realtimefmri

Real-time collection and preprocessing of functional magnetic resonance imaging data along with stimulus presentation. This software allows you to construct custom real-time data analysis pipelines without much fuss.

It consists of three parts: a collection system that interfaces with the Siemens MRI scanner to make brain volumes available as soon as they are acquired, a preprocessing system that can be configured to do common operations like motion correction and detrending, and a stimulation system that can present the data to the operator or subject in a variety of ways.

1. Data collection

2. Preprocessing

Images are passed through a series of preprocessing steps specified in the preprocessing configuration file.

Configuring your own pipeline

Preprocessing steps are subclasses of the parent class, PreprocessingStep. The only requirement is that they implement a .run() method that performs some operation on input.

Configuration file

The overall preprocessing pipeline is specified in a YAML configuration file saved in the ~/.config/realtimefmri/pipelines. The configuration file should contain at least one dictionary that has pipeline and a list of steps. An individual step has the following format:

• Required keys:
  • name (string): a descriptive name for this step
  • step (string): YAML tag specifying a python object (one that subclasses PreprocessingStep), which executes the step
  • input (list of strings): string keys to the data_dict specifying which of the entries should be passed as inputs to the .run() method for this step
• Optional keys:
  • kwargs (dictionary): values provided upon initialization of the step
  • output (list of strings): keys that will be paired with returned values and added to the data_dict when this step is finished
  • send (list of strings): keys indicating values from data_dict that will be sent to the stimulation code when this step is finished

Here is an example of a simple preprocessing step, the first step in fact, which converts input Dicom image into a more usable Nifti volume.

- name: dicom_to_nifti
  step: realtimefmri.core.preprocessing.DicomToNifti
  input:
    - raw_image_binary
  output:
    - image_nifti
  send:
    - image_nifti

The send field is optional. Here, it indicates that the Nifti volume should be sent along to the stimulation code.

3. Stimulation

Things get interesting when we make stimuli that rely on data gathered in real-time. All you need to do is build some software that manages a zmq SUB socket subscribed to one of the topics published by the preprocessing code. This can be implemented in any language that has has a zmq library, which is pretty much any modern language. Some generally useful stimuli are included in this library including a pycortex viewer.

Timing

The main processes (collection, preprocessing, stimulation) are able to run on separate machines, which could have different clocks. The real-time computer receives a pulse whenever a volume is acquired. This time stamp is sent alongside each volume in the code.