tech-pi / srf Goto Github PK

I tested and showed the first and last z slice of d18, (110x110x15, 2mm).
The last slice illustrated a normal shape (img21) with a cylinder shaped image (img22).
In the first slice, the effmap has some weird part at the bottom (img11). In its image, there are some significant point errors at the bottom-right corner (img12).
This effect can be observed in other perspectives.

col1 effmap, col2 img.
row 1, z slice 1
row 2, z slice 14
row 3, y slice 55
row 4, x slice 55

• unified shape descriptor:
  json <-> dxl.shape object
• pygate support for shape to mac
• srf.external using pygate to generate mac from shape
• rearrange the functions in external which can be reused (e.g. the get_scanner() function).

There is still an error within the psf process as shown in the figure below.

The similar result appears only in tor model, so I think the problem is caused by the unmatched permutation between psf operation and ordinary tor "recon" step.
Since the problem does not influence the image quality, it would be left aside and solved later.

this commit changes interface of constructor of class RingGeometry is changed.

However it seems inconsistent to RingGeometry's field names.

I'd prefer to change it back to nb_rings and nb_blocks_per_ring.
Note changing examples is also required.

there were errors when multiple customized TF ops were loaded in single run session.

A convenient constructor for scanners. Currently a typical scanner is construct by:

def make_scanner():
    """
    Create a cylindrical PET scanner.
    """
    config = get_mct_config()
    ring = RingGeometry(config['ring'])
    block = Block(block_size=config['block']['size'],
                  grid=config['block']['grid'])
    name = config['name']
    tof = TOF(res=config['tof']['resolution'], bin=config['tof']['bin'])
    return CylindricalPET(name, ring, block, tof)

It seems not clear/simple enough, better hierarchy design maybe helpful .

AbstractPETScanner might be a better place for interfaces like make_ring_pairs_lors .

Mixed usage of efficiency map and 1/efficiency map maybe confusing and cost time for debug, we need to extract two different class for this and use multi dispatch to make our algorithms works with both efficiency maps.

After I pulled the latest SRF, I found it cannot work correctly in reconstruction. Could you please have a test on it. if the reconstruction works, could you give me a demo / example? @chengaoyu

Desired structure might be

• api
  |---- cli
  |---- web
  |---- app