CC implant segmentation

Simple implementation of largest-connected-component implant segmenting. #6

TODO: Hierarchical CC to get to full resolution.

src/segmentation/segment-implant-cc.py

@@ -1,4 +1,4 @@
-import h5py, sys, os.path, pathlib, numpy as np, scipy.ndimage as ndi
+import h5py, sys, os.path, pathlib, numpy as np, scipy.ndimage as ndi, tqdm
 sys.path.append(sys.path[0]+"/../")
 from config.constants import *
 from config.paths import hdf5_root, binary_root, commandline_args
@@ -8,47 +8,44 @@
 
 NA = np.newaxis
 
-sample, scale, chunk_size, pa_scale = commandline_args({"sample":"<required>","scale":1,'chunk_size':256,"pa_scale":8})
+sample, scale, chunk_size = commandline_args({"sample":"<required>","scale":8, "chunk_size":256})
 
 # Load metadata. TODO: Clean up, make automatic function.
 meta_filename = f"{hdf5_root}/hdf5-byte/msb/{sample}.h5"
 h5meta     = h5py.File(meta_filename,'r')
 vm_shifts  = h5meta['volume_matching_shifts']
 full_Nz, Ny, Nx = h5meta['voxels'].shape    # Full image resolution
-Nz         = full_Nz - np.sum(vm_shifts)
-nz,ny,nx   = np.array([Nz,Ny,Nx])//pa_scale   # Coarse image resolution for PA-analysis
+Nz         = full_Nz - np.sum(vm_shifts)    # Full volume matched image resolution
+nz,ny,nx   = np.array([Nz,Ny,Nx])//scale    # Volume matched image resolution at chosen scale
 
 voxelsize   = h5meta['voxels'].attrs['voxelsize'] * scale
 global_vmin = np.min(h5meta['subvolume_range'][:,0])
 global_vmax = np.max(h5meta['subvolume_range'][:,1])
 values      = np.linspace(global_vmin,global_vmax,2**16)
+implant_threshold_u16 = np.argmin(np.abs(values-implant_threshold))
+print(f"Implant threshold {implant_threshold} -> {implant_threshold_u16} as uint16")
 h5meta.close()
 
 output_dir = f"{binary_root}/masks/implant/{scale}x"
 pathlib.Path(output_dir).mkdir(parents=True, exist_ok=True)
 
-pa_voxels = np.fromfile(f"{binary_root}/voxels/{pa_scale}x/{sample}.uint16",count=nz*ny*nx*2, offset=0, dtype=np.uint16).reshape(nz,ny,nx)
-implant_threshold_u16 = np.argmin(np.abs(values-implant_threshold))
-
-print(f"Implant threshold {implant_threshold} -> {implant_threshold_u16} as uint16")
-pa_noisy_implant = pa_voxels > implant_threshold_u16
+noisy_implant = np.empty((nz,ny,nx),dtype=bool)
+voxel_chunk   = np.empty((chunk_size,ny,nx),dtype=np.uint16)
 
-pa_label, n_features = ndi.label(pa_noisy_implant)
-pa_bincnts    = np.bincount(pa_label[pa_label>0],minlength=n_features+1)
-largest_cc_ix = np.argmin(pa_bincnts)
-
-implant_mask=(pa_label==largest_cc_ix)
-np.savez_compressed(f"{output_dir}/{sample}.npz",implant_mask=implant_mask)
-# for z0 in range(0,Nz,chunk_size):
-#     z1 = min(Nz,z0+chunk_size)
-#     print(f"Reading and thresholding chunk {z0}:{z1} of {voxels_in.shape} {voxels_in.dtype}.")
-#     implant_chunk       = voxels_in[z0:z1] >= byte_implant_threshold
-#     print(f"Max inddata: {voxels_in[z0:z1].max()}; Number of matching voxels: {np.sum(implant_chunk)}")
-#     if(sphere_diameter>1):
-#         print(f"Binary opening with {sphere_diameter*voxelsize} micrometer sphere ({sphere_diameter} voxel radius).")
-#         implant_chunk[sphere_diameter//2:-sphere_diameter//2] = ndi.binary_opening(implant_chunk,sph5)[sphere_diameter//2:-sphere_diameter//2]
-#     print("Writing chunk")
-#     voxels_out[z0:z1]  = implant_chunk
+for z in tqdm.tqdm(range(0,nz,chunk_size),"Loading and thresholding voxels"):
+    chunk_length = min(chunk_size,nz-z)
+    load_slice(voxel_chunk, f"{binary_root}/voxels/{scale}x/{sample}.uint16",
+               (chunk_length,0,0), (nz,ny,nx))
+    noisy_implant[z:z+chunk_length] = voxel_chunk[:chunk_length]
 
-# h5in.close()
-# h5out.close()
+
+print(f"Computing connected components")
+label, n_features = ndi.label(noisy_implant)
+print(f"Counting component volumes")
+bincnts           = np.bincount(label[label>0],minlength=n_features+1)
+
+print(f"Writing largest connected component to {output_dir}/{sample}.npz")
+largest_cc_ix     = np.argmax(bincnts)
+implant_mask=(label==largest_cc_ix)
+np.savez_compressed(f"{output_dir}/{sample}.npz",implant_mask=implant_mask)
+