Automated Segmentation and Analysis of Membrane-Associated Microtubules

Tomograms were bandpass filtered using Bsoft. Segmentations of the filtered tomograms were guided by a tensor voting algorithm using TomoSegMemTV^{77 (link),78} . The parameters were optimised for each dataset, respectively. Clusters containing the IMC segmentations were manually extracted by visual analysis. The clusters were then converted to a 3D point cloud and further processed using the Open3D library⁷⁸ . Statistical outlier analysis was used to remove excess noise from the segmentations. Subsequently, the DBSCAN algorithm was used to separate individual membrane sections and the outer side of the IMC was selected manually for subsequent distance measurements. The angle φ_IMC was measured between two vectors for each SPMT particle: a vector of the SPMT particle X axis and a vector from the particle centre to the nearest segmented IMC coordinate (roughly equivalent to IMC normal vector intersecting the SPMT particle). Particles outside segmented membrane patches were excluded. Measurements from individual microtubules were reduced to a median value for plotting and statistical analyses.
The requirement for a clear membrane density for automated segmentation substantially reduced the number of microtubules available for analysis. Thus, due to the rarity of SPMTs in merozoites, doublets in gametocytes and minus termini in ookinetes, a small number of tomograms were segmented manually in IMOD.