Five disciplines along the path of a radiology report

A report is delivered when it lands in the practice software the referring doctor actually uses, filed against the right patient and the right clinician. I measure our results service that way rather than by counting messages sent. The work is unglamorous: matching each referrer's provider number and location, retiring channels that misroute, and chasing messages that practice software rejected. Urgent and unexpected findings are different. Our radiologists telephone the referrer or the doctor on call, and the system records that the call happened; no message replaces that conversation. We also upload reports to My Health Record, where patients and their other clinicians can see them, so I watch upload failures as closely as messaging failures. When a practice changes software or a doctor moves, my team should know before the first report goes astray.

Most of our integration failures have not been faulty interfaces. They have been correct messages carrying wrong identities: a patient registered twice, a referrer recorded at a practice they had left, a hospital order number that did not match its scan. I therefore treat patient matching, healthcare identifiers and the provider directory as clinical infrastructure with named owners and weekly exception reports. We exchange HL7 version 2 messages with practice software, hospital systems and our own RIS, and we are adding FHIR-based exchange where partners support it. Every new connection passes conformance testing with real examples, including the awkward ones: amended reports, cancelled examinations and studies split into several parts. I would rather delay a hospital connection by a month than open it with unresolved identity mismatches, because a misfiled result harms a clinician who cannot see the error.

Imaging carries an awkward security problem. A CT or MRI scanner may remain clinically sound for fifteen years while the operating system on its console loses support much sooner, and altering it outside the manufacturer's process can compromise the device's regulated configuration. We cannot patch every console, so we isolate them: modality networks are segmented, remote vendor access is brokered and recorded, and nothing on a scanner network can browse the internet. Health information is sensitive information under Australia's Privacy Act, and the notifiable data breaches scheme requires us to assess suspected breaches promptly and notify people likely to suffer serious harm. I report cyber risk to our audit and risk committee each quarter with the discipline expected of financial risk. We rehearse downtime twice a year at a clinic, because a ransomware plan nobody has practised is a document, not a capability.

Between 2022 and 2024 we moved an archive of about 1.9 petabytes from two Perth data centres to cloud storage hosted in Australia. The seven-year economics favoured the move, but the clinical question was different: what happens in Geraldton or Kalgoorlie when the link fails during a busy list? We kept a local cache at every clinic holding recent studies and the comparison images for that day's booked patients, which cost about A$1.3 million more than a pure cloud design. Retention obligations for medical records, and the radiologists' need for old comparisons, meant we could not tier data by age alone. I negotiated retrieval charges explicitly, because an archive that is cheap to store and expensive to read penalises the clinical use it exists for. Each clinic moved separately, with radiologists confirming comparison studies before every cut-over.

Software that detects findings on medical images is regulated by the Therapeutic Goods Administration as a medical device, and we use only products included on the ARTG, Australia's register of therapeutic goods, for the purpose we intend. Inclusion begins our assessment rather than ending it. Before any clinical use we test a tool on our own scanners, protocols and patient population, and the radiologists' clinical governance committee decides its scope. Our rule is that software may change the order in which studies are read or draw attention to a region, but a radiologist reads every study and signs every report. We review performance monthly, including findings the software missed, and we tell referring hospitals what it does and does not do. Any proposal to let software clear apparently normal studies is a clinical decision for radiologists, not a technology upgrade.