The amount of data generated in clinical practice is increasing substantially due to advances in computing, health care information technology, and policy [1,2]. Health records were historically paper-based. The Health Information Technology for Economic and Clinical Health Act, enacted into law in 2009, spurred the widespread adoption of electronic health records (EHRs) among health care providers in the United States [3,4]. The increasing availability of point-of-care clinical data in EHRs has benefited the use of real-world data (RWD) for real-world evidence (RWE) in biomedical research. RWD is observational data routinely collected in clinical settings, including EHR, health insurance claims, and patient-reported outcomes. RWD also includes data collected outside the traditional clinical setting, such as wearables and emerging data sources such as social media [5]. RWE is clinical evidence derived from RWD for a variety of applications, particularly regarding the use, effectiveness, and safety of medical products [6]. Early RWE efforts focused on structured EHR and insurance claims data [2]. Advances in analytical methods, such as AI, are expected to further enhance the ability to obtain deeper insights from RWD [7]. In this current paper, EHR data refer to structured and semistructured clinical data, including coded variables (eg, diagnoses, medications, and laboratory values) and unstructured clinical text. In parallel, more complex data types such as genomics and structured laboratory test results are increasingly integrated into RWE frameworks [8,9]. Most recently, real-world imaging data (RWiD) has emerged as a novel and valuable resource. RWiD is enabled by improvements in imaging infrastructure, data standardization, and deidentification technologies [10]. Although imaging studies are technically part of the EHR, they are treated here as a distinct data modality (RWiD) due to their unique technical, analytical, and infrastructural characteristics [11].

Medical imaging plays a key role in patient management by providing more definitive diagnostic information and guiding treatment decisions across multiple medical specialties (Figure 1) [12-14]. While its applications are wide-ranging, imaging is especially critical in therapeutic areas where it plays a central role in both routine care and research, most notably in oncology, cardiology, and neurology (Table 1) [15]. Typically, medical imaging includes x-ray, computed tomography (CT), magnetic resonance imaging (MRI), and ultrasound, as well as (hybrid) nuclear medicine modalities such as positron emission tomography and single-photon emission computed tomography [16].

As imaging technologies continue to evolve, the importance of imaging data in health care, both in routine clinical care and in research, is expected to grow even further. Medical imaging is foundational to clinical trials, and the emergence of RWiD is creating new opportunities for biopharmaceutical and life sciences research [17]. Medical imaging plays a central role in therapeutic development and clinical trials. In oncology, for example, standardized imaging criteria such as response evaluation criteria in solid tumors (RECIST) are widely used to evaluate treatment response and define trial endpoints [18]. More broadly, imaging provides quantitative and spatially resolved assessments of disease burden and progression. This information is difficult to capture using structured EHR or insurance claims data alone. These traditional data sources primarily reflect health care usage, diagnoses, and procedures, but often lack detailed phenotypic information, longitudinal disease characterization, and objective measures of treatment response [7]. However, while biopharmaceutical and life sciences researchers typically use traditional data modalities such as insurance claims and clinical notes (EHR), the use of RWiD is currently limited. The most important reason for the limited use is the lack of awareness and novelty of the availability of this data modality.

A recent review by Navarro-Garcia et al [19] highlights the growing role of RWiD in AI-driven biomarker development in oncology. The authors emphasize the value of large, heterogeneous imaging datasets derived from routine clinical care, while also noting key challenges related to data quality, standardization, and validation. The present editorial paper takes a broader perspective on RWiD, extending beyond AI-centric applications to include its role across the life cycle of biopharmaceutical research. In addition to biomarker discovery, we focus on practical considerations such as deidentification, data governance, and infrastructure, and outline applications in clinical trial design, external control arms, postmarketing surveillance, and drug repurposing. The goal of this editorial paper is to increase the awareness of RWiD for biopharmaceutical researchers.

Imaging datasets provide direct visual evidence of disease, enabling a deeper understanding of anatomical and physiological aspects compared to textual radiology reports alone. Relying solely on textual radiology reports results in limited contextual information. Subtle changes may be missed, and interpretation of images by radiologists is known to have varying degrees of interobserver variability. Other challenges include issues such as no use of standard radiology reporting formats, poor quality or incomplete radiology reports, and inability to visually confirm findings and validate the radiology reports [45].

Recent developments, such as improved imaging data handling and transfer technology, have catalyzed the ability to include the actual images in research projects [10]. Applying advanced analytics and AI methods to imaging datasets enables a variety of research and development applications [17].

Biomarker discovery encompasses the identification, validation, and characterization of measurable indicators that reflect specific biological states or conditions [57]. Biomarkers may be molecular, physiological, histological, genomic, or radiographic in nature, and are widely used for diagnosis, disease monitoring, and prognostication. Medical imaging contributes uniquely to this process by providing high-resolution phenotypic information that captures structural, functional, and morphological alterations associated with disease [58,59].

Imaging-based biomarkers offer spatially resolved and anatomically contextualized insights that are not available from molecular or genetic markers alone [60]. Spatial information can reveal where pathological processes occur within tissues or organs, such as regional variations captured through parametric mapping in cardiac MRI, and thereby improve mechanistic understanding. Imaging also enables longitudinal assessment, allowing quantitative tracking of biomarker dynamics in relation to disease progression or therapeutic response.

Several clinically meaningful quantitative metrics can only be derived from imaging. Examples include coronary artery calcium scores or psoas muscle area, which, although not routinely reported, can serve as robust prognostic biomarkers. Imaging modalities have also supported the discovery of novel biomarkers across disease domains. For instance, Mastrodicasa et al [61] demonstrated how CT angiography can yield imaging-derived biomarkers in aortic dissection, showing that partial false lumen thrombosis is a predictor of mortality in patients with Stanford type B dissections [60]. Similarly, Holland et al [62] leveraged large-scale optical coherence tomography datasets to train deep learning–based clustering models for biomarker discovery in early age–related macular degeneration [56]. Their approach rediscovered established biomarkers and identified a potentially novel hypertransmission phenotype [62].

Radiogenomics uses advanced image processing and machine learning to extract quantitative features from medical images such as texture, shape, and intensity. This is correlated with underlying genomic profiles, offering a noninvasive way to capture tumor heterogeneity and biological endpoints such as genetic mutations, microsatellite instability, and hypoxia. Recent studies used radiogenomics to predict complex molecular states directly from standard-of-care imaging. For instance, PRISM-CRC (Patho-Radiomic Integrative Survival Model–Colorectal Cancer) is a notable deep-learning framework that integrates radiology (CT), histopathology, and clinical data to predict microsatellite instability status and five-year disease-free survival in colorectal cancer with high accuracy (area under the receiver operating characteristic curve of 0.91) [63].

The increasing availability of large, curated imaging datasets further accelerates biomarker discovery by enabling data-driven pattern recognition and hypothesis generation through AI. As such, imaging continues to expand its role as an indispensable source of quantitative, contextualized, and longitudinal phenotypic information for biomarker research.

Characterizing the natural history of a disease is essential for elucidating its progression, phenotypic variability, and untreated outcomes. Such knowledge informs mechanistic understanding, identifies critical windows for therapeutic intervention, and establishes benchmarks for clinical endpoints. In rare diseases, however, natural history studies are often constrained by limited patient populations and geographic dispersion, resulting in small, heterogeneous cohorts. Novel diseases, such as COVID-19, present additional challenges due to initial data scarcity and limited clinical experience.

Imaging plays a central role in addressing these limitations by enabling a longitudinal, quantitative assessment of structural and functional changes. Serial imaging allows investigators to monitor disease evolution with high fidelity, supporting objective measurement of progression. For example, Yu et al [64] developed deep learning models to classify progression on chest radiographs [57]. They demonstrated that automated analysis can detect interval changes and emerging pathology in pulmonary disease [64].

Integration of imaging with clinical, molecular, and histopathologic data can further uncover mechanistic insights and novel biomarkers. Vanguri et al [65] used multimodal fusion of radiographic, pathological, and genomic features to identify biomarkers predictive of immunotherapy response in nonsmall cell lung cancer [58]. Their research illustrated how imaging augments biological interpretation in heterogeneous diseases [65].

Large-scale imaging datasets also facilitate the identification of disease subtypes and variability. Jiang et al [66] leveraged brain MRI and machine learning to define four biotypes of focal temporal lobe epilepsy, each characterized by distinct neuroanatomical signatures associated with divergent long-term seizure outcomes. Functional and multiparametric imaging modalities, including ultrasound, echocardiography, and perfusion imaging, provide complementary physiological information. This enables a more holistic understanding of disease processes.

Finally, imaging-derived quantitative metrics reduce reliance on subjective assessment and improve reproducibility in natural history studies. In Duchenne muscular dystrophy, Rooney et al [67] used MRI-derived muscle biomarkers to model disease trajectory, identifying metrics capable of monitoring therapeutic effects and predicting functional outcomes. Collectively, these examples highlight imaging’s essential role in advancing natural history research. This is especially true for diseases where data scarcity or biological heterogeneity poses substantial challenges.

Effective clinical trial design requires careful selection of endpoints, patient populations, and methodological approaches to ensure validity, reliability, and statistical power. Although RWiD is typically retrospective in nature, it can inform prospective clinical trial design by providing insights into disease characterization, outcome variability, and imaging-derived biomarkers. In this context, RWiD can be considered complementary to, rather than a replacement for, randomized clinical trials.

The US Food and Drug Administration has signaled openness to novel evidence pathways that incorporate digital biomarkers, AI-derived endpoints, and nontraditional data modalities as complements to randomized trial data, including for external control arms and postmarketing commitments [6]. Imaging increasingly contributes standardized, quantitative endpoints to these frameworks. Therefore, RWiD is positioned to play a more important role in regulatory submissions.

Large imaging datasets enable the development and validation of AI models for detection, diagnosis, disease monitoring, and progression assessment. Such models can streamline trial operations through automated or semiautomated image analysis. This reduces variability and increases throughput. Imaging enables stratification of patients into biologically meaningful subgroups, supporting enrichment strategies and improving trial efficiency. For example, the MRI-derived subtypes of focal temporal lobe epilepsy identified by Jiang et al [66] could inform stratification strategies that ensure appropriate representation of biologically distinct subgroups. In oncology, Tomaszewski et al [68] demonstrated that computational analysis of preaccrual CT scans in a failed soft-tissue sarcoma trial (SARC021) could have enabled patient enrichment and potentially allowed the study to meet its primary survival endpoint.

Incorporating imaging into trial design can also support the selection of sensitive and specific biomarkers that reflect disease progression or therapeutic response. Metrics such as hippocampal volume on brain MRI in Alzheimer disease have been widely used as imaging endpoints [62]. They may also help refine inclusion and exclusion criteria to identify patients most likely to benefit from investigational therapies [69]. More broadly, imaging-derived quantitative measures, such as aortic aneurysm diameter, tumor volume, or breast tissue density, provide objective endpoints. They reduce subjective interpretation and enhance reproducibility across sites.

Collectively, these capabilities position imaging as a critical component of modern clinical trial design. They enable more precise patient selection, more informative endpoints, and more efficient trial execution.

Identifying eligible participants is a foundational step in clinical trial execution. It requires locating individuals who meet specific demographic, clinical, and phenotypic criteria. Traditional approaches, such as patient registries, partnerships with health care providers, and EHR mining, can help surface potential candidates. However, they may miss important biological or imaging-defined subgroups [70].

Imaging can be used as a screening tool to identify eligible participants at scale, particularly when eligibility depends on phenotypic features not captured in structured clinical data. As demonstrated by Jiang et al [66], MRI-based stratification uncovered four biotypes of focal temporal lobe epilepsy that share similar clinical profiles but differ markedly in neuroanatomical phenotype, highlighting how imaging can refine inclusion and exclusion criteria.

Imaging also supports more nuanced patient stratification based on predicted therapeutic response. In a retrospective analysis of multiple clinical trials in nonsmall cell lung cancer, Dercle et al [71] applied radiomics to standard-of-care CT scans to identify patients with differential sensitivity to treatments such as nivolumab, docetaxel, and gefitinib. Their findings suggest that imaging-derived signatures could enhance clinical decision-making and improve survival prediction in heterogeneous patient populations.

Beyond improving selection accuracy, imaging can accelerate large-scale screening, particularly when combined with advanced AI methods. AI models have been developed to automatically detect and characterize a wide range of conditions, including interstitial lung disease [72], vascular pathology [61], and oncological abnormalities [73]. These tools can rapidly filter imaging datasets to identify individuals meeting trial eligibility criteria, thereby reducing the burden of manual review. Imaging may be especially valuable for detecting patients with rare diseases who might otherwise be overlooked in conventional recruitment pipelines. Wojtara et al [74] describe in a review paper that recent advancements in medical imaging have enabled researchers to train AI models based on large datasets and subsequently fine‐tune these models on smaller datasets typically associated with rare diseases.

Together, these capabilities underscore the value of imaging data in expanding and refining participant identification strategies, ultimately supporting more efficient and targeted clinical trial enrollment.

An external control arm is a comparator group derived from data sources outside the randomized clinical trial, such as real-world datasets, historical cohorts, or prior observational studies [75]. This approach is particularly valuable when enrolling a traditional control group is impractical or ethically constrained, including in rare diseases and oncology. Notably, approximately 199 of 659 (30%) rare disease trials are terminated prematurely, most often due to slow patient accrual [76]. By replacing or reducing the need for prospectively enrolled placebo or control patients, external control arms can improve trial feasibility, reduce participant burden, while maintaining the ability to evaluate treatment safety and efficacy.

RWiD can significantly strengthen the external control arm construction. Imaging provides standardized, longitudinal measurements that support consistent outcome assessment and comparability between cohorts. For instance, CT imaging in idiopathic pulmonary fibrosis can detect disease progression before forced vital capacity decline or overt clinical worsening, ensuring more accurate characterization of control populations [77]. Comprehensive imaging-derived phenotypes, such as lesion size, location, morphology, and temporal trajectory, enhance the ability to match patients in the investigational and external cohorts, thereby improving comparability.

Incorporating imaging may also reduce bias. Traditional external controls often rely heavily on clinical metrics or patient-reported outcomes. This may introduce measurement inconsistencies or misclassification. Imaging-derived quantitative measures offer greater objectivity and reproducibility, strengthening internal validity and improving confidence in treatment-effect estimation. Shukla-Dave et al [78] demonstrated that RWiD increased dataset diversity and representation compared to traditional clinical trials [79]. Thereby, RWiD enabled the identification and mitigation of selection biases that cause inaccurate model performance.

Collectively, integrating RWiD into external control arm development can enhance precision, mitigate bias, and support more efficient and ethically feasible clinical trial designs.

Effective postmarketing surveillance relies on the integration of multiple RWD sources, including clinical records (EHR), laboratory data, and outcomes information. RWiD is most informative when linked with other data modalities rather than used in isolation. Postmarketing surveillance is essential for evaluating the safety, effectiveness, and real-world performance of drugs and medical devices following regulatory approval. By collecting data from clinicians, patients, and health systems, postmarketing surveillance efforts can detect adverse events, identify long-term outcome patterns, and inform regulatory actions such as labeling updates or the initiation of additional studies [80].

RWiD can enhance postmarketing evaluation when integrated with other RWD sources [81]. Imaging contributes detailed, objective assessments of anatomical and functional changes that complement clinical, laboratory, and outcomes data, enabling more precise monitoring of treatment effects. RWiD also offers a means to evaluate performance degradation in AI-based diagnostic devices by providing diverse, real-world image distributions over time. When linked with other data sources, such as EHR data, imaging supports a more comprehensive characterization of patient trajectories, including treatment exposure, comorbidities, and longitudinal outcomes.

Imaging further provides visual and quantitative evidence for the early detection of subtle physiological changes that may signal emerging risks or therapeutic impacts. For example, reductions in spleen volume can serve as an early predictor of radiation-induced leukopenia. Steinhelfer et al [82] demonstrated that automated splenic volume assessment in patients receiving 177Lu-DOTATATE therapy outperformed conventional laboratory parameters in predicting long-term leukopenia risk. As splenic volume is rarely measured during routine clinical care, access to raw imaging is critical for enabling such analyses.

Overall, the incorporation of RWiD into postmarketing surveillance frameworks can strengthen signal detection, refine risk assessment, and improve long-term monitoring of therapeutic and device performance.

Drug repurposing and repositioning aim to identify new therapeutic applications for existing pharmacologic agents beyond their originally approved indications. This strategy can accelerate development timelines, reduce costs, and capitalize on established safety profiles.

RWiD can meaningfully contribute to repurposing efforts by enabling direct visualization and quantification of drug effects across tissues and organ systems. Imaging allows investigators to identify and validate off-label therapeutic benefits that may not be apparent from clinical or laboratory data alone. For example, although metformin is primarily used as a first-line therapy for type 2 diabetes, imaging-based studies have shown that metformin use may slow abdominal aortic diameter expansion and reduce mortality risk following aortic aneurysm repair [83].

RWiD also enhances mechanistic understanding by capturing spatially resolved and temporal drug responses. When integrated into multimodal datasets, including EHRs, transcriptomic and proteomic profiles, and other digital health information, imaging can reveal biological pathways and phenotypic effects. These insights are relevant to new therapeutic hypotheses. Yin and Wong [84], for instance, describe multimodal pipelines for drug repositioning in Alzheimer disease and related dementias that leverage imaging alongside molecular and clinical data to identify promising candidates and understand their potential mechanisms of action.

Collectively, RWiD provides complementary phenotypic evidence that strengthens drug repurposing pipelines, supporting both discovery and validation of novel therapeutic uses.

In conclusion, RWiD is an emerging and increasingly valuable asset for biopharma and life sciences. Challenges such as data harmonization, storage and transfer, computational demands, and deidentification remain. However, datasets that include actual images provide richer insight than radiology reports alone, which are optimized for clinical decision-making rather than research. When incorporated into broader RWD frameworks, RWiD enables biomarker discovery, characterization of disease natural history, more efficient clinical trial design, external control arms, participant identification, postmarketing evidence generation, and opportunities for drug repurposing and repositioning. As infrastructure and standards mature, the role of imaging in RWE generation will continue to expand.