Welcome to NeurDICOM documentation

Welcome to everyone

Hello, everyone! I'm very excited that Github's community is interested in my project. Unfortunatly, this project needs improvements in performance and stability. I plan to rewrite some code and replace technologies which I used in first version. It would be great if you bring new fresh ideas to enhance application.

NeurDICOM is portable and easy-to-deploy RESTful DICOM and PACS server that allows physicians to use state-of-the-art methods of machine learning and neural networks to make a diagnosis based on medical images processing and interpetating.

Content

1. API

1.1. Patients
1.2. Studies
1.3. Series
1.4. Instances
1.5. Plugins

2. Plugins

🔴 1. API

• Users
• Patients
• Studies
• Series
• Instances
• Plugins
• Models
• DICOM nodes

📗 1.1. Patients

Base URL: /api

GET /patients

Description: Find all patients

GET /patients/:id

Description: Find patient by id

GET /patients/:id/studies

Description: Find patient's studies

PUT /patients/:id

Description: Update patient

DELETE /patients/:id

Description: Delete patient

📗 1.3 Studies

Base URL: /api

GET /studies

Description: Find all studies

GET /studies/:id

Description: Find study by id

GET /studies/:id/series

Description: Find study's series

PUT /studies/:id

Description: Update study

DELETE /studies/:id

Description: Delete study

📗 1.3 Series

Base URL: /api

GET /series

Description: Find all series

GET /series/:id

Description: Find series by id

GET /series/:id/instances

Description: Find series' instances

PUT /series/:id

Description: Update series

DELETE /series/:id

Description: Delete series

📗 1.4 Instances

Base URL: /api

GET /instances

Description: Find all instances

GET /instances/:id

Description: Find instances by id

GET /instances/:id/tags

Description: Get instance's tags

GET /instances/:id/image

Description: Get instance's image

GET /instances/:id/raw

Description: Get instance's raw pixel data in bytes

GET /instances/:instance_id/process/by_plugin/:plugin_id

Description: Process instance by plugin

POST /instances

Description: Save instance

PUT /instances/:id

Description: Update instance

DELETE /instances/:id

Description: Delete instance

📗 1.5 Plugins

Base URL: /api

GET /plugins

Description: Find all plugins

GET /plugins/:id

Description: Find plugin by id

POST /plugins

Description: Upload plugin

DELETE /plugins/:id

Description: Delete plugin

🔴 2. Plugins

Neurdicom provides API for creating plugins to extend the functionality of DICOM files processing. Plugins should be written in Python and packed as ZIP archive. Plugin should have at least two files 'META.json' and 'plugin.py'. 'META.json' describes the plugin, author and main plugin information. Expected strusture of 'META.json' is shown below:

{
	"plugin_guid": "...",
	"author": "John Doe",
	"name": "hello",
	"info": "Print hello greeting",
	"help": "To greet say 'Hello'",
	"params": {
		"name": {
			"type": "text"
		}
	}
}

Field "params" list all parameters which plugin expects. Each "params" field has one of several options "type", "isarray", "values", "isnull". Field "type" is obligatory and takes one of the following values:

[
	"text",
	"int",
	"double"
]

Fields "isarray", "isnull" are boolearn or nullable. Nullable value are is considered as false.
Plugin should have a public "Plugin" class. "Plugin" class should have at least three of the following methods "init", "process" and "destory". Example of "Plugin" class is shown below:

class Plugin:
    def init(self):
        print('Init method called')

    def process(self, ds: Dataset = None, *args, **kwargs):
        print('Hello,', kwargs['name'], '!')
        return {
            'greeting': 'Hello, %s!' % kwargs['name']
        }

    def destroy(self):
        print('Destroy method called')

Also you can write plugins in C/C++. All C/C++ plugins should be createad as a shared library and compiled at the specific target. Library should contain exposed extern C-functions to provide an access to plugin class. For example:

Our library contains 4 files: library.h, library.cpp, plugin.h, plugin.cpp library.h and library.cpp contain defintions of C-functions which provide access to plugin initialization, processing and destruction:

library.h

#include "Plugin.h"

extern "C" Plugin* InitPlugin();
extern "C" double* Process(Plugin *plugin, double *img, int width, int height, const char* params);
extern "C" void DestroyPlugin(Plugin* plugin);

library.cpp

#include "library.h"
extern "C" {

    Plugin* InitPlugin(){
        return new Plugin;
    }

    double* Process(Plugin *plugin, double *img, int width, int height, const char* params) {
        return plugin->process(img, width, height, params);
    }

    void DestroyPlugin(Plugin *plugin){
        delete plugin;
    }
}

Plugin class should contain constructor, processing function and destructor. Processing function should accept at least 4 parameters: image as double array, width, height and additional parameters passed as json string.

plugin.h

class Plugin {
public:
    Plugin();
    double *process(double *, int, int, const char *params);
    ~Plugin();
};

plugin.cpp

#include "Plugin.h"
#include "json.hpp"
#include <iostream>

#include "json.hpp"

// for convenience
using json = nlohmann::json;

using namespace std;

double *Plugin::process(double *img, int w, int h, const char *params) {
    cout << "Processing image" << endl;
    cout << params << endl;
    auto j = json::parse(params);
    double factor = j["factor"].get<double>();
    for (int r = 0; r < w * h; r++) {
        img[r] = factor - img[r];
    }
    return img;
}

Plugin::Plugin() {
    cout << "Initializing plugin" << endl;
}

Plugin::~Plugin() {
    cout << "Destroying plugin" << endl;
}

References

1. http://bikulov.org/blog/2013/10/01/using-cuda-c-plus-plus-functions-in-python-via-star-dot-so-and-ctypes/