Python plugin for Orthanc


This plugin can be used to write Orthanc plugins using the Python programming language instead of the more complex C/C++ programming languages.

Python plugins have access to more features and a more consistent SDK than Lua scripts. The Python API is automatically generated from the Orthanc plugin SDK in C using the Clang compiler front-end.

As of release 2.0 of the plugin, the coverage of the C SDK is about 75% (119 functions are automatically wrapped in Python out of a total of 157 functions in the Orthanc SDK 1.7.2).


Pay attention to the fact that this plugin is licensed under the terms of the AGPL license.

This has an important consequence: If you distribute Orthanc to clients together with one Python script, or if you put an Orthanc server equipped with one Python script on a Web portal, you must disclose the source code of your Python script to the Orthanc community under the terms of the AGPL license.

We suggest you to put the source code of your Python scripts on the dedicated “OrthancContributed” repository on GitHub, and/or to send it to the Orthanc Users discussion group.

Check out the FAQ about licensing for more context.



The most direct way of starting Orthanc together with the Python plugin is through Docker. Let’s create the file /tmp/ that contains the following basic Python script:

print('Hello world!')

Let’s also create the file /tmp/orthanc.json that contains the following minimal configuration for Orthanc:

  "StorageDirectory" : "/var/lib/orthanc/db",
  "RemoteAccessAllowed" : true,
  "Plugins" : [
  "PythonScript" : "/etc/orthanc/"

Given these two files, Orthanc can be started as follows:

$ docker run -p 4242:4242 -p 8042:8042 --rm \
  -v /tmp/orthanc.json:/etc/orthanc/orthanc.json:ro \
  -v /tmp/ \

You’ll see the following excerpt in the log, which indicates that the Python plugin is properly loaded:

W0331 15:48:12.990661 PluginsManager.cpp:269] Registering plugin 'python' (version mainline)
W0331 15:48:12.990691 PluginsManager.cpp:168] Python plugin is initializing
W0331 15:48:12.990743 PluginsManager.cpp:168] Using Python script "" from directory: /etc/orthanc
W0331 15:48:12.990819 PluginsManager.cpp:168] Program name: /usr/local/sbin/Orthanc
Hello world!

Here is a full example of a more complex setup using the osimis/orthanc images.

Compiling from source

The procedure to compile this plugin from source is similar to that for the core of Orthanc. The following commands should work for most UNIX-like distribution (including GNU/Linux):

$ mkdir Build
$ cd Build
$ make

Before running CMake, make sure that the Python interpreter and its associated development library are installed. On Ubuntu 18.04 LTS, you would for instance install packages libpython3.7-dev and python3.7.

The compilation will produce the shared library OrthancPython, that can be loaded by properly setting the Plugins configuration option of Orthanc.

Warning: The shared library is only compatible with the Python interpreter whose version corresponds to the value of the PYTHON_VERSION argument that was given to CMake.

Note for OS X: As indicated by Stephen Douglas Scotti, here is a sample invocation of CMake to force the version of Python to be used on OS X:

        -DPYTHON_LIBRARY=/usr/local/Cellar/python@3.8/3.8.5/Frameworks/Python.framework/Versions/3.8/lib/libpython3.8.dylib \

Microsoft Windows

Pre-compiled binaries for Microsoft Windows are also available.

Beware that one version of the Python plugin can only be run against one version of the Python interpreter. This version is clearly indicated in the filename of the precompiled binaries.

You are of course free to compile the plugin from sources. You’ll have to explicitly specify the path to your Python installation while invoking CMake. For instance:

C:\orthanc-python\Build> cmake .. -DPYTHON_VERSION=2.7 -DPYTHON_WINDOWS_ROOT=C:/Python27 \
                                  -DSTATIC_BUILD=ON -DCMAKE_BUILD_TYPE=Release -G "Visual Studio 15 2017"

Note about debug builds: Usually, building Python modules such as the Python plugin for Orthanc in debug mode (where _DEBUG is defined) leads to a module (.exe or .dll) that requires a debug build of Python, and debug versions of all the Python libraries. This is quite cumbersome, for it requires building Python on your own or downloading additional debug files.

Since using a debug build of Python is only necessary in very specific cases (such as the debugging of code at the boundary between Python and an extension), we have changed the default behavior to use the release Python library by default.

This means that you are able to build this plugin in debug mode with the standard Python distribution.

In case you need to use the Python debug libraries, you can instruct the build system to do so by setting the PYTHON_WINDOWS_USE_RELEASE_LIBS CMake option, that is ON by default, to OFF. The previous build example would then be, should you require a full debug build:

C:\orthanc-python\Build> cmake .. -DPYTHON_VERSION=2.7 -DPYTHON_WINDOWS_ROOT=C:/Python27 \
                                  -DSTATIC_BUILD=ON -DPYTHON_WINDOWS_USE_RELEASE_LIBS=OFF \
                                  -DCMAKE_BUILD_TYPE=Debug -G "Visual Studio 15 2017"

Please note that this CMake option only impacts debug builds under Windows, when using (any version of) the Microsoft Visual Studio compiler.

The precompiled binaries all use release (i.e. non-debug) versions of Python.

Configuration options

The only two configuration options that are available for this plugin are the following:

  • PythonScript indicates where the Python script is located. If this configuration option is not provided, the Python plugin is not started.
  • PythonVerbose is a Boolean value to make the Python interpreter verbose.


Extending the REST API

Here is a basic Python script that registers two new routes in the REST API:

import orthanc
import pprint

def OnRest(output, uri, **request):
    print('Accessing uri: %s' % uri)
    output.AnswerBuffer('ok\n', 'text/plain')

orthanc.RegisterRestCallback('/(to)(t)o', OnRest)
orthanc.RegisterRestCallback('/tata', OnRest)

Here is the associated minimal configuration file for Orthanc (provided the Python script is saved as

  "Plugins" : [ "." ],
  "PythonScript" : "",
  "PythonVerbose" : false

The route can then be accessed as:

$ curl http://localhost:8042/toto

Listening to changes

This sample uploads a DICOM file as soon as Orthanc is started:

import orthanc

def OnChange(changeType, level, resource):
    if changeType == orthanc.ChangeType.ORTHANC_STARTED:

        with open('/tmp/sample.dcm', 'rb') as f:

     elif changeType == orthanc.ChangeType.ORTHANC_STOPPED:

     elif changeType == orthanc.ChangeType.NEW_INSTANCE:
         print('A new instance was uploaded: %s' % resource)



In releases <= 3.0 of the Python plugin, deadlocks might emerge if you call other core primitives of Orthanc (such as the REST API) in your callback function. This issue has been fixed in release 3.1.

As a temporary workaround against such deadlocks in releases <= 3.0, if you have to call other primitives of Orthanc, you should make these calls in a separate thread, passing the pending events to be processed through a message queue. Here is the template of a possible solution to postpone such deadlocks as much as possible by relying on the multithreading primitives of Python:

import orthanc
import threading

def OnChange(changeType, level, resource):
    # One can safely invoke the "orthanc" module in this function
    orthanc.LogWarning("Hello world")

def _OnChange(changeType, level, resource):
    # Invoke the actual "OnChange()" function in a separate thread
    t = threading.Timer(0, function = OnChange, args = (changeType, level, resource))


Beware that this workaround is imperfect and deadlocks have been observed even if using it! Make sure to upgrade your plugin to solve this issue for good. Note that this temporary workaround is not needed in releases >= 3.1 of the plugin.

Accessing the content of a new instance

import orthanc
import json
import pprint

def OnStoredInstance(dicom, instanceId):
    print('Received instance %s of size %d (transfer syntax %s, SOP class UID %s)' % (
        instanceId, dicom.GetInstanceSize(),

    # Print the origin information
    if dicom.GetInstanceOrigin() == orthanc.InstanceOrigin.DICOM_PROTOCOL:
        print('This instance was received through the DICOM protocol')
    elif dicom.GetInstanceOrigin() == orthanc.InstanceOrigin.REST_API:
        print('This instance was received through the REST API')

    # Print the DICOM tags



Your callback function will be called synchronously with the core of Orthanc. This implies that deadlocks might emerge if you call other core primitives of Orthanc in your callback (such deadlocks are particular visible in the presence of other plugins or Lua scripts). It is thus strongly advised to avoid any call to the REST API of Orthanc in the callback. If you have to call other primitives of Orthanc, you should make these calls in a separate thread, passing the pending events to be processed through a message queue.

Calling pydicom

Here is a sample Python plugin that registers a REST callback to dump the content of the dataset of one given DICOM instance stored in Orthanc, using pydicom:

import io
import orthanc
import pydicom

def DecodeInstance(output, uri, **request):
    if request['method'] == 'GET':
        # Retrieve the instance ID from the regular expression (*)
        instanceId = request['groups'][0]
        # Get the content of the DICOM file
        f = orthanc.GetDicomForInstance(instanceId)
        # Parse it using pydicom
        dicom = pydicom.dcmread(io.BytesIO(f))
        # Return a string representation the dataset to the caller
        output.AnswerBuffer(str(dicom), 'text/plain')

orthanc.RegisterRestCallback('/pydicom/(.*)', DecodeInstance)  # (*)

This can be called as follows:

$ curl http://localhost:8042/pydicom/19816330-cb02e1cf-df3a8fe8-bf510623-ccefe9f5

Auto-routing studies

Here is a sample Python plugin that routes any stable study to a modality named samples (as declared in the DicomModalities configuration option):

import orthanc

def OnChange(changeType, level, resourceId):
    if changeType == orthanc.ChangeType.STABLE_STUDY:
        print('Stable study: %s' % resourceId)
        orthanc.RestApiPost('/modalities/sample/store', resourceId)


Note that, if you want to use an orthanc plugin to transfer the study, you should use the RestApiPostAfterPlugins() method:

import orthanc

def OnChange(changeType, level, resourceId):
    if changeType == orthanc.ChangeType.STABLE_STUDY:
        print('Stable study: %s' % resourceId)
        orthanc.RestApiPostAfterPlugins('/dicom-web/servers/sample/store', resourceId)


Rendering a thumbnail using PIL/Pillow

from PIL import Image
import io
import orthanc

def DecodeInstance(output, uri, **request):
    if request['method'] == 'GET':
        # Retrieve the instance ID from the regular expression (*)
        instanceId = request['groups'][0]

        # Render the instance, then open it in Python using PIL/Pillow
        png = orthanc.RestApiGet('/instances/%s/rendered' % instanceId)
        image =

        # Downsize the image as a 64x64 thumbnail
        image.thumbnail((64, 64), Image.ANTIALIAS)

        # Save the thumbnail as JPEG, then send the buffer to the caller
        jpeg = io.BytesIO(), format = "JPEG", quality = 80)
        output.AnswerBuffer(, 'text/plain')


orthanc.RegisterRestCallback('/pydicom/(.*)', DecodeInstance)  # (*)

Inspecting the available API

Thanks to Python’s introspection primitives, it is possible to inspect the API of the orthanc module in order to dump all the available features:

import inspect
import numbers
import orthanc

# Loop over the members of the "orthanc" module
for (name, obj) in inspect.getmembers(orthanc):
    if inspect.isroutine(obj):
        print('Function %s():\n  Documentation: %s\n' % (name, inspect.getdoc(obj)))

    elif inspect.isclass(obj):
        print('Class %s:\n  Documentation: %s' % (name, inspect.getdoc(obj)))

        # Loop over the members of the class
        for (subname, subobj) in inspect.getmembers(obj):
            if isinstance(subobj, numbers.Number):
                print('  - Enumeration value %s: %s' % (subname, subobj))
            elif (not subname.startswith('_') and
                print('  - Method %s(): %s' % (subname, inspect.getdoc(subobj)))

Scheduling a task for periodic execution

The following Python script will periodically (every second) run the function Hello() thanks to the threading module:

import orthanc
import threading

TIMER = None

def Hello():
    global TIMER
    TIMER = None
    orthanc.LogWarning("In Hello()")
    # Do stuff...
    TIMER = threading.Timer(1, Hello)  # Re-schedule after 1 second

def OnChange(changeType, level, resource):
    if changeType == orthanc.ChangeType.ORTHANC_STARTED:
        orthanc.LogWarning("Starting the scheduler")

    elif changeType == orthanc.ChangeType.ORTHANC_STOPPED:
        if TIMER != None:
            orthanc.LogWarning("Stopping the scheduler")


Filtering and returning metadata

Besides the main DICOM tags, Orthanc associates some metadata to each resource it stores (this includes the date of last update, the transfer syntax, the remote AET...). People are often interested in getting such metadata while calling the /tools/find route in the REST API, or even in filtering this metadata the same way they look for DICOM tags.

This feature is not built in the core of Orthanc, as metadata is not indexed in the Orthanc database, contrarily to the main DICOM tags. Filtering metadata requires a linear search over all the matching resources, which induces a cost in the performance.

Nevertheless, here is a full sample Python script that overwrites the /tools/find route in order to give access to metadata:

import json
import orthanc
import re

# Get the path in the REST API to the given resource that was returned
# by a call to "/tools/find"
def GetPath(resource):
    if resource['Type'] == 'Patient':
        return '/patients/%s' % resource['ID']
    elif resource['Type'] == 'Study':
        return '/studies/%s' % resource['ID']
    elif resource['Type'] == 'Series':
        return '/series/%s' % resource['ID']
    elif resource['Type'] == 'Instance':
        return '/instances/%s' % resource['ID']
        raise Exception('Unknown resource level')

def FindWithMetadata(output, uri, **request):
    # The "/tools/find" route expects a POST method
    if request['method'] != 'POST':
        # Parse the query provided by the user, and backup the "Expand" field
        query = json.loads(request['body'])

        if 'Expand' in query:
            originalExpand = query['Expand']
            originalExpand = False

        # Call the core "/tools/find" route
        query['Expand'] = True
        answers = orthanc.RestApiPost('/tools/find', json.dumps(query))

        # Loop over the matching resources
        filteredAnswers = []
        for answer in json.loads(answers):
                # Read the metadata that is associated with the resource
                metadata = json.loads(orthanc.RestApiGet('%s/metadata?expand' % GetPath(answer)))

                # Check whether the metadata matches the regular expressions
                # that were provided in the "Metadata" field of the user request
                isMetadataMatch = True
                if 'Metadata' in query:
                    for (name, pattern) in query['Metadata'].items():
                        if name in metadata:
                            value = metadata[name]
                            value = ''

                        if re.match(pattern, value) == None:
                            isMetadataMatch = False

                # If all the metadata matches the provided regular
                # expressions, add the resource to the filtered answers
                if isMetadataMatch:
                    if originalExpand:
                        answer['Metadata'] = metadata
                # The resource was deleted since the call to "/tools/find"

        # Return the filtered answers in the JSON format
        output.AnswerBuffer(json.dumps(filteredAnswers, indent = 3), 'application/json')

orthanc.RegisterRestCallback('/tools/find', FindWithMetadata)

Warning: In the sample above, the filtering of the metadata is done using Python’s library for regular expressions. It is evidently possible to adapt this script in order to use the DICOM conventions about attribute matching.

Here is a sample call to retrieve all the studies that were last updated in 2019 thanks to this Python script:

$ curl http://localhost:8042/tools/find -d '{"Level":"Study","Query":{},"Expand":true,"Metadata":{"LastUpdate":"^2019.*$"}}'

Implementing basic paging

As explained in the FAQ, the Orthanc Explorer interface is low-level, and is not adapted for end-users. One common need is to implement paging of studies, which calls for server-side sorting of studies. This can be done using the following sample Python plugin that registers a new route /sort-studies in the REST API of Orthanc:

import json
import orthanc

def GetStudyDate(study):
    if 'StudyDate' in study['MainDicomTags']:
        return study['MainDicomTags']['StudyDate']
        return ''

def SortStudiesByDate(output, uri, **request):
    if request['method'] == 'GET':
        # Retrieve all the studies
        studies = json.loads(orthanc.RestApiGet('/studies?expand'))

        # Sort the studies according to the "StudyDate" DICOM tag
        studies = sorted(studies, key = GetStudyDate)

        # Read the limit/offset arguments provided by the user
        offset = 0
        if 'offset' in request['get']:
            offset = int(request['get']['offset'])

        limit = 0
        if 'limit' in request['get']:
            limit = int(request['get']['limit'])

        # Truncate the list of studies
        if limit == 0:
            studies = studies[offset : ]
            studies = studies[offset : offset + limit]

        # Return the truncated list of studies
        output.AnswerBuffer(json.dumps(studies), 'application/json')

orthanc.RegisterRestCallback('/sort-studies', SortStudiesByDate)

Here is a sample call to this new REST route, that could be issued by any JavaScript framework (the json_pp command-line pretty-prints a JSON file):

$ curl http://localhost:8042/sort-studies | json_pp

This route also implement paging (i.e. it can limit and offset the returned studies):

$ curl 'http://localhost:8042/sort-studies?offset=2&limit=2' | json_pp

Obviously, this basic sample can be improved in many ways. To improve performance, one could for instance cache the result of /studies?expand in memory by listening to changes in the list of studies (cf. orthanc.ChangeType.NEW_STUDY and orthanc.ChangeType.DELETED).

Creating a Microsoft Excel report

As Orthanc plugins have access to any installed Python module, it is very easy to implement a server-side plugin that generates a report in the Microsoft Excel .xls format. Here is a working example:

import StringIO
import json
import orthanc
import xlwt

def CreateExcelReport(output, uri, **request):
    if request['method'] != 'GET' :
        # Create an Excel writer
        excel = xlwt.Workbook()
        sheet = excel.add_sheet('Studies')

        # Loop over the studies stored in Orthanc
        row = 0
        studies = orthanc.RestApiGet('/studies?expand')
        for study in json.loads(studies):
            sheet.write(row, 0, study['PatientMainDicomTags'].get('PatientID'))
            sheet.write(row, 1, study['PatientMainDicomTags'].get('PatientName'))
            sheet.write(row, 2, study['MainDicomTags'].get('StudyDescription'))
            row += 1

        # Serialize the Excel workbook to a string, and return it to the caller
        b = StringIO.StringIO()
        output.AnswerBuffer(b.getvalue(), 'application/')

orthanc.RegisterRestCallback('/report.xls', CreateExcelReport)

If opening the http://localhost:8042/report.xls URI, this Python will generate a workbook with one sheet that contains the list of studies, with the patient ID, the patient name and the study description.

Forbid or allow access to REST resources (authorization)

The following Python script installs a callback that is triggered whenever the HTTP server of Orthanc is accessed:

import orthanc
import pprint

def Filter(uri, **request):
    print('User trying to access URI: %s' % uri)
    return True  # False to forbid access


If access is not granted, the Filter callback must return False. As a consequence, the HTTP status code would be set to 403 Forbidden. If access is granted, the Filter must return true. The request argument contains more information about the request (such as the HTTP headers, the IP address of the caller and the GET arguments).

Note that this is similar to the IncomingHttpRequestFilter() callback that is available in Lua scripts.

Thanks to Python, it is extremely easy to call remote Web services for authorization. Here is an example using the requests library:

import json
import orthanc
import requests

def Filter(uri, **request):
    body = {
        'uri' : uri,
        'headers' : request['headers']
    r ='http://localhost:8000/authorize',
                      data = json.dumps(body))
    return r.json() ['granted']  # Must be a Boolean


This filter could be used together with the following Web service implemented using Node.js:

const http = require('http');

const requestListener = function(req, res) {
  let body = '';
    req.on('data', function(chunk) {
    body += chunk;
  req.on('end', function() {
    var answer = {
      'granted' : false  // Forbid access


Performance and concurrency

Important: This section only applies to UNIX-like systems. The multiprocessing package will not work on Microsoft Windows as the latter OS has a different model for forking processes.

Using slave processes

Let us consider the following sample Python script that makes a CPU-intensive computation on a REST callback:

import math
import orthanc
import time

# CPU-intensive computation taking about 4 seconds
def SlowComputation():
    start = time.time()
    for i in range(1000):
        for j in range(30000):
    end = time.time()
    duration = (end - start)
    return 'computation done in %.03f seconds\n' % duration

def OnRest(output, uri, **request):
    answer = SlowComputation()
    output.AnswerBuffer(answer, 'text/plain')

orthanc.RegisterRestCallback('/computation', OnRest)

Calling this REST route from the command-line returns the time that is needed to compute 30 million times a squared root on your CPU:

$ curl http://localhost:8042/computation
computation done in 4.208 seconds

Now, let us call this route three times concurrently (we use bash):

$ (curl http://localhost:8042/computation & curl http://localhost:8042/computation & curl http://localhost:8042/computation )
computation done in 11.262 seconds
computation done in 12.457 seconds
computation done in 13.360 seconds

As can be seen, the computation time has tripled. This means that the computations were not distributed across the available CPU cores. This might seem surprising, as Orthanc is a threaded server (in Orthanc, a pool of C++ threads serves concurrent requests).

The explanation is that the Python interpreter (CPython actually) is built on the top of a so-called Global Interpreter Lock (GIL). The GIL is basically a mutex that protects all the calls to the Python interpreter. If multiple C++ threads from Orthanc call a Python callback, only one can proceed at any given time. Note however that the GIL only applies to the Python script: The baseline REST API of Orthanc is not affected by the GIL.

The solution is to use the multiprocessing primitives of Python. The “master” Python interpreter that is initially started by the Orthanc plugin, can start several children processes, each of these processes running a separate Python interpreter. This allows to offload intensive computations from the “master” Python interpreter of Orthanc onto those “slave” interpreters. The multiprocessing library is actually quite straightforward to use:

import math
import multiprocessing
import orthanc
import signal
import time

# CPU-intensive computation taking about 4 seconds
# (same code as above)
def SlowComputation():
    start = time.time()
    for i in range(1000):
        for j in range(30000):
    end = time.time()
    duration = (end - start)
    return 'computation done in %.03f seconds\n' % duration

# Ignore CTRL+C in the slave processes
def Initializer():
    signal.signal(signal.SIGINT, signal.SIG_IGN)

# Create a pool of 4 slave Python interpreters
POOL = multiprocessing.Pool(4, initializer = Initializer)

def OnRest(output, uri, **request):
    # Offload the call to "SlowComputation" onto one slave process.
    # The GIL is unlocked until the slave sends its answer back.
    answer = POOL.apply(SlowComputation)
    output.AnswerBuffer(answer, 'text/plain')

orthanc.RegisterRestCallback('/computation', OnRest)

Here is now the result of calling this route three times concurrently:

$ (curl http://localhost:8042/computation & curl http://localhost:8042/computation & curl http://localhost:8042/computation )
computation done in 4.211 seconds
computation done in 4.215 seconds
computation done in 4.225 seconds

As can be seen, the calls to the Python computation now fully run in parallel (the time is cut down from 12 seconds to 4 seconds, the same as for one isolated request).

Note also how the multiprocessing library allows to make a fine control over the computational resources that are available to the Python script: The number of “slave” interpreters can be easily changed in the constructor of the multiprocessing.Pool object, and are fully independent of the threads used by the Orthanc server.

Obviously, an in-depth discussion about the multiprocessing library is out of the scope of this document. There are many references available on Internet. Also, note that threading is not useful here, as Python multithreading is also limited by the GIL, and is more targeted at dealing with costly I/O operations or with the scheduling of commands.

Slave processes and the “orthanc” module

Very importantly, pay attention to the fact that only the “master” Python interpreter has access to the Orthanc SDK. The “slave” processes have no access to the orthanc module.

You must write your Python plugin so as that all the calls to orthanc are moved from the slaves process to the master process. For instance, here is how you would parse a DICOM file in a slave process:

import pydicom
import io

def OffloadedDicomParsing(dicom):
    # No access to the "orthanc" library here, as we are in the slave process
    dataset = pydicom.dcmread(io.BytesIO(dicom))
    return str(dataset)

def OnRest(output, uri, **request):
    # The call to "orthanc.RestApiGet()" is only possible in the master process
    dicom = orthanc.RestApiGet('/instances/19816330-cb02e1cf-df3a8fe8-bf510623-ccefe9f5/file')
    answer = POOL.apply(OffloadedDicomParsing, args = (dicom, ))
    output.AnswerBuffer(answer, 'text/plain')

Communication primitives such as multiprocessing.Queue are available to exchange messages from the “slave” Python interpreters to the “master” Python interpreter for more advanced scenarios.

NB: Starting with release 3.0 of the Python plugin, it is possible to call the REST API of Orthanc from a slave process in a more direct way. The function orthanc.GenerateRestApiAuthorizationToken() can be used to create an authorization token that provides full access to the REST API of Orthanc (without have to set credentials in your plugin). Any HTTP client library for Python, such as requests, can then be used to access the REST API of Orthanc. Here is a minimal example:

import json
import multiprocessing
import orthanc
import requests
import signal

TOKEN = orthanc.GenerateRestApiAuthorizationToken()

def SlaveProcess():
    r = requests.get('http://localhost:8042/instances',
                     headers = { 'Authorization' : TOKEN })
    return json.dumps(r.json())

def Initializer():
    signal.signal(signal.SIGINT, signal.SIG_IGN)

POOL = multiprocessing.Pool(4, initializer = Initializer)

def OnRest(output, uri, **request):
    answer = POOL.apply(SlaveProcess)
    output.AnswerBuffer(answer, 'text/plain')

orthanc.RegisterRestCallback('/computation', OnRest)