Simple Soma Detection

Counting soma labels by skimage.measure.regionprop()

ero5is the binary image after thresholding at 5% followed by erosion.

• step 1: label regions by skimage.measure.label()
• step 2: filter out smaller regions by skimage.measure.regionprops()
• step 3: print out possible number of soma.

ReLab, NumLab = skimage.measure.label(ero5, return_num=True)
print('number of labels = ',NumLab)

props = skimage.measure.regionprops(ReLab)
CNumLab = NumLab
for x in range(NumLab):
    D = props[x].equivalent_diameter
    if D < 10:
        CNumLab-=1

print('Possible number of somas is',str(CNumLab))

Load location of known soma by _get_voxel()

Specify the radius and the resolution (mip) of the corresponding subvolume to be labeled. Note that different resolution will lead to different number of known cells listed in voxs.

dir = "s3://open-neurodata/brainlit/brain1"
dir_segments = "s3://open-neurodata/brainlit/brain1_segments"
mip = 1; radius = 150; v_id = 0
voxs = [];
ngl_sess = NeuroglancerSession(mip=mip, url=dir, url_segments=dir_segments)
for seg_id in range(1000):
    try:
        vox = ngl_sess._get_voxel(seg_id, v_id) 
        voxs.append(vox)
    except:
        pass

Because the bounding box bboxfor the subvolume is known, we can check the total number of known cells in the subvolume.

Nbbox = [10342,6265,4751,10643,6566,5052]
KnoSomNum = 0
for vox in voxs:
    if vox[0] > Nbbox[0] and vox[0] < Nbbox[3]:
        if vox[1] > Nbbox[1] and vox[1] < Nbbox[4]:
            if vox[2] > Nbbox[2] and vox[2] < Nbbox[5]:
                KnoSomNum +=1

print('Known number of soma within the subvolume is',str(KnoSomNum))