Fig 1.
Results display from GraFT-App.
(A) The distinguishable colors ROIs figure shows the overlay of all of the dictionary elements found, with their color matching their time trace in (B).
Fig 2.
Comparison of ASM (MPC) and IPM (quadprog) solvers.
(A) Runtime of solvers normalized by window size (pixel area). (B) Sample ROIs spatial results of the NeuroFinder dataset between both solvers. (C) Comparison of runtime between MPC and quadprog solvers by varying dictionary elements over 50x50 pixel area using patchGraFT. (D) Sample time trace of the same ROI (#3) from IPM and ASM solvers.
Fig 3.
Effects of time compression on temporal components identified by non-normalized and normalized GraFT.
Histogram of highest time trace correlations to the NAOMi ground truth for (A) non-normalized GraFT and (B) normalized GraFT. Heatmap for (C) non-normalized and (D) normalized GraFT results across compressions levels. Each entry is the percentage of traces in the row-wise compression result that correlated at with the column-wise result. Example highly correlated traces (
for all compression levels) in (E) non-normalized GraFT and (F) normalized GraFT.
Fig 4.
Effects of time compression on spatial profiles identified by non-normalized and normalized GraFT.
(A) Plot of spatial profiles comparing uncompressed GraFT and 64 × , non-normalized GraFT which had highly correlated time traces () and medium correlated time traces to the ground truth (
). (B) Plot of all spatial profiles with corresponding time trace correlations
.(C) Spatial profile comparison within a defined ROI between non-normalized,
, 64 × compressed GraFT and normalized, 64 × GraFT. The non-normalized ROI was post-proceesed using different filtering parameters to improve visualization, while the normalized ROI used the default visualization parameters across different sparsity λ coefficients. (D) Comparison of non-normalized,
, GraFT (top) and normalized GraFT,
, GraFT (bottom) spatial profiles comparing to the time traces shown in Fig 3.
Fig 5.
Time speedup and RAM allocation improvement by compression level.
(A) Relative time speedup as a function of compression. (B) Relative RAM usage as a function of compression.
Fig 6.
GraFT algorithm results on calcium imaging of pial arterioles.
(A) Spatial profiles found along with an overlay of all five ROIs with distinguishable colors. (B) Respective time traces of the spatial profiles detected. (C) Component spatial and temporal linear correlations using Pearson’s correlation coefficient (r).
Fig 7.
Results of GraFT on axonal imaging.
(A) Mean image (left) and four identified components (right) show that GraFT finds networks of axons across the full field-of-view. (B) The four components shown separately for clarity. (C) Example component time traces, color coded to match panel B. Local temporal averages for three time-points, each with the components displayed with intensity weighted by the respective integrated fluorescence over that time-window, and the local video average of the same time window show that the full axonal group is active when the components have high activity levels. (D) Suite2p results on the same data breaks up the axons into multiple segments, similar to dendritic imaging [18].
Table 1.
Parameters used for experimental results between the three different sample datasets analyzed. Highlighted in gray are parameters that changed across datasets.