Reconstruction as a service: a data space for off-site image reconstruction in magnetic particle imaging

TL;DR

The paper addresses the challenge of SM-based MPI requiring extensive, on-site calibration by proposing a data space architecture aligned with IDS RAM to enable off-site calibration data and image reconstruction as a service. It defines a minimum information model using MDF as the vocabulary, analyzes use-cases including online reconstruction, retrospective reprocessing, maintenance analysis, and ML data collection, and maps data-space roles to MPI stakeholders. The approach aims to reduce downtime for scanners, provide access to updated reconstruction algorithms, and facilitate maintenance and training data generation, with potential applicability to other imaging modalities. This work lays the groundwork for a practical, governance-enabled infrastructure that could enhance clinical workflows and enable flexible reconstruction pipelines in MPI.

Abstract

Magnetic particle imaging (MPI) is an emerging medical imaging modality which offers a unique combination of high temporal and spatial resolution, sensitivity and biocompatibility. For system-matrix (SM) based image reconstruction in MPI, a huge amount of calibration data needs to be acquired prior to reconstruction in a time-consuming procedure. Conventionally, the data is recorded on-site inside the scanning device, which significantly limits the time that the scanning device is available for patient care in a clinical setting. Due to its size, handling the calibration data can be challenging. To solve these issues of recording and handling the data, data spaces could be used, as it has been shown that the calibration data can be measured in dedicated devices off-site. We propose a data space aimed at improving the efficiency of SM-based image reconstruction in MPI. The data space consists of imaging facilities, calibration data providers and reconstruction experts. Its specifications follow the reference architecture model of international data spaces (IDS). Use-cases of image reconstruction in MPI are formulated. The stakeholders and tasks are listed and mapped to the terminology of IDS. The signal chain in MPI is analysed to identify a minimum information model which is used by the data space.