Opportunistic Screening with Imaging

The growing volume of clinical imaging and the emergence of artificial intelligence technologies present a unique opportunity to extract additional value from existing imaging data that would otherwise go unused, providing patients with health benefits beyond the original imaging purpose through value-added opportunistic screening. By targeting clinically significant diseases and major public health concerns, opportunistic screening can enhance existing risk assessment, prevention, and treatment paradigms, expanding radiology’s reach and impact for individual patients and at the population level.

Key points

•
Opportunistic screening (OS) leverages existing imaging data that go unused in clinical practice to improve health.
•
OS targets high-value clinical diseases with established screening and treatment strategies to improve presymptomatic disease detection, risk stratification, and clinical outcomes.
•
OS extracts value from unused imaging data and is fundamentally different than “incidentaloma” management that can result in low value care.
•
Artificial intelligence tools will accelerate the capabilities of OS and the scope and impact of radiology.
•
OS can be integrated into multimodal disease risk prediction approaches and augment existing clinical risk models.

Abbreviations

AAC	abdominal aortic calcification
AI	artificial intelligence
ASCVD	atherosclerotic CVD
CAC	coronary artery calcification
CT	computed tomography
CTBA	CT biological age
CVD	cardiovascular disease
CXRs	chest radiographs
DL	deep learning
DXA	dual-energy X-ray absorptiometry
HCAs	health care applications
LSVR	liver segmental volume redistribution
MASLD	metabolic dysfunction-associated steatotic liver disease
MRI PDFF	MR imaging proton density fat fraction
OS	opportunistic screening
OSCAR	Opportunistic Screening Consortium in Abdominal Radiology
PCP	primary care provider
SAT	subcutaneous fat
T2D	type 2 diabetes
VAT	visceral adipose tissue
VBQ	vertebral bone quality score

Introduction

Imaging routinely captures valuable tissue and organ body composition data beyond the intended purpose of the examination, providing a unique opportunity to gain health insights that have been historically underappreciated and underutilized. Opportunistic screening (OS) generally refers to the practice of “exercising prevention through an unorganized program or chance encounter.” In radiology, OS involves purposefully leveraging unused imaging data to address health needs, add value, and provide health benefit beyond the primary indication. Although terms like resourceful, fortuitous, serendipitous, and even “opportuneful” apply, opportunistic has endured. OS offers a unique opportunity to identify presymptomatic disease and risk factors for major causes of human disease that could translate to improved longevity, quality of life, and clinical outcomes, particularly when early lifestyle modification or other meaningful interventions are implemented. Indeed, a more fitting description may be “value-added opportunistic screening.” ^,

In this review, we provide an overview of OS with imaging with an emphasis on high impact clinical use cases, highlighting diseases that have established prevention programs and treatments. Examples of OS are presented along with a discussion of emerging artificial intelligence (AI) tools, showing how AI can enable and expand the reach and impact of radiology through OS. Finally, we address challenges, successes, and future directions of OS in clinical practice.

Opportunistic screening: high-value clinical targets and population impact

The origins of OS in radiology may date back to the early 1990s with OS for abdominal aortic aneurysm using ultrasound, but it was not until the 2010s that the concept gained traction, particularly through OS for osteoporosis using computed tomography (CT). OS targets common, clinically relevant diseases that are public health concerns including cardiovascular disease (CVD), osteoporosis, type 2 diabetes (T2D), obesity, metabolic disorders, and chronic liver disease ( Fig. 1 , Table 1 ) that have known risk factors with established screening protocols and management strategies. This approach requires no additional imaging time, specialized equipment, or patient radiation dose (eg, CT and radiography).

Two panels show image-based opportunistic screening clinical use cases and a timeline chart illustrating the growth of imaging-related opportunistic screening publications over time. — Fig. 1

Table 1

Image biomarkers for opportunistic screening

Tissue or Organ	Imaging Biomarker	High-value Clinical Use	Modalities
Bone	Vertebral trabecular attenuation, femoral neck DXA equivalent	Osteoporosis, spine surgical planning	CT, MR imaging, and radiography
Muscle	Muscle attenuation, muscle area	Sarcopenia, sarcopenic obesity, fragility fracture, and frailty	CT and MR imaging
Fat	Visceral and subcutaneous fat area and attenuation, ectopic fat (organ fat deposition)	ASCVD, MASLD, and mortality risk	CT and MR imaging
Cardiovascular	CAC, AAC, and heart size	ASCVD, heart failure, and abdominal aortic aneurysm	CT, MR imaging, and radiography
Liver	Attenuation, liver surface nodularity, liver segmental volume ratio	MASLD and fibrosis	CT and MR imaging
Pancreas	Attenuation	T2D and MASLD	CT and MR imaging
Spleen	Volume	Cirrhosis and portal hypertension	CT and MR imaging
Kidneys	Volume	T2D	CT and MR imaging

Abbreviations : ASCVD, atherosclerotic cardiovascular disease; DXA, dual-energy X-ray absorptiometry; MASLD, metabolic dysfunction-associated steatotic liver disease; T2D, type 2 diabetes.

OS is aligned with core screening principles. It targets high prevalence diseases that represent significant public health concerns and have effective and widely available treatment with potential for improved outcomes through early detection. Tests should reliably detect preclinical disease, be safe, accessible, and show benefit. OS offers a value-adding opportunity for radiology and has been identified as a way for radiology to engage in and improve population health outcomes.

The transformative role of artificial intelligence

AI tools have streamlined onerous time-intensive tasks like image segmentation ( Fig. 2 ) and enabled multimodal predictions, overcoming historical barriers and paving a way for large-scale OS implementation. Although “black box” AI models are rapid, efficient, and accurate, their lack of interpretability limits adoption for health care applications (HCAs). In contrast, explainable models are understandable to human users, improve accountability, allow for quality assurance, ^, and improve understanding of AI model predictions, which explains trends toward prioritizing explainability. For example, explainable deep learning (DL) tools can rapidly extract valuable objective body composition measures from CT scans ( Fig. 3 ) offering underappreciated insights into cardiometabolic health and unsuspected disease, ^,^, which can lead to actionable opportunities for prevention and intervention.

A grid of seven axial computed tomography images at the L 3 vertebral level shows automated deep learning segmentation of fat, muscle, abdominal aortic calcification, bone, liver, and spleen using color overlays. — Fig. 2

A workflow diagram shows an explainable automated deep learning pipeline that processes chest, abdomen, and pelvis C T scans to extract multiple quantitative imaging biomarkers. — Fig. 3

Imaging modalities

Although OS can be applied to any imaging modality, CT dominates the OS landscape due to its high-volume use, role as a first-line modality for the assessment of acute and chronic disorders, and accessibility. The value of CT for OS is related to the objective and reproducible nature of CT and the advent of automated AI tools that can rapidly extract explainable predictive biomarkers, supporting personalized medicine and value-added initiatives. Although a strong argument can be made for CT as the optimal modality for OS, radiography (eg, osteoporosis and CVD ) offers a cost-effective and accessible alternative, and MR imaging (eg, osteoporosis, ^,^, mortality, and CVD ; Fig. 4 ) shows promise despite higher cost and lower availability.

A three-part infographic shows a deep learning framework for whole-body M R I segmentation, dataset demographics from U K Biobank and N A K O, and analysis methods for predicting mortality. — Fig. 4

Targets for opportunistic screening: high-value clinical use cases

Osteoporosis

Osteoporosis, characterized by reduced bone mass and quality, leads to bone fragility and increased fracture risk, but remains underdiagnosed and undertreated, affecting 13% of US adults aged over 50 years. Although bone attenuation is not a direct bone mineral density (BMD) equivalent, CT is ideal for bone assessment because it can easily distinguish between cortical and trabecular bone with trabecular bone being more metabolically active and primarily affected by metabolic bone disease ( Fig. 5 ). CT has proven to be a more accurate predictor of osteoporosis than the reference standard dual-energy X-ray absorptiometry (DXA), with stronger performance for detecting vertebral compression fractures, and showing promise for OS among different patient populations ( Fig. 6 ). ^,^,^,^,^,^,

Three panels of an elderly woman show low lumbar bone attenuation on C T, automated body composition overlays, and a later pelvic radiograph with a left proximal femur fracture. — Fig. 5

Three sagittal C T of the thoracic spine in three women with osteoporosis show automated detection of vertebral fractures and bone mineral density levels. — Fig. 6

Despite measurement variability across studies and some variation in bone attenuation with tube voltage (eg, lower energies producing higher attenuation values), simple L1 trabecular attenuation measures correlate with DXA and osteoporotic vertebral compression fractures. A study of adults undergoing abdominal CT and DXA showed an attenuation threshold of 110 HU (at 120 kV) had 90% specificity for DXA-based osteoporosis and substantial increases in fragility fracture risk have been observed with attenuation less than 90 HU in adults aged over 65 years. Practically, an L1 attenuation of 100 HU (at 120 kV) suggests osteoporosis and higher fracture risk ( Fig. 7 ). ^,

Five panels show axial C T slices with automated body composition overlays, and a later pelvic radiograph demonstrates a displaced left proximal femur fracture in an elderly woman. — Fig. 7

Using T1 signal intensity from standard T1-weighted sequences, MR imaging can assess bone quality from marrow cellularity and fat content. Methods like signal-to-noise ratio (SNR)-based M-score and cerebrospinal fluid (CSF)-normalized vertebral bone quality score (VBQ) are moderately correlated with DXA T-scores for differentiating healthy from osteopenic and osteoporotic bone. Like CT, an L1 level approach with VBQ is a simple, rapid, and effective approach for predicting osteopenia and osteoporosis with MR imaging. ^,

Cardiovascular Disease

CVD is the most common cause of death worldwide making early detection, risk stratification, and prevention major public health priorities. Coronary artery calcification (CAC) and abdominal aortic calcification (AAC) are key OS biomarkers for CVD risk. CAC is a key predictor of atherosclerotic CVD (ASCVD) events ( Fig. 8 ) and is considered the most predictive single CV risk marker in asymptomatic patients. Incidental CAC is found in 34% to 63% patients undergoing noncardiac CT and can be identified without electrocardiogram (ECG)-gating ( Fig. 9 ). It offers superior risk discrimination and reclassification of risk compared to models like pooled cohort equations in certain patient populations. ^,^, Adding CAC to the 2023 PREVENT equations also improves myocardial infarction prediction. Recently, the NOTIFY-1 prospective quality improvement project found that notifying ordering providers and patients of incidental CAC on nongated chest CT increased statin prescriptions, showing its impact on preventative care.

Two panels with axial C T images of the chest showing extensive coronary artery calcifications in a 42-year-old patient, highlighted by an automated A I tool to show vascular calcification. — Fig. 8

Two panels with axial non-contrast C T images of the chest showing incidental coronary artery calcification in a 69-year-old woman after a fall, highlighted with a color overlay in the left anterior descending artery. — Fig. 9

Compared to CAC, AAC is an underappreciated but valuable marker that can identify individuals at an increased risk for CV events and all-cause mortality. ^,^, A study of 3599 undergoing noncontrast abdominal CT and cardiac CT for CAC assessment showed that the presence of any AAC was predictive of CAC and associated with a 2 fold risk of major adverse cardiovascular events (MACE). DL tools make rapid quantification of AAC feasible on noncontrast as well as portal venous CT, expanding the scope of OS beyond what is possible with CAC ( Fig. 10 ). Soft tissue biomarkers such as muscle attenuation (ie, myosteatosis), visceral fat, and hepatic steatosis associated with CV disease risk ^,^,^, are emerging targets for OS. ^,

Three panels show abdominal aortic calcification with automated color overlays on coronal and axial C T images in a 65-year-old man who presented to the emergency department with abdominal pain. — Fig. 10

DL models can now estimate CVD risk from chest radiographs (CXRs). A recent DL CXR CVD-Risk model identified individuals at high 10 year risk of MACE and improved on traditional risk scores in patients with known ASCVD, highlighting potential for population-level OS using CXRs.

Obesity, Diabetes, and Metabolic Syndrome

Obesity is a global epidemic and major health crisis, with adult obesity rates more than doubling over the past 3 decades. These trends are alarming given the association between obesity and T2D, metabolic syndrome, CVD, and cancer, which contribute to adverse health outcomes and rising health care costs. Metabolic syndrome affects one-third of adults in the United States but is underrecognized due to insufficient screening. It is defined by 3 of 5 factors: elevated waist circumference, blood pressure, fasting glucose, serum triglycerides, and reduced high-density lipoprotein cholesterol. Although CT biomarkers such as visceral adipose tissue (VAT), subcutaneous adipose tissue (SAT), hepatic steatosis, and muscle can identify at-risk individuals and track phenotypic changes of metabolic syndrome, imaging features are not included in the definition of metabolic syndrome.

Recent studies have shown that AI-driven CT biomarkers can identify patients with T2D. Patients with T2D have lower pancreatic CT attenuation and higher VAT. CT predictor variables also include intrapancreatic fat, AAC, and hepatic steatosis ( Fig. 11 ). Additionally, abdominal CT biomarkers change significantly with glycemic control ( Fig. 12 ) and can track treatment changes in patients undergoing glucagon-like peptide-1 receptor agonist treatment ( Fig. 13 ).