The study was evaluated by our Institutional Review Board, and the requirement for informed consent was waived. The four hospitals involved were as follows: ASST Grande Ospedale Metropolitano Niguarda, Fondazione IRCCS Ca’ Granda Ospedale Maggiore Policlinico, ASST Fatebenefratelli Sacco and Ospedale dei Bambini V. Buzzi. The first three are general hospitals, while the last one is paediatric.
To analyse the dosimetric archive, the associated large amount of data was clustered according to the RadLex® playbook [15] as in the previously cited papers [9,10,11].
Data collected in 2017 were first analysed according to facilities, age and sex, to get descriptive statistics. In a second step, the distributions of dosimetric quantities were compared with values from the literature to check the strength of the cloud database. A systematic comparison with currently available reference levels was beyond the scope of this work. The same data were also used to assess the quantity of studies with series not matching with the original requirement and to evaluate their effect on dosimetric quantities.
Description of the RDIM software and cloud server architecture
The four hospitals were equipped with the RDIM software Bracco Injeneering’s NEXO [DOSE]® (Bracco Injeneering S.A., Lausanne, Switzerland), developed by PACSHealth, LLC, integrated with the different PACS of each hospital (Agfa, Fuji, Carestream). NEXO [DOSE]® is a web-based software which collects patient information (age, sex, etc.) and dosimetric data.
Relevant data could be extracted from different sources: digital imaging and communications in medicine (DICOM) header (for each series), patient protocol (overall exam) and radiation dose structured report. Dealing with several CT scanners, the most suitable data sources were chosen for each device.
Data from the hospitals were collected both in local servers and in a cloud one.
Each individual institution has complied with its internal procedures to ensure the highest level of security and privacy of patient personal data. These procedures required the appointment of a person in charge of data processing, in this case, an external subject, the specification of access methods and the definition of the persons authorised to access the data, who undertake to behave in absolute confidentiality. In the case of the cloud server, data were anonymised prior to leaving the local site server following DICOM PS3.15 [16] and Integrating the Healthcare Enterprise Radiation Exposure Monitoring RAD-63 profiles [17]. Patient data were removed, replaced, or modified in accordance with the reference DICOM standard. There were some exceptions, such as patient characteristics (age, sex, height, weight) and device information (facility, device, exposure parameters), for relevant data needed for statistical aims and analysis. Only the relevant data were transmitted, not the entire studies.
Finally, all data were collected in the Cloud NEXO [DOSE]® Server – Microsoft Azure for Healthcare, in compliance with Health Insurance Portability and Accountability Act, International Organization for Standardization and European Union data protection directives, received via DICOM over transport layer security (TLS).
Each hospital could access the cloud database, and the software was able to report, for each exam, patient demographics (age, sex), and scan protocol information (CTDIvol, DLP), with all the previously anonymised sensitive data. More detailed information relative to the single series could also be retrieved.
Global descriptive analysis
The analysis of the whole data stored in the cloud server can give a general description of the distribution of radiological exams among the population depending on facilities, age and sex.
Through NEXO [DOSE]®, data were filtered according to facility, device, age, sex, and other characteristics in order to develop descriptive statistics of parameters relevant for the risk associated to radiation exposure, such as the percentages of males and females undergoing exams and the distribution of the number of patients as a function of age. Moreover, the number of exams of the different hospitals was tracked.
A first retrospective analysis regarding CT exams performed during 2017 was carried out.
Detailed analysis of CT studies
Clustering
The RadLex® playbook is a project of the Radiological Society of North America [8] that provides a standardised system for naming radiological procedures. As in other studies [9,10,11], the RadLex® playbook was used to cluster a great quantity of data for the subsequent analysis.
The arrangement of exams in homogeneous groups, according to scan region and acquisition task, is difficult due to the differences in types of CT scanner, radiology information systems (RIS) and picture archiving and communication system (PACS). We solved this problem using the radiological information stored in the different DICOM tags and inside the hospital reporting database. In two hospitals, the DICOM tag “study description” (0008,1030) generated by the RIS was used; in another hospital, the same tag generated by the scanner was considered, whereas in the last one, the DICOM tag “protocol name” (0018,1030) compiled with the scanner protocol name was chosen.
A study common name (SCN) from the RadLex® playbook was associated to each description or protocol name to classify exams in a consistent way, according to scan region and use of contrast media, as represented in general terms in Fig. 1. Since the procedures within a RadLex® label should have homogeneous exposure parameters, data organised in this way were used to analyse population and dosimetric quantities in a consistent way in order to evaluate the different radiological procedures.
Through NEXO [DOSE]®, data were filtered according to SCN and both studies and series were exported in Excel format. Files relative to studies include mean CTDIvol and total DLP, allowing to evaluate their distributions and to calculate median values, 25th and 75th percentiles. Variations of DLP with sex were also evaluated.
In the first phase of the study, 76,171 exams on adult patients (age range 18–109 years), the most frequent ones without administration of contrast agent (without contrast, “WO”) were considered, in particular, those belonging to SCNs “CT Head WO” (35%, 27,030 exams), “CT Chest WO” (9%, 6,635 exams), and “CT Abd/Pelv WO” (2%, 1,778 exams) in order to evaluate radiation exposure in different body regions. This analysis was performed at first with the whole data in the cloud, excluding those studies with coarse problems in the data transfer from the DICOM tag. In the case of “CT Chest WO” and “CT Abd/Pelv WO”, only data from three hospitals were analysed since the few studies of the fourth one were not enough for statistical aims. CTDIvol and DLP values were depicted through histograms, including in the highest class values higher than the depicted range, but to check the strength of this clustering, only median values of dosimetric quantities were compared with recent DRLs, as indicated by ICRP 135 [2].
Check of clustered data
In the second phase of the study, a more accurate check of data within the SCNs was performed through series analysis following the methodology proposed in other publications [12, 13]. Studies including series not in agreement with the SCN in terms of anatomical region or use of contrast media (exams not in line with the SCN) were quantified and removed from the subsequent analysis. For example, within the SCN “CT Head WO” studies including series descriptions as “TorAdd 3.0 B40f”, “C_Spine”, “HeadAngio 0.75 H30f”, and “Spine 2.0 B30s” were found and removed; the same for description as “HeadSeq 4.8 H31s” within the SCN “CT Chest WO”.
The analysis of CTDIvol and DLP was thus repeated only with the studies in line with the SCNs in order to evaluate potential changes in median values and to test the strength of the database.