def cellMasks(ops, stat, Ly, Lx, allow_overlap):
'''creates cell masks for ROIs in stat and computes radii
inputs:
stat, Ly, Lx, allow_overlap
from stat: ipix, ypix, xpix, lam
allow_overlap: boolean whether or not to include overlapping pixels in cell masks (default: False)
outputs:
stat, cell_pix (Ly,Lx), cell_masks (ncells,Ly,Lx)
assigned to stat: radius (minimum of 3 pixels)
'''
ncells = len(stat)
cell_pix = np.zeros((Ly,Lx))
cell_masks = np.zeros((ncells,Ly,Lx), np.float32)
for n in range(ncells):
if allow_overlap:
overlap = np.zeros((stat[n]['npix'],), bool)
else:
overlap = stat[n]['overlap']
ypix = stat[n]['ypix'][~overlap]
xpix = stat[n]['xpix'][~overlap]
lam = stat[n]['lam'][~overlap]
if xpix.size:
# compute radius of neuron (used for neuropil scaling)
radius = utils.fitMVGaus(ypix/ops['diameter'][0], xpix/ops['diameter'][1],lam,2)[2]
stat[n]['radius'] = radius[0] * np.mean(ops['diameter'])
stat[n]['aspect_ratio'] = 2 * radius[0]/(.01 + radius[0] + radius[1])
# add pixels of cell to cell_pix (pixels to exclude in neuropil computation)
cell_pix[ypix[lam>0],xpix[lam>0]] += 1
# add pixels to cell masks
cell_masks[n, ypix, xpix] = lam / lam.sum()
else:
stat[n]['radius'] = 0
stat[n]['aspect_ratio'] = 1
cell_pix = np.minimum(1, cell_pix)
return stat, cell_pix, cell_masks
def activity_stats(parent):
i0 = int(1 - parent.iscell[parent.ichosen])
ypix = np.array([])
xpix = np.array([])
lam = np.array([])
footprints = np.array([])
F = np.zeros((0, parent.Fcell.shape[1]), np.float32)
Fneu = np.zeros((0, parent.Fcell.shape[1]), np.float32)
for n in np.array(parent.imerge):
ypix = np.append(ypix, parent.stat[n]["ypix"])
xpix = np.append(xpix, parent.stat[n]["xpix"])
lam = np.append(lam, parent.stat[n]["lam"])
footprints = np.append(footprints, parent.stat[n]["footprint"])
F = np.append(F, parent.Fcell[n, :][np.newaxis, :], axis=0)
Fneu = np.append(Fneu, parent.Fneu[n, :][np.newaxis, :], axis=0)
# remove overlaps
ipix = np.concatenate((ypix[:, np.newaxis], xpix[:, np.newaxis]), axis=1)
_, goodi = np.unique(ipix, return_index=True, axis=0)
ypix = ypix[goodi]
xpix = xpix[goodi]
lam = lam[goodi]
stat0 = {}
stat0["ypix"] = ypix.astype(np.int32)
stat0["xpix"] = xpix.astype(np.int32)
stat0["lam"] = lam
stat0['med'] = [np.median(stat0["ypix"]), np.median(stat0["xpix"])]
stat0["npix"] = ypix.size
d0 = parent.ops["diameter"]
radius = utils.fitMVGaus(ypix / d0[0], xpix / d0[1], lam, 2)[2]
stat0["radius"] = radius[0] * d0.mean()
stat0["aspect_ratio"] = 2 * radius[0] / (.01 + radius[0] + radius[1])
npix = np.array([parent.stat[n]['npix']
for n in range(len(parent.stat))]).astype('float32')
stat0["npix_norm"] = stat0["npix"] / npix.mean()
# compactness
rs, dy, dx = sparsedetect.circleMask(d0)
rsort = np.sort(rs.flatten())
r2 = ((ypix - stat0["med"][0]) / d0[0])**2 + (
(xpix - stat0["med"][1]) / d0[1])**2
r2 = r2**.5
stat0["mrs"] = np.mean(r2)
stat0["mrs0"] = np.mean(rsort[:r2.size])
stat0["compact"] = stat0["mrs"] / (1e-10 + stat0["mrs0"])
# footprint
stat0["footprint"] = footprints.mean()
F = F.mean(axis=0)
Fneu = Fneu.mean(axis=0)
dF = F - parent.ops["neucoeff"] * Fneu
# activity stats
stat0["skew"] = stats.skew(dF)
stat0["std"] = dF.std()
# compute outline and circle around cell
iext = fig.boundary(ypix, xpix)
stat0["yext"] = ypix[iext]
stat0["xext"] = xpix[iext]
ycirc, xcirc = fig.circle(stat0["med"], stat0["radius"])
goodi = ((ycirc >= 0)
& (xcirc >= 0)
& (ycirc < parent.ops["Ly"])
& (xcirc < parent.ops["Lx"]))
stat0["ycirc"] = ycirc[goodi]
stat0["xcirc"] = xcirc[goodi]
# deconvolve activity
spks = dcnv.oasis(dF[np.newaxis, :], parent.ops)
# add cell to structs
parent.stat = np.concatenate((parent.stat, np.array([stat0])), axis=0)
print(parent.stat[-1]["ypix"].shape)
parent.Fcell = np.concatenate((parent.Fcell, F[np.newaxis, :]), axis=0)
parent.Fneu = np.concatenate((parent.Fneu, Fneu[np.newaxis, :]), axis=0)
parent.Spks = np.concatenate((parent.Spks, spks), axis=0)
iscell = np.array([parent.iscell[parent.ichosen]], dtype=bool)
parent.iscell = np.concatenate((parent.iscell, iscell), axis=0)
def merge_activity_masks(parent):
print('merging activity... this may take some time')
i0 = int(1-parent.iscell[parent.ichosen])
ypix = np.array([])
xpix = np.array([])
lam = np.array([])
footprints = np.array([])
F = np.zeros((0,parent.Fcell.shape[1]), np.float32)
Fneu = np.zeros((0,parent.Fcell.shape[1]), np.float32)
probcell = []
probredcell = []
merged_cells = []
remove_merged = []
for n in np.array(parent.imerge):
if len(parent.stat[n]['imerge']) > 0:
remove_merged.append(n)
for k in parent.stat[n]['imerge']:
merged_cells.append(k)
else:
merged_cells.append(n)
merged_cells = np.unique(np.array(merged_cells))
for n in merged_cells:
ypix = np.append(ypix, parent.stat[n]["ypix"])
xpix = np.append(xpix, parent.stat[n]["xpix"])
lam = np.append(lam, parent.stat[n]["lam"])
footprints = np.append(footprints, parent.stat[n]["footprint"])
F = np.append(F, parent.Fcell[n,:][np.newaxis,:], axis=0)
Fneu = np.append(Fneu, parent.Fneu[n,:][np.newaxis,:], axis=0)
probcell.append(parent.probcell[n])
probredcell.append(parent.probredcell[n])
probcell = np.array(probcell)
probredcell = np.array(probredcell)
pmean = probcell.mean()
prmean = probredcell.mean()
# remove overlaps
ipix = np.concatenate((ypix[:,np.newaxis], xpix[:,np.newaxis]), axis=1)
_, goodi = np.unique(ipix, return_index=True, axis=0)
ypix = ypix[goodi]
xpix = xpix[goodi]
lam = lam[goodi]
stat0 = {}
if 'aspect' in parent.ops:
d0 = np.array([int(parent.ops['aspect']*10), 10])
else:
d0 = parent.ops['diameter']
if isinstance(d0, int):
d0 = [d0,d0]
### compute statistics of merges
stat0["imerge"] = merged_cells
stat0["ypix"] = ypix.astype(np.int32)
stat0["xpix"] = xpix.astype(np.int32)
stat0["lam"] = lam / lam.sum() * merged_cells.size
stat0['med'] = [np.median(stat0["ypix"]), np.median(stat0["xpix"])]
stat0["npix"] = ypix.size
radius = utils.fitMVGaus(ypix/d0[0], xpix/d0[1], lam, 2)[2]
stat0["radius"] = radius[0] * d0.mean()
stat0["aspect_ratio"] = 2 * radius[0]/(.01 + radius[0] + radius[1])
npix = np.array([parent.stat[n]['npix'] for n in range(len(parent.stat))]).astype('float32')
stat0["npix_norm"] = stat0["npix"] / npix.mean()
# compactness
rs,dy,dx = sparsedetect.circleMask(d0)
rsort = np.sort(rs.flatten())
r2 = ((ypix - stat0["med"][0])/d0[0])**2 + ((xpix - stat0["med"][1])/d0[1])**2
r2 = r2**.5
stat0["mrs"] = np.mean(r2)
stat0["mrs0"] = np.mean(rsort[:r2.size])
stat0["compact"] = stat0["mrs"] / (1e-10 + stat0["mrs0"])
# footprint
stat0["footprint"] = footprints.mean()
# inmerge
stat0["inmerge"] = 0
### compute activity of merged cells
F = F.mean(axis=0)
Fneu = Fneu.mean(axis=0)
dF = F - parent.ops["neucoeff"]*Fneu
# activity stats
stat0["skew"] = stats.skew(dF)
stat0["std"] = dF.std()
### for GUI drawing
# compute outline and circle around cell
iext = fig.boundary(ypix, xpix)
stat0["yext"] = ypix[iext].astype(np.int32)
stat0["xext"] = xpix[iext].astype(np.int32)
ycirc, xcirc = fig.circle(stat0["med"], stat0["radius"])
goodi = (
(ycirc >= 0)
& (xcirc >= 0)
& (ycirc < parent.ops["Ly"])
& (xcirc < parent.ops["Lx"])
)
stat0["ycirc"] = ycirc[goodi]
stat0["xcirc"] = xcirc[goodi]
# deconvolve activity
spks = dcnv.oasis(dF[np.newaxis, :], parent.ops)
### remove previously merged cell (do not replace)
for k in remove_merged:
remove_mask(parent, k)
np.delete(parent.stat, k, 0)
np.delete(parent.Fcell, k, 0)
np.delete(parent.Fneu, k, 0)
np.delete(parent.Spks, k, 0)
np.delete(parent.iscell, k, 0)
np.delete(parent.probcell, k, 0)
np.delete(parent.probredcell, k, 0)
np.delete(parent.redcell, k, 0)
np.delete(parent.notmerged, k, 0)
# add cell to structs
parent.stat = np.concatenate((parent.stat, np.array([stat0])), axis=0)
parent.stat = sparsedetect.get_overlaps(parent.stat, parent.ops)
parent.stat = np.array(parent.stat)
parent.Fcell = np.concatenate((parent.Fcell, F[np.newaxis,:]), axis=0)
parent.Fneu = np.concatenate((parent.Fneu, Fneu[np.newaxis,:]), axis=0)
parent.Spks = np.concatenate((parent.Spks, spks), axis=0)
iscell = np.array([parent.iscell[parent.ichosen]], dtype=bool)
parent.iscell = np.concatenate((parent.iscell, iscell), axis=0)
parent.probcell = np.append(parent.probcell, pmean)
parent.probredcell = np.append(parent.probredcell, prmean)
parent.redcell = np.append(parent.redcell, prmean > parent.chan2prob)
parent.notmerged = np.append(parent.notmerged, False)
# recompute binned F
parent.mode_change(parent.activityMode)
for n in merged_cells:
parent.stat[n]['inmerge'] = parent.stat.size-1
add_mask(parent, parent.iscell.size-1)