2015 Summer Project Week:Dicom parsing with DCMJS

• 

  Projects List

Key Investigators

• Nicolas Rannou
• Michael Onken
• Steve Pieper
• Yves Martelli

Project Description

Objective

• Define target DICOM types
• Define a good workflow to parse targets with DCMJS.
• Define required JS models to support the workflow.
• Evaluate DicomParser from cornerstone team.

Approach, Plan

• Talk with DICOM/DCMTK experts to get some feedback on best strategy.
• Write a demo to showcase js models and dicom parsing workflows.

Progress

• Evaluation of current solutions
  • note: compilation of dcmjs with -o3 or -of drasticly improves performance of dcmjs
• Proposed workflow
  • https://docs.google.com/drawings/d/1BX0NBIDPrrM-JGk9b1awL_w57f9rMvdTJSeVjT9bM5U
  • note: workflow relies on new dcmtk functionnalities. It will have to be evaluated in term of performance.
  • https://github.com/andywer/threadpool-js
• Take dicom segmentations/models into account
• DICOM testing database discussion with web-torrents

Benchmark

Benchmark of DICOM parser libraries using benchmark.js (what-do-the-results-from-benchmark-js-mean).

These results were obtained on a Lenovo Flex2 laptop (Intel Core i3-4030U) under Microsoft Windows 8.1 using Mozilla Firefox (ff) 38.0.5 and Google Chrome v43.0.

Parsers: dcmjs (3fe78dd), dicomParser (v1.0.2) and dwv (v0.10.0). The test just calls the parse method of each library, no values are checked. Note that the 'dicomParser' does not interpret the data until asked for when the others do at parsing time.

Units are in operations/seconds.

	dcmjs		dicomParser		dwv
	ff	chrome	ff	chrome	ff	chrome
gdcm-CT-MONO2-8-abdo	320	112	16115	9974	4551	3640
gdcm-CR-MONO1-10-chest	249	83.13	8412	9004	3603	2804
osirix-toutatix-100	170	48.65	3245	3843	1411	995
osirix-goudurix	189	53.97	6346	3995	1885	1363
dicompyler-ct.0	187	60.86	8414	5156	2612	1930
gdcm-US-RGB-8-epicard	148	54.04	10418	10217	31.42	256
gdcm-US-RGB-8-esopecho	501	110	10309	10198	4642	3340
Average	252	75	9037	7484	2676	2047
				3.4
					3.7
		0.03
			0.01