The PET Raw Data Standardization Initiative

Introduction: In modern PET scanners, listmode acquisition records the raw unprocessed data that are produced during a scan. Listmode files encapsulate information on all of the coincident gamma detection events as well as associated data. For historical and practical reasons, listmode files have been considered as an intermediate format, retained only transiently to reconstruct images. These DICOM formatted images comprise the current standard for reading, storing, and sharing PET data. As our field progresses, and data processing and storage capacities continue to expand, we can reconsider the role of listmode data, and our practices for handling the high-fidelity information they contain. Our goals are to define an open, extendable, standardized and vendor-agnostic description of PET listmode and associated data; define a standard for storage and transmission of that data; and develop a standardized description of software to access and manipulate the standardized data, as well as prototype software. Such a standard will facilitate a new paradigm for PET innovation, including new opportunities for inter-scanner and inter-vendor harmonization and AI applications, and aide in development of advances in novel PET applications and analysis tools.

Methods: A working group was created with the support of the SNMMI’s PIDS council that brought together experts in PET imaging from multiple academic centers and from industry. A first organizational meeting was held in February 2022 and regular meetings have been held thereafter. Initial efforts have focused on identifying the data elements that will form the content of the standard and a methodology for defining the data objects and structure. Development of use cases is also planned to provide both testing and examples of applications which archive, access and manipulate the data enclosed within the standard.

Results: The "Data Elements" subgroup has defined a content description that is compatible with the scanners and scan-acquisition approach of most of the major PET vendors. The content includes the coincidence listmode data itself; a scanner geometry description which allows going beyond current cylindrical scanners; periodic summary information for data corrections such as singles histograms and dead-time signals for correction of dead-time and randoms; normalization information; and concurrent physiologic signals for identifying cardiac contraction and respiratory motion.

The "Container" subgroup has been developing a meta-language, called Yardl, for defining the structure of the data and protocols for accessing, transferring and storing the data. Automated tools can then be used to process the meta-language definition and generate subroutines to support manipulation of the data in multiple programming languages, starting with C++ and Python. It also allows for different formulations to be created such as storage in an HDF5 file format or transmission on a wire as a binary data stream. A single meta-language human-readable definition will facilitate maintenance and updating of the standard as well as verification of the validity of any proposed changes. The meta-language approach and associated tooling are also applicable to other modalities and are, for example, under consideration for a revision of the MR Data standard.

A working prototype of the meta-language standard definition, software for processing the definition into C++ subroutines and example simple applications of the subroutines are available in open source on GitHub. The standard is still currently under development and will be updated as its content is refined and the use of the standard is evaluated.

Conclusions: The development of a PET raw data standard is bringing together industry and academic experts to open the doors on access to the richness inherent in listmode information. We believe that this will facilitate harmonization across platforms, spark innovative applications and advance the capability and success of PET imaging.