Real-World Imaging Data Expands Research Possibilities

Real-world imaging data is emerging as a distinct resource for biomedical research, adding the actual images collected during routine care to more familiar clinical datasets. A recent article published in JMIR Medical Informatics links wider use to improvements in imaging infrastructure, standardisation and de-identification. Medical imaging supports screening, diagnosis, treatment planning, response assessment and surveillance across specialties including oncology, cardiology and neurology. Unlike radiology reports, image files retain detailed anatomical, functional and quantitative information. Their use, however, introduces substantial technical, methodological and governance demands that must be addressed before reliable evidence can be generated. 

Must Read: Radiology AI Moves Inside the Diagnostic Viewer 

Challenges in Standardising Imaging Data 

Medical images are commonly stored in digital imaging and communications in medicine files, which combine image pixels with metadata about the examination and equipment. Although the format supports exchange, imaging appearance and quantitative measurements can still vary across scanner vendors, hardware, acquisition settings, contrast administration and reconstruction methods. These differences can reduce comparability across institutions and introduce systematic bias into multicentre analyses. Vendor-specific metadata and local implementation practices add further heterogeneity. 

Harmonisation may reduce technical variation, but it requires caution. Statistical methods can remove scanner or protocol effects, yet inappropriate use may also remove clinically meaningful signals. This risk is particularly important when disease status is closely linked to a specific site or scanner, when cohorts are small or when one institution represents a distinct patient group. Standardised protocols are therefore valuable where prospective coordination is possible. Retrospective work depends more heavily on detailed metadata, sensitivity analyses and external validation across diverse institutions and populations. 

Selection bias remains another limitation because imaging is ordered for clinical reasons rather than at random. Patients undergoing advanced examinations may differ from those who do not in disease severity, comorbidities, access to care or other characteristics. Data from tertiary and academic centres may also overrepresent complex cases. Linking images with electronic health records, laboratory information, outcomes and insurance claims can add clinical context and support adjustment for measured differences. These methods cannot fully remove unmeasured confounding, so causal conclusions require restraint. 

Why Actual Images Add Research Value 

Radiology reports are designed to answer clinical questions, not to capture every feature that may later matter to research. They may omit details about shape, texture, surrounding tissues, acquisition protocols or findings outside the reason for the examination. A chest computed tomography scan performed for lung cancer screening, for example, may also contain information relevant to cardiovascular disease or osteoporosis that is not systematically included in the report. 

Report quality may vary because of time pressure, differing expertise, non-standard formats and interobserver variability. Different radiologists can interpret the same study differently, creating inconsistency in retrospective datasets. Actual images allow findings to be reassessed using a defined protocol, trained readers or automated methods. They also help confirm whether required acquisition settings were used and support consistent definitions of disease progression across institutions and time periods. 

These advantages bring greater operational complexity. Imaging files are large, difficult to transfer and computationally demanding. Efficient use depends on structured metadata, vendor-neutral archives and systems that link images with other clinical variables. De-identification must address both metadata and pixels because identifying information may be embedded within images or recoverable from facial structures in head imaging. In Europe, this work must comply with the General Data Protection Regulation. Federated learning may limit central data aggregation, but still requires local computing capacity, consistent curation, secure parameter exchange and compatible infrastructure. 

Applications Across Evidence Generation 

Real-world imaging data can support biomarker discovery by providing spatially resolved and longitudinal information about disease. Images capture structural, functional and physiological changes that may not appear in clinical records, laboratory results or genetic data. Quantitative measures such as calcium scores, muscle area, tissue volume and lesion characteristics may provide prognostic information even when they are not routinely reported. Advanced image processing can also connect imaging features with genomic or pathological patterns. 

Longitudinal imaging can improve understanding of disease natural history, particularly where patient numbers are limited or phenotypes are heterogeneous. Serial examinations allow objective tracking of progression and may help identify disease subtypes, treatment-response patterns and clinically meaningful changes. Integration with molecular, pathological and clinical information can link visible changes with underlying biological characteristics. 

Clinical trial planning is another potential application. Retrospective imaging can inform endpoint selection, patient stratification, eligibility criteria and expected outcome variability. Automated or semiautomated analysis may reduce reader variability and increase throughput. Imaging-derived phenotypes can identify participants whose relevant features are not captured in structured records. 

External control arms may also benefit from objective longitudinal measurements that improve matching between groups. In postmarketing surveillance, linked imaging and clinical data can support monitoring of treatment effects, emerging risks and diagnostic-device performance. Raw images may reveal subtle changes that are not routinely measured. Imaging may also contribute to drug repurposing by showing effects across tissues and organs that are not apparent from clinical or laboratory data alone. 

Real-world imaging data offers information that radiology reports and structured clinical records cannot provide alone. Its value lies in direct access to detailed, quantitative and longitudinal evidence from routine care. Reliable use depends on careful cohort design, harmonisation, secure storage, validated de-identification, appropriate infrastructure and transparent acknowledgement of bias and residual confounding. When images are linked with clinical, laboratory and outcomes data, they can support biomarker research, disease characterisation, trial planning, external controls, surveillance and drug repurposing. They remain a complementary evidence source rather than a substitute for randomised trials or broader clinical datasets. 

Source: JMIR Medical Informatics 

Image Credit: iStock