary, development delays, epilepsy, hyperactivity disorder, rickets, cystic fibrosis, achondroplasia, and concussion with internal bleeding. PCs presented in the table may have been identified at any point in the participant's life. Note, the PC termed “spinal dysraphism” exclude atlantoaxial assimilation, Klippel-Feil syndrome, and scoliosis. The PC termed “other endocrine diseases” excludes growth hormone deficiency, empty sella syndrome, and abnormal pituitary. PC denoted as seizures is a symptom where the cause of the seizures was not specified.

Consent for anonymized use of demographic data, health-related data, and MR images were provided by participants who contributed to the Chiari1000 project. The Chiari1000 project was created at The University of Akron in 2015 as a web-accessed database that obtains and stores participant information such as self-reported diagnosis, symptomatology, and cognitive and socio-emotional characteristics. Images were provided through online sharing (e.g., Dropbox), direct mail, or through participant release. An OsiriX DICOM PACS server (Pixmeo SARL, Geneva, Switzerland) was used to store anonymized MR images at The University of Akron. Inclusion of CMI participants was based on diagnosis of CMI by a physician and pre-surgical status of participant images. Pre-surgical status was determined using volunteered surgical and radiology records, patient disclosure, and inspection of MR images. CMI participants were excluded when surgery status could not be ascertained. Healthy female MR images and demographic information were obtained through the University of Minnesota, Washington University, Oxford University Human Connectome Project consortium (Van Essen et al., 2012), and Akron General Medical Center. No control participants presented with conditions that are known to affect brain morphology and or any other chronic illnesses.

DICOM MR images from 1.5T and 3T scanners that had been collected in the process of medical treatment were submitted by CMI patient participants to the Chiari1000 project. MMs were taken by four operators using participants' midsagittal MR images. Operators were blinded to CMI diagnosis. Midsagittal selection by each operator was standardized across all participants and completed through a visual inspection of specific morphological structures in each image to mitigate morphological changes due to head tilt. Visual inspection criteria consisted of the visibility of three out of four structures in the midsagittal plane: (1) the genu of the corpus callosum, (2) the splenium of the corpus callosum, (3) the pituitary infundibulum, and (4) the cerebral aqueduct. Images that failed to meet these criteria were excluded from further analysis. Where available, measurements were taken on T1-weighted images. As a result, measurements from all but ten participants were taken on a T1-weighted image. The measurements from the remaining ten participants were taken from a T2-weighted image.

MorphPro, a semi-automated software that was developed in MATLAB (Mathworks, Natick, MA) at The University of Akron, was used to measure 28 soft tissue and bone structures within the PCF compartment, craniocervical junction, oral cavity, and intracranial area using different linear, angular, and area measurements shown in Figure 1. These 28 measurements were selected in a previous study (Houston et al., 2017). Prior to the development of MorphPro, morphometric measurements were conducted manually using existing commercial DICOM readers such as OsiriX (Pixmeo SARL, Geneva, Switzerland). MorphPro provided the operator with a more efficient method for measuring specific features within a midsagittal image with real-time calculation of lengths, angles, and areas. Additionally, this software provided automatic data recording and storage of all morphometric measurements on the midsagittal image. In a separate study, Houston et al. (2017) demonstrated the reproducibility of results using MorphPro through the comparison of identical morphometric parameters measured using OsiriX software.

Measurement methodology was ratified by a board-certified neuroradiologist (JRB) with experience in neuroimaging and CMI through visual confirmation. Descriptions of all 28 measurements below are previously described in Houston et al. (2017) and are closely related to those described by Urbizu and colleagues (Urbizu et al., 2013, 2014, 2017a). Morphometric measurement number in parenthesis corresponds to number labels in Figure 1.

Morphometric measurement comparisons for 12 PCs were determined using two-tailed independent t-tests that compared participants with and without a specific PC. Measurement analysis for each of the 12 PCs was conducted separately (e.g., syringomyelia vs. no syringomyelia) due to the overlap of PC prevalence within the sample. Percentage of multiple conditions within our dataset can be seen in Table 3. To correct for family-wise error, a corrected probability value of 0.002 (0.05/28) was used to determine reliably different measures between the two groups (Tabachnick and Fidell, 2007). All analyses were performed in SPSS 24 (SPSS Inc., Chicago, IL) and Excel (Microsoft, Redmond, WA). For group comparisons where MRI age and/or BMI were found to be significantly different, a one-way analysis of covariance (ANCOVA) was used to control for the effects of age and/or BMI. Controlling for the demographic effects on morphology was motivated by previous findings (see Roller et al., 2015; Houston et al., 2017), which found that differences in morphology were reduced when controlling for demographic variables such as BMI and age. Mann-Whitney u-tests were used for group comparisons in which samples did not demonstrate a normal distribution.

In addition, we further tested our dataset by combining the many PCs into four groups. Three groups were previously defined by Milhorat et al. (2010) as CMI and cranial settling (CS), CMI and space-occupying intracranial lesions (SIL), and classical CMI. A fourth group was classified as CMI and endocrine diseases (ED) due to the high saturation of participants with ED in our sample. These groups are comprised of the following PCs: CS - EDS, Marfan syndrome, MASS, scoliosis, Klippel Feil syndrome, or CCI; SIL - hydrocephalus or brain tumors; and ED - growth hormone deficiency, abnormal pituitary, or other endocrine diseases. Classical Chiari participants do not have any of the three previously mentioned groupings of PCs.

No significant morphometric differences were observed between CMI participants with and without CS, SIL, or ED. Compared to controls, measurements from all four groups exhibited lower TP, and a smaller fastigium, pons, CC and clivus length. Additionally, a reduced PCF osseous area, a wider basal angle, and a narrower Wackenheim angle were identified compared to controls. MM statistical results of these large groupings can be seen in Table 6.

In our study, few MMs were found to distinguish between CMI participants with and without the examined PCs. McRae line length was the only MM that differed based on syringomyelia status and was smaller in CMI participants with syringomyelia. Additionally, McRae line length was not significantly different in syringomyelia patients than in controls. Previous case-control studies that evaluated both male and female CMI patients showed that McRae line length was larger in CMI patients than in healthy controls (Nishikawa et al., 1997; Milhorat et al., 1999; Karagöz et al., 2002; Aydin et al., 2005; Dagtekin et al., 2011; Houston et al., 2017). These findings suggest that a normal-sized McRae line length is unique to adult women CMI patients with syringomyelia. Additional follow-up studies may be necessary to examine whether CMI adult female patients with a normal McRae line length are at greater risk for syrinx development.

We found TP to be the only MM to distinguish between CMI patients with and without EDS. In contrast to this observation, 71% of the CMI patients diagnosed with HDCT from Milhorat et al. (2007) showed an increased retro-odontoid pannus formation with basilar impression. However, Milhorat et al. (2007) evaluated both male and female CMI patients, while our study only evaluated adult women. Additionally, due to a small occurrence of additional HDCT conditions within our dataset, we could not evaluate the remaining HDCT conditions discussed in Milhorat's et al (2007) study. Discrepancies between results from this previous study and our results may be due to the aforementioned sample differences. A final morphometric distinction was found between CMI patients with and without scoliosis, where basion to PAL length was significantly longer in CMI patients with scoliosis. However, to our knowledge, no prior morphological study has examined similar midsagittal 2D morphometrics in CMI patients with scoliosis. Thus this finding requires further investigation. Aside from syringomyelia, EDS, and scoliosis participants, no other significant morphometric differences were identified for CMI participants with any of the remaining RCs, or APCs in the present study.

When evaluating the etiological factor groups, no significant morphometric differences were found when comparing CMI participants with CS, SIL, or ED to classical CMI participants. Milhorat et al. (2010) compared only groups with CS, SIL and classical CMI participants to healthy controls. However, examination of their results indicated that differences may have been present and thus, motivated this additional test. The cause of these differences across studies is unclear and should be further examined in future studies.

Many MMs within our study were found to distinguish CMI patients with specific PCs from controls including four that were consistently different in all 12 examined PCs. These MM differences included a lower TP (9.5 mm), and shorter clivus, fastigium, and corpus callosum heights that differed by a small margin (2–5 mm). We previously identified similar differences in a subset 302 individuals from the present study (Houston et al., 2017). Eighteen MMs, including the four aforementioned MMs, were significantly different for all CMI participants compared to controls. In the present study, no specific PCs reflected this exact pattern of morphometric differences. This suggests that the four case-control MM differences found across PCs in the present study are the most robust identifiers of CMI. Alternatively, these varying morphometric results could also be attributed to differing demographic characteristics, and smaller samples sizes for the examined PCs.

In addition to these four MMs, CMI participants with EDS had a larger basion to PAL length, and a larger basal angle compared to controls. In a previous analysis of HDCT patients that incorporated EDS, Milhorat et al. (2007) found basion-dens interval and clivus-axial angle (Wackenheim angle) to be significantly different in CMI patients compared to controls. However, our analysis did not show differences in basion-dens interval or Wackenheim angle.

In an analysis of sub-groupings of CMI participants similar to those described by Milhorat et al. (2010), we found that classical CMI participants demonstrated a significantly smaller clivus length compared to controls, a finding that replicates Milhorat et al. (2010). However, no case-control differences were observed for McRae line length. In addition, Milhorat et al. (2010) found no MMs that distinguish between CS or SIL and controls. In contrast, we did find significantly smaller clivus lengths for CS, SIL, and ED participants compared to controls. This discrepancy may be due to the inclusion of male participants, in addition to the MMs being evaluated on CT images in Milhorat et al. (2010). Therefore, by excluding male CMI participants from our MRI analysis, we were able to better identify unique morphology for adult female participants with specific etiological factors.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.