Cylarim overview
Cylarim is an innovative software package for laminar fMRI analysis that addresses a major limitation of many existing approaches: their reliance on pre-defined regions of interest (ROIs). The definition of ROIs often involves arbitrary choices, making them difficult to justify and prone to bias. To overcome this problem, Cylarim utilizes small, overlapping cylinders that traverse the cortical ribbon, enabling a comprehensive mapping of laminar-specific activity across extensive cortical regions.
Initial steps
Cylarim assumes the availability of high-precision segmentations of grey matter, white matter, and cerebrospinal fluid (CSF) with a spatial resolution of approximately 0.2mm. Data with lower resolution should be upsampled accordingly.
The initial processing steps are similar to the LayNii software package (Huber et al., 2021). Specifically, we extract cortical rims delineating the boundaries between CSF and white matter. Subsequently, we derive a normalized ‘metric’ image representing cortical depth, scaled to the range [0,1]. Cylrim offers two programs for this purpose: vrim and vmetric. The outputs of LayNii may be used as well. Note that it is mandatory that the resulting “rim” image uses the following codes: 1(CSF-GM boundary), 2(WM-GM boundary), 3(GM interior), with GM(grey matter), WM(white matter), and CSF(cerebral spinal fluid).
The main steps
vcylarim is the core tool in this package. It first constructs cortical cylinders by connecting pairs of voxels: one on the gray matter GM/WM rim and the other on the GM/CSF rim. Voxel pairs are identified by finding the nearest neighbor on the opposite rim for each voxel. After excluding redundant pairs, cylinders are defined by dilating the line connecting each pair of voxels with a user-specified radius (typically 2mm).
After the cylinders have been constructed, vcylarim analyses activation profiles using various methods. For example, vcylarim can be instructed to compute three depth strata (deep, middle, superficial), and perform pairwise t-tests to compare them. Alternatively, it can be instructed to identify peaks in the cortical profiles, or identify regions where the profile is concave up or down. The output of vcylarim is a set of files depending on the various type of analysis selected by the user.
The output files are 4D images that can be further processed using the program vcylarim_probe.
The input images can be either in Lipsia’s own ‘vista’-format (.v), or in nifti-format (.nii). The output images are always in ‘vista’-format. They can be visualized using the viewer vini (github.com/MPI-Neuroinformatics/vini), or converted to nifti using the program vnifti and visualized using other neuroimaging tools.
Some typical workflows are shown below.
Example workflow 1
vrim -in segmentation.v -out rim.v vmetric -in rim.v -out metric.v vcylarim -in zmap.v -rim rim.v -metric metric.v -radius 2 -3bins true -nperm 1000 -out cyl vcylarim_probe -in cyl_3bins_zvals.v -out result.v -type bottom_m vfdr -in result.v -out result_fdrcorrected.v -alpha 0.01 vnifti -in result_fdrcorrected.v -out result_fdrcorrected.nii
In this example, the cortical ribbon is divided into three depth strata (‘deep’, ‘middle’, ‘superficial’), and two-sample tests are performed to obtain three contrasts: deep-middle, deep-superficial, and middle-superficial. Z-values uncorrected for multiple comparisons are obtained using permutation tests by randomly reassigning cortical depth values.
The program vcylarim_probe then identifies predominant deep layer activations (‘-type bottom_m’) using a conjunction analysis in which the conjunction between the contrasts ‘deep-middle’ and ‘deep-superficial’ is computed. Finally, the resulting map is corrected for the false discovery rate (FDR) using a significance level of 0.01. The output image can be visualized either using the program vini or converted to nifti and visualized using other imaging software.
Example workflow 2
vrim -in segmentation.v -out rim.v vmetric -in rim.v -out metric.v vcylarim -in zmap.v -rim rim.v -metric metric.v -radius 2 -peaks true -out cyl vnifti -in cyl_peaks_coeff.v -out cyl_peaks_coeff.nii
In this example, the locations of peaks and valleys are estimated in each cylinder. It involves fitting a fourth-order Chebyshev polynomial to the z-values within each cylinder using a least-squares approach. The largest and smallest points of the fitted curve within the interval [0,1] are then identified and reported as the output. The resulting file “cyl_peaks_coeff.v” contains two volumes. The first volume shows the location of valleys, the second shows the location of peaks.
Reference:
Lohmann, G. et al (2025), bioRxiv, Beyond ROIs: Laminar fMRI Mapping with Cylarim, https://doi.org/10.1101/2025.03.25.645237