Introduction

Research in the field of MR image reconstruction has made continuous advances over the last decades. Input to image reconstruction is raw MR data as well as k-space data, which is often hardly accessible on clinical MRI systems and comes in many different vendor specific formats. Towards the goal of reproducible research, the ISMRM has taken steps to implement a discipline-specific, open raw data file format – the ISMRM raw data format (ISMRMRD) ¹. The purpose of this format is to facilitate exchange of raw data along with the required physics parameters of the acquisition process to provide all information necessary for vendor agnostic image reconstruction. The ISMRMRD format is based on the widely used HDF5 file format and has programming libraries available for C++, Matlab and Python, as well as data converters for GE, Siemens, Philips and Bruker datasets. Developing, debugging and maintaining of related software, such as file format converters, sequence raw data labeling, raw data processing, as well as reconstruction algorithms often requires developers to visually inspect the acquisition headers and data chunks contained in the ISMRMRD files, which can easily grow to sizes of tens of gigabytes in case of diffusion or fMRI experiments. Finding labeling or data inconsistencies can then often results in “find the needle in a haystack” operations, wasting precious development time. While there are several free HDF5 data exploration tools available, the ISMRMRD ecosystem lacks a specific, fast raw data browser. In this work, we present such a tool, which allows fast browsing through large ISMRMRD datasets. The presented cross platform ISMRMRD viewing tool is written in Python, freely available as open-source project on GitHub ² and builds on the available ISMRMRD programming libraries.

Implementation

An ISMRMRD file consists of a flexible XML header and a set of fixed structures for the actual coil and trajectory data. Each acquisition, which typically corresponds to one k-space line, is described by a fixed size data header consisting of several encoding counters and various data specific numbers such as number of samples, channels and acquisition dwell time. These fixed size headers are most conveniently displayed in a table with each column representing a header field and each row corresponding to an acquisition. A model-view-viewmodel software architecture based on the PyQt library ³ was selected for implementation. An intermediate data buffer in the viewmodel class enables browsing through very large files without running into performance and memory issues. Plotting of individual acquisitions (coil and trajectory data), triggered by selecting a corresponding row in the header table, is implemented using the PyQtGraph library ⁴. The data can be displayed in various forms (magnitude, real, imaginary, FFT, phase), easily extendable and selectable from a drop down list above the plot area. Figure 1 shows the main application window and Figure 2 a close-up of the acquisition header table, delivering fast access to all encoding counters and header fields.

Results and Discussion

A fast ISMRMRD file browser was implemented which can easily handle large datasets of tens of gigabytes. Opening a 2.6 GB diffusion dataset (64 diffusion direction, 217800 acquisitions) takes less than 1.46 seconds on an Intel Core i7-6700 @ 3.4GHz with 480 GB Intel SSD DC S3500 disk. It show all acquisition header fields in a table and selected coil and trajectory data (if available) can be visualized in a plot area (see Figure 3). This makes it a very useful tool for all activities involving ISMRM raw data. Further developments will involve more viewing capabilities as well as adaptation to newer versions of the ISMRMRD format.

Acknowledgements

No acknowledgement found.