dicomParser is a lightweight library for parsing DICOM P10 byte streams in modern web browsers (IE10+), Node.js and Meteor. dicomParser is fast, easy to use and has no external dependencies.

The best way to see the power of this library is to actually see it in use. A number of live examples are included that are not only useful but also show how to use dicomParser. Click here for a list of all live examples

Get a packaged source file:

• dicomParser.js
• dicomParser.min.js

Or install via NPM:

npm install dicom-parser

Or install via Bower:

bower install dicomParser

Or install via atmosphere for Meteor applications

meteor add chafey:dicom-parser

// create a Uint8Array with the contents of the DICOM P10 byte stream
// you want to parse (e.g. XMLHttpRequest to a WADO server)
var arrayBuffer = new ArrayBuffer(bufferSize);
var byteArray = new Uint8Array(arrayBuffer);

// Parse the byte array to get a DataSet object that has the parsed contents
try
{
    var dataSet = dicomParser.parseDicom(byteArray);

    // access elements by tag
    var sopInstanceUid = dataSet.string('x0020000d');
    // access 16 bit unsigned pixel data for single frame sop instance
    // NOTE: use Uint8Array for for 8 bit gray data and Int16Array for 16 bit signed data
    var pixelData = new Uint16Array(dataSet.byteArray.buffer, dataSet.elements.x7fe00010.dataOffset);
}
catch(err)
{
   // catch parse errors
   console.log('Error parsing byte stream' - err);
}

See the live examples for more in depth usage of the library

Note that actually displaying DICOM images is quite complex due to the variety of pixel formats and compression algorithms that DICOM supports. If you are interested in displaying images, please take a look at the cornerstone library and the cornerstoneWADOImageLoader which uses this library to extract the pixel data from DICOM files and display the images with cornerstone library. You can find the actual code that extracts grayscale pixel data using this library here.

• Parses DICOM Part 10 byte arrays in all encodings
  • Explicit and implicit
  • Little endian and big endian
• Supports all VR's including sequences
• Supports elements with undefined length
• Supports sequence items with undefined length
• Provides functions to convert from all VR types to native Javascript types
• Does not require a data dictionary
• Designed for use in the browser
• Each element exposes the offset and length of its data in the underlying byte stream
• Packaged using the module pattern, as an AMD module and as a CommonJS module for Node.js
• No external dependencies
• Supports extraction of encapsulated pixel data frames
• Convenient utility functions to parse strings formatted in DA, TM and PN VRs and return JavaScript objects
• Convenient utility function to create a string version of an explicit element
• Convenient utility function to convert a parsed explicit dataSet into a javascript object
• Supports reading incomplete/partial byte streams
  • By specifying a tag to stop reading at
  • By returning the elements parsed so far in the exception thrown during a parse error

This project uses grunt to build the software.

NodeJs - click to visit web site for installation instructions.

grunt-cli

npm install -g grunt-cli

bower

npm install -g bower

Update dependencies (after each pull):

npm install

bower install

Running the build:

grunt

Automatically running the build and unit tests after each source change:

grunt watch

Future:

• Add unit tests for sequence parsing functionality and encapsulated pixel frames
• Figure out how to automatically generate documentation from the source (jsdoc)
• Optimize findItemDelimitationItemAndSetElementLength() for speed
• Optimize functions in byteArrayParser.js for speed
• Add example that allows you to compare two sop instances against each other
• Figure out how to not have a global dicomParser object when used with an AMD loader
• See what needs to be done to support different character sets (assumes ASCII currently)
• Support for parsing from streams on Node.js and Meteor
• Switch to JavaScript ES6
• Separate the parsing logic from the dataSet creation logic (e.g. parsing generates events which dataSet creation logic creates the dataSet from)
  • dataSet creation logic could filter out unwanted tags to improve performance of parse
  • dataSet creation logic could defer creation of sequence dataSets to improve performance of parse

• @neandrake for help with getting Node.js support
• @ggerade for implementing support for floats/doubles with VM > 1
• @bryan-cool for bug fix related to parsing implicit little endian files and big endian support
• @snagytx, @doncharkowsky - for bug fix related to reading encapsulated frames

While building the WADO Image Loader for cornerstone, I couldn't find a Javascript DICOM parser that exactly met my needs. DICOM really isn't that hard to parse so I figured I would just make my own. Here are some of the key things that I really wanted out of a DICOM library that I am hoping to deliver:

• License is extremely liberal so it could be used in any type of project
• Only deals with parsing DICOM - no code to actually display the images
• Designed to work well in a browser (modern ones at least)
• Follows modern javascript best practices
• Has documentation and examples on how to use it
• Does not hide the underlying data stream from you
• Does not require a data dictionary
• Decodes individual elements "on demand" - this goes with not needing a data dictionary
• Code guards against corrupt or invalid data streams by sanity checking lengths and offsets
• Does not depend on any external dependencies - just drop it in and go
• Has unit tests
• Code is easy to understand

Interested in knowing why the above goals are important to me? Here you go:

License is extremely liberal so it could be used in any type of project

DICOM is an open standard and parsing it is easy enough that it should be freely available for all types of products - personal, open source and commercial. I am hoping that the MIT license will help it see the widest possible adoption (which will in the end help the most patients). I will dual license it under GPL if someone asks.

Only deals with parsing DICOM - no code to actually display the images

I am a big believer in small reusable pieces of software and loose coupling. There is no reason to tightly couple the parser with image display. I hope that keeping this library small and simple will help it reach the widest adoption.

Designed to work well in a browser (modern ones at least)

There are some good javascript DICOM parsing libraries available for server development on node.js but they won't automatically work in a browser. I needed a library that let me easily parse WADO responses and I figured others would also prefer a simple library to do this with no dependencies. The library does make use of the ArrayBuffer object which is widely supported except for IE (it is available on IE10+). I have no current plans to add support for older versions of IE but would be open to contributions if someone wants to do the work.

Follows modern javascript best practices

This of course means different things to different people but I have found great benefit from making sure my javascript passes jshint and leveraging the module pattern. I also have a great affinity to AMD modules but I understand that not everyone wants to use them. So for this library I am shooting for simply making sure the code uses the module pattern and passes jshint.

Has documentation and examples on how to use it

Do I really need to convince you that this is needed?

Does not hide the underlying data stream from you

I have used many DICOM parsing libraries over the years and most of them either hide the underlying byte stream from you or make it difficult to access. There are times when you need to access the underlying bytes - and it is frustrating when the library works against you. A few examples of the need for this include UN VR's, private attributes, encapsulated pixel data and implicit little endian transfer syntaxes (which unfortunately are still widely being used) when you don't have a complete data dictionary.

This library addresses this issue by exposing the offset and length of the data portion of each element. It also defers parsing (and type converting) the data until it is actually asked to do so. So what you get from a parse is basically a set of pointers to where the data for each element is in the byte stream and then you call the function you want to extract the type you want. An awesome side effect of this is that you don't need a data dictionary to parse a file even if it uses implicit little endian. It also turns out that parsing this way is very fast as it avoids doing unneeded type conversions.

Note that you cannot 100% reliably parse sequence elements in an implicit little endian transfer syntax without a data dictionary. I therefore strongly recommend that you work with explicit transfer syntaxes whenever possible. Fortunately most Image Archives should be able to give you an explicit transfer syntax encoding of your sop instance even if it received it in implicit little endian.

Note that WADO's default transfer syntax is explicit little endian so one would assume that an Image Archive supporting WADO would have a good data dictionary management system. Initially I wasn't going to support parsing of implicit data at all but decided to mainly for convenience (and the fact that many of my test data sets are in little endian transfer syntax and I am too lazy to convert them to explicit transfer syntax).

Does not require a data dictionary

As a client, you usually you know which elements you want to access and know what type they are so designing a client oriented parser around a data dictionary is adding unnecessary complexity, especially if you can stick to explicit transfer syntaxes. I also believe it is the the server's responsibility to provide the client safe and easily digestable data (i.e. explicit transfer syntaxes). A server typically supports many types of clients so it makes sense to centralize data dictionary management in one place rather than burden each client with it.

Data dictionaries are not required for most client use cases anyway so I decided not to support it in this library at all. For those use cases that do require a data dictionary, you can layer it on top of this library. An example of doing so is provided in the live examples. If you do want to know the VR, request the instance in an explicit transfer syntax and you can have it. If your Image Archive can't do this for you, get a new one - seriously.

Decodes individual elements "on demand" - this goes with not needing a data dictionary

See above, this is related to not requiring a data dictionary. Usually you know exactly what elements you need and what their types are. The only time this is not the case is when you are building a DICOM Dump utility or you can't get an explicit transfer syntax and have one of those problematic elements that can be either OB or OW (and you can usually figure out which one it is without the VR anyway)

Code guards against corrupt or invalid data streams by sanity checking lengths and offsets

Even though you would expect an Image Archive to never send you data that isn't 100% DICOM compliant, that is not a bet I would make. As I like to say - there is no "DICOM police" to penalize vendors who ship software that creates bytes streams that violate the DICOM standard. Regardless, it is good practice to never trust data from another system - even one that you are in full control of.

Does not depend on any external dependencies - just drop it in and go

Sort of addressed above as maximizing adoption requires that the library minimize the burden on its users. I did find a few interesting libraries that were targeted at making it easier and safer to parse byte streams but they just seemed like overkill so I decided to do it all in one to keep it as simple as it could be. In general I am a big fan of building complex systems from lots of smaller simpler pieces. Some good references on this include the microjs site and the cujo.js manifseto

Has unit tests

I generally feel that units tests are often a waste of time for front end development. Where unit tests do make sense is code that is decoupled from the user interface - like a DICOM parsing module. I did use TDD on this project and had unit tests covering ~ 80% of the code paths passing before I even tried to load my first real DICOM file. Before I wrote this library, I did a quick prototype without unit tests that actually took me much less time (writing tests takes time....). So in the end I don't think it saved me much time getting to a first release, but I am hoping it will pay for itself in the long run (especially if this library receives wide adoption). I also know that some people out there won't even look at it unless it has good test coverage.

Interesting note here - I did not write unit tests for sequence parsing and undefined lengths mainly because I found the standard difficult to understand in these areas and didn't want to waste my time building tests that were not correct. I ended up making these work by throwing a variety of data sets at it and fixing the issues that I found. Getting this working took about 3x longer than everything else combined so perhaps it would have been faster if I had used TDD on this part.

Code is easy to understand

In my experience, writing code that is easy to understand is far more important than writing documentation or unit tests for that code. The reason is that when a developer needs to fix or enhance a piece of code, they almost never start with the unit tests or documentation - they jump straight into the code and start thrashing about in the debugger. If some other developer is looking at your code, you probably made a mistake - either a simple typo or a design issue if you really blew it. In either case, you should have mercy on them in advance and make their unenviable task of fixing or extending your code the best it can be. Some principles I try to follow include:

• Clear names for source files, functions and variables. These names can get very long but I find that doing so is better than writing comments in the source file
• Small source files. Generally I try to keep each source file to under 300 lines or so. The longer it gets, the harder it is to remember what you are looking at
• Small functions. The longer the function is, the harder it is to understand

You can find out more about this by googling for "self documenting code"

Copyright 2014 Chris Hafey chafey@gmail.com