Usability of PETSIRD, the PET Raw Data open format of the Emission Tomography Standardization Initiative (ETSI): results from ETSI’s first hackathon

Introduction: With the advent of Positron Emission Tomography (PET) scanners of increasingly higher sensitivity alongside the evolving capacity of Artificial Intelligence (AI) to support Nuclear Medicine, the opportunity for accessing high fidelity information through standardized PET raw data has become apparent. The Emission Tomography Standardization Initiative (ETSI) was founded in 2022 from an international consortium of nuclear medicine experts to define open, extendable, standardized and vendor-agnostic formats of emission tomography raw data. Currently, the initiative focuses on developing the PET ETSI Raw Data (PETSIRD) list-mode format comprised of (i) data elements (e.g. coincidence events, geometry, correction factors, etc.), (ii) a container describing the element’s structure and the protocols for access, storage and transfer using Microsoft Research YARDL meta-language, and (iii) a use-cases software library. We report on the progress of developing use-case applications for generating, accessing and processing PETSIRD data.

Methods: On November 13-14^th 2023, ETSI organized its in Vancouver, BC, Canada, aiming at the development of proof-of-concept open-source software for PETSIRD use-cases to demonstrate its usability. A group of ~30 international participants from industry and academia worked together to produce software tools showcasing how to create, access and process PETSIRD data for well-defined user tasks in quantitative PET imaging. Specifically, four subgroups were formed to work on the following tasks: (1) apply split/merge, gating and sub-sampling operations on PETSIRD data according to a user-defined or physiological signal (e.g. ECG) input, (2) build a convertor from GATE Monte Carlo simulations output data in ROOT format to PETSIRD and subsequently generate simulated demo PETSIRD data from realistic Monte Carlo simulations, (3) add a PETSIRD read/write/processing interface within existing open-source reconstruction software, and (4) analytically simulate PETSIRD data from images as well as independently reconstruct PET images from PETSIRD data alone. In all use-cases, the highly versatile YARDL container model of PETSIRD was exploited.

Results: All four subgroups were able to make substantial progress during the 2-day hackathon and delivered useful and practical software tools demonstrating PETSIRD’s high usability and functionality. Specifically, subgroup 1 implemented tools for both sub-sampling and gating operations on PETSIRD data guided by external inputs. Subgroup 2 built a ROOT-to-PETSIRD converter, to be used as a possible roadmap for future converters from vendor proprietary formats to PETSIRD and generated realistic demo PETSIRD data of a simplified NEMA IEC phantom. Subgroup 3 added basic PETSIRD input/output interfaces to STIR, CASToR and PyTomography open-source libraries. Subgroup 4 used PARALLELPROJ’s open-source projectors to analytically generate synthetic PETSIRD data from images. Finally, subgroup 4 successfully reconstructed an estimate of the NEMA phantom PET image from the PETSIRD demo data of subgroup 2, thus independently validating the respective software of the two subgroups and the format’s capability to contain sufficient information for standalone reconstruction. The role of PETSIRD’s YARDL container was pivotal in all developments, as its automatic code generation capabilities in both C++ and Python enabled rapid software prototyping. Prototype PETSIRD definitions are available on . All code will be made available there.

Conclusions: The successful implementation of several practical and functional use cases during ETSI’s first hackathon demonstrated PETSIRD’s usability as a standalone, open-source, standardized PET list-mode data format. Moreover, it highlighted its value as a vendor- and scanner-agnostic standard to aid the development of novel and high-impact PET applications and analysis tools especially of those relying on data sharing and harmonization principles such as AI models.

• Download figure
• Open in new tab
• Download powerpoint