Y = f['LogNcountsMmus'][:] # gene expression matrix tech_noise = f['LogVar_techMmus'][:] # technical noise genes_het_bool=f['genes_heterogen'][:] # index of heterogeneous(??!??) genes geneID = f['sym_names'][:] # gene names cellcyclegenes_filter = SP.unique(f['cellcyclegenes_filter'][:].ravel() -1) # idx of cell cycle genes cellcyclegenes_filterCB600 = f['ccCBall_gene_indices'][:].ravel() -1 # idxof cell cycle genes ... #ground truth from Hoechst staining phase_vec = f['labels'][:].ravel() if max(phase_vec)==4: phase_vec[phase_vec==1]=2 phase_vec = phase_vec-1 KG1 = SP.zeros((Y.shape[0],Y.shape[0])) for iph in range(Y.shape[0]): for jph in range(Y.shape[0]): if SP.bitwise_and(phase_vec[iph]==phase_vec[jph], phase_vec[iph]==1): KG1[iph,jph]=1 KS = SP.zeros((Y.shape[0],Y.shape[0])) for iph in range(Y.shape[0]): for jph in range(Y.shape[0]): if SP.bitwise_and(phase_vec[iph]==phase_vec[jph], phase_vec[iph]==2): KS[iph,jph]=1 KG2M = SP.zeros((Y.shape[0],Y.shape[0])) for iph in range(Y.shape[0]): for jph in range(Y.shape[0]): if SP.bitwise_and(phase_vec[iph]==phase_vec[jph], phase_vec[iph]==3): KG2M[iph,jph]=1 #intra-phase variations in cell size
fnames = sp.array(fnames).reshape(len(fnames)/2, 2)
pos_items = [(a,b) for a in 'tp','fn' for b in 'tp','fn']
neg_items = [(a,b) for a in 'tn','fp' for b in 'tn','fp']
all_items = pos_items + neg_items
for xitem, yitem in all_items:
idx, x = d1[xitem]
idy, y = d2[yitem]
assert (idx==idy).all()
sel = sp.bitwise_and(x, y)
mfnames = fnames[idx][sel]
nmfnames = len(mfnames)
if nmfnames <= 0:
continue
mpath = path.join(output_path, xitem+'_'+yitem)
os.makedirs(mpath)
for n, (fname1, fname2) in enumerate(mfnames):
arr1 = sp.atleast_3d(sp.misc.imread(fname1).astype('float32'))
arr2 = sp.atleast_3d(sp.misc.imread(fname2).astype('float32'))
# grayscale?
if arr1.shape[2] == 1:
arr1 = sp.concatenate([arr1,arr1,arr1], 2)
'genes_heterogen'][:] # index of heterogeneous(??!??) genes
geneID = f['sym_names'][:] # gene names
cellcyclegenes_filter = SP.unique(f['cellcyclegenes_filter'][:].ravel() -
1) # idx of cell cycle genes
cellcyclegenes_filterCB600 = f['ccCBall_gene_indices'][:].ravel(
) - 1 # idxof cell cycle genes ...
#ground truth from Hoechst staining
phase_vec = f['labels'][:].ravel()
if max(phase_vec) == 4:
phase_vec[phase_vec == 1] = 2
phase_vec = phase_vec - 1
KG1 = SP.zeros((Y.shape[0], Y.shape[0]))
for iph in range(Y.shape[0]):
for jph in range(Y.shape[0]):
if SP.bitwise_and(phase_vec[iph] == phase_vec[jph],
phase_vec[iph] == 1):
KG1[iph, jph] = 1
KS = SP.zeros((Y.shape[0], Y.shape[0]))
for iph in range(Y.shape[0]):
for jph in range(Y.shape[0]):
if SP.bitwise_and(phase_vec[iph] == phase_vec[jph],
phase_vec[iph] == 2):
KS[iph, jph] = 1
KG2M = SP.zeros((Y.shape[0], Y.shape[0]))
for iph in range(Y.shape[0]):
for jph in range(Y.shape[0]):
if SP.bitwise_and(phase_vec[iph] == phase_vec[jph],
phase_vec[iph] == 3):
KG2M[iph, jph] = 1
print "bad file: %s" % (fn)
continue
#store
#pdb.set_trace()
RV['K'] = K
RV['cc_genes_het'] = cc_genes_het
dumpDictHdf5(RV,fout)
fout.close()
Ycorr = RV['Ycorr'][:]
SP.savetxt(out_file_base + '.Ycorr.txt', Ycorr)
SP.savetxt(out_file_base + '.isConverged.txt', RV['is_converged'])
#plot variance decomposition
from scLVM_cfg import *
indsconv = RV['is_converged'].ravel() == 1
is_nocc_conv = SP.bitwise_and(indsconv, cc_genes_het.ravel() == 1)
vars_normConv = RV['varsnorm'][indsconv == 1, :]
# ['hidden_0', 'biol_noise', 'tech_noise']
# ==>
# ['hidden_0', 'tech_noise', 'biol_noise']
#pdb.set_trace()
vars_normConv = vars_normConv[:, [0, 1, 2]]
H2 = 1 - vars_normConv[:, 2]
out_file_pdf = out_file_base+'.pdf'
#pdb.set_trace()
var_plot(vars_normConv, H2, CFG['var_comp_fields'][SP.array([0,1,2])],V_range=SP.linspace(0,1,11),plot_element_count=False,
filename=out_file_pdf, normalize=True, figsize=[6,5])
# var_plot(vars_normConv, H2, CFG['var_comp_fields'][SP.array([0,1,2])],V_range=SP.linspace(0,1,8),plot_element_count=True,
# filename=out_file_pdf, normalize=True, figsize=[6,5])
# calculate average variance components across all genes and visualize
var_mean = vars_normConv.mean(0)
def extract_palm_mask(input_image, sample_spacing=20, debug=False, visual_debug=False, vd_last_step=True):
centre_xy, inner_circle_radius, outer_circle_radius = detect_palm_circle(input_image, visual_debug=visual_debug and not vd_last_step, debug=debug)
outer_circle_points = bresenhamCircle(centre_xy[0], centre_xy[1], outer_circle_radius)
#sorting points in clockwise direction
outer_circle_points = sorted(outer_circle_points, key = lambda c:atan2(c[0] - centre_xy[0], c[1] - centre_xy[1]))
spacing = max(int((sample_spacing*len(outer_circle_points))/float(360)), 1)
#subset the sampling space
sample_points = []
for i in range(0, len(outer_circle_points), spacing):
sample_points.append([outer_circle_points[i][0], outer_circle_points[i][1]])
if debug:
print("[DEBUG] extract_palm_mask : outer circle point - %d, sample space - %d and sample points - %d" % (len(outer_circle_points), spacing, len(sample_points)))
boundary_points = get_boundary_points(sample_points, input_image, debug=debug)
convex_hull, palm_mask = np.zeros(input_image.shape, dtype=np.uint8), np.zeros(input_image.shape, dtype=np.uint8)
for points in boundary_points:
convex_hull[points[0], points[1]] = 255
convex_hull = convex_hull_image(convex_hull)
palm_mask[convex_hull==True] = 255
palm_mask[convex_hull==False] = 0
wrist_point_one, wrist_point_two = wrist_detection(boundary_points)
palm_mask = scipy.bitwise_and(input_image, palm_mask)
display_image = np.stack([input_image, input_image, input_image], axis=2)
#remove the pixel below the wristine
remove_pixel_below_wrist_line(input_image, wrist_point_one, wrist_point_two)
#image rotation
#slope of wristline and angle of rotation
m = (wrist_point_two[0]-wrist_point_one[0])/(wrist_point_two[1]-wrist_point_one[1])
theta = atan(m)*180/pi
rotation_center = ((wrist_point_one[0]+wrist_point_two[0])/2, (wrist_point_one[1]+wrist_point_two[1])/2)
#executing rotation with above parameters
rotated_image, rotation_marix = img_rotation(input_image, theta, rotation_center, debug=debug)
palm_mask, _ = img_rotation(palm_mask, theta, rotation_center, debug=debug)
if visual_debug:
# displaying the palm centre
rr, cc = circle_perimeter(centre_xy[0], centre_xy[1], 3)
rr, cc = valid_circle_points(rr, cc, display_image.shape[0], display_image.shape[1])
display_image[rr, cc] = [0, 0, 255]
# displaying the palm inner circle
rr, cc = circle_perimeter(centre_xy[0], centre_xy[1], inner_circle_radius)
rr, cc = valid_circle_points(rr, cc, display_image.shape[0], display_image.shape[1])
display_image[rr, cc] = [0, 255, 0]
# displaying the palm outer circle
rr, cc = circle_perimeter(centre_xy[0], centre_xy[1], outer_circle_radius)
rr, cc = valid_circle_points(rr, cc, display_image.shape[0], display_image.shape[1])
display_image[rr, cc] = [0, 255, 128]
# displaying the wirst line
rr, cc = line(wrist_point_one[0], wrist_point_one[1], wrist_point_two[0], wrist_point_two[1])
display_image[rr, cc] = [255, 0, 0]
# imgnew = cv2.rotate(img,theta)
fig = plt.figure()
ax0 = fig.add_subplot(2, 2, 1)
ax0.set_title("original input image")
ax0.imshow(display_image)
ax1 = fig.add_subplot(2, 2, 2)
ax1.set_title("rotated image wrt wrist line")
ax1.imshow(rotated_image, cmap='gray')
ax2 = fig.add_subplot(2, 2, 3)
ax2.set_title("palm mask with wrist line")
ax2.imshow(palm_mask, cmap='gray')
ax3 = fig.add_subplot(2, 2, 4)
ax3.set_title("segmentation")
ax3.imshow((rotated_image - palm_mask), cmap='gray')
# wristline_len = np.sqrt(np.square(wrist_point_one[0] - wrist_point_two[0]) + np.square(wrist_point_one[1] - wrist_point_two[1]))
# ax3.imshow(img_rotation(input_image, theta, (wrist_point_one[0]-wristline_len, wrist_point_one[0]-wristline_len), debug=debug), cmap='gray')
plt.show()
return palm_mask, rotated_image
