Opportunistic Screening at Abdominal CT: Use of Automated Body Composition Biomarkers for Added Cardiometabolic Value

Abstract

Abdominal CT is a frequently performed imaging examination for a wide variety of clinical indications. In addition to the immediate reason for scanning, each CT examination contains robust additional data on body composition that generally go unused in routine clinical practice. There is now growing interest in harnessing this additional information. Prime examples of cardiometabolic information include measurement of bone mineral density for osteoporosis screening, quantification of aortic calcium for assessment of cardiovascular risk, quantification of visceral fat for evaluation of metabolic syndrome, assessment of muscle bulk and density for diagnosis of sarcopenia, and quantification of liver fat for assessment of hepatic steatosis. All of these relevant biometric measures can now be fully automated through the use of artificial intelligence algorithms, which provide rapid and objective assessment and allow large-scale population-based screening. Initial investigations into these measures of body composition have demonstrated promising performance for prediction of future adverse events that matches or exceeds the best available clinical prediction models, particularly when these CT-based measures are used in combination. In this review, the concept of CT-based opportunistic screening is discussed, and an overview of the various automated biomarkers that can be derived from essentially all abdominal CT examinations is provided, drawing heavily on the authors' experience. As radiology transitions from a volume-based to a value-based practice, opportunistic screening represents a promising example of adding value to services that are already provided. If the potentially high added value of these objective CT-based automated measures is ultimately confirmed in subsequent investigations, this opportunistic screening approach could be considered for intentional CT-based screening. ^©RSNA, 2021.

Graphical abstract

Figure 1.

Automated CT-based cardiometabolic tools for assessment of (second row of images, left to right) bone, aortic calcium, visceral to subcutaneous fat ratio, muscle attenuation, and liver attenuation biomarkers from original abdominal CT data. In practice, a visual correlate allows for quality assurance for the automated segmentation results in individual patients. The specific CT biomarkers shown have all been validated in prior works. (Adapted and reprinted under a CC BY license from reference .)

Figure 2a.

(a) CT-based assessment of bone mineral density with the use of trabecular attenuation at the L1 vertebral level. Axial CT images show the L1 vertebral level (top row) in adult patients of a variety of ages. Magnified axial CT views show the L1 vertebra in soft tissue (second row) and bone (third row) windows. Standard manual placement of the ROI for measurement of trabecular attenuation and the mean attenuation value in the ROI are shown in the third row. Automated ROI placement was programmed to match this location. Sagittal reconstruction images (bottom row, soft-tissue [left] and bone [right] windows) show placement of the ROI (yellow oval) at the L1 vertebra. Typically, trabecular attenuation values progressively decrease with increasing patient age. The loss of bone mineral density is more apparent with the soft-tissue window. (Reprinted, with permission, from reference .) (b) Graph (left) shows normative reference CT-based L1 trabecular attenuation values based on more than 20 000 examinations. The mean attenuation values show that age-related L1 trabecular bone loss is fairly linear. Error bars indicate standard deviations, which are remarkably uniform throughout the age spectrum. Table (right) shows the median and the mean (± standard deviation [SD] ) values for L1 trabecular attenuation for each age group. These normative reference ranges, which are derived from a combination of manual and automated measurements, can serve as a quick reference for radiologists when reading body CT examinations performed for other clinical indications. Note that these values apply to scanning at 120 kVp. (Reprinted, with permission, from reference .)

Figure 2b.

(a) CT-based assessment of bone mineral density with the use of trabecular attenuation at the L1 vertebral level. Axial CT images show the L1 vertebral level (top row) in adult patients of a variety of ages. Magnified axial CT views show the L1 vertebra in soft tissue (second row) and bone (third row) windows. Standard manual placement of the ROI for measurement of trabecular attenuation and the mean attenuation value in the ROI are shown in the third row. Automated ROI placement was programmed to match this location. Sagittal reconstruction images (bottom row, soft-tissue [left] and bone [right] windows) show placement of the ROI (yellow oval) at the L1 vertebra. Typically, trabecular attenuation values progressively decrease with increasing patient age. The loss of bone mineral density is more apparent with the soft-tissue window. (Reprinted, with permission, from reference .) (b) Graph (left) shows normative reference CT-based L1 trabecular attenuation values based on more than 20 000 examinations. The mean attenuation values show that age-related L1 trabecular bone loss is fairly linear. Error bars indicate standard deviations, which are remarkably uniform throughout the age spectrum. Table (right) shows the median and the mean (± standard deviation [SD] ) values for L1 trabecular attenuation for each age group. These normative reference ranges, which are derived from a combination of manual and automated measurements, can serve as a quick reference for radiologists when reading body CT examinations performed for other clinical indications. Note that these values apply to scanning at 120 kVp. (Reprinted, with permission, from reference .)

Figure 3.

Opportunistic CT measurement of L1 trabecular attenuation in an asymptomatic 76-year-old woman who presented with abdominal pain due to mild pancreatitis. Axial contrast-enhanced CT image (left) shows the relative placement of the automated bone mineral density ROI (green oval) in the anterior trabecular space of the L1 vertebral body. An attenuation value of 100 HU indicates low bone mineral density. Placement of the ROI was intended to match that placed with the previously established manual approach. Sagittal CT image (middle) from an examination performed 8 years later for unintended weight loss shows manual placement of the L1 ROI and an attenuation value of 81 HU, which indicates osteoporosis. An interval vertebral compression fracture at T11 (arrow) is now seen. Note also the depiction of abdominal aortic calcium quantification (red shading) between the L1 and L4 vertebral levels. Frontal radiograph (right) obtained 9 years after the initial CT examination shows a left hip fracture (arrow), the risk for which could have been identified at an earlier CT examination.

Figure 4.

CT for prediction of an osteoporotic fracture in an asymptomatic 59-year-old woman undergoing colorectal cancer screening. Index axial CT images (top row, left and middle) show results from automated assessment of bone with the automated placement of the bone mineral density ROI (green oval, top row, left) (63 HU) and muscle (−1.7 HU) attenuation (red shading, top row, middle), which were in the 99th and 98th percentiles, respectively, relative to the entire screening study cohort. However, the 10-year FRAX scores for this patient were 6.7% for any fracture and 0.5% for hip fracture, which are results well below the recommended treatment threshold. Frontal radiograph (top row, right) from 3 months later shows that the patient sustained a left femoral neck fracture. Axial noncontrast CT images (bottom row) from multiple prior examinations for urolithiasis over the years show, in retrospect, a progressive decrease in L1 bone attenuation at the ROI (green oval, automated measures shown). Unfortunately, this information is typically not considered in routine clinical practice for CT examinations performed for other indications. (Adapted and reprinted, with permission, from reference .)

Figure 5a.

Automated detection of a vertebral fracture with two different algorithms in two patients. (a) Composite CT image in a 62-year-old woman with a severe osteoporotic compression fracture of the L1 vertebral body shows the use of an algorithm for automated detection and characterization of fractures that involves the use of a height compass approach (red and green lines, left). The geometric arrangement of the compass-like layout consists of a central circular sector surrounded by two ring-shaped finite thickness concentric bands. The sagittal CT image (left) shows vertebral column segmentation and partitioning. The adjacent image (right) shows the stacked-height compass of the entire vertebral column. The circular images show the height compasses for a grade-3 wedge fracture at L1 (middle) and preserved vertebral height at T12 (top) and L5 (bottom). The patient underwent spine DXA less than 2 months earlier that was interpreted as normal (T-score, −0.5). (b) Sagittal CT image in an 85-year-old woman shows a severe T11 compression deformity (yellow rectangle), which was detected with a different automated deep-learning algorithm. (Fig 5b courtesy of Einav Blumenfeld, Zebra Medical Systems.)

Figure 5b.

Automated detection of a vertebral fracture with two different algorithms in two patients. (a) Composite CT image in a 62-year-old woman with a severe osteoporotic compression fracture of the L1 vertebral body shows the use of an algorithm for automated detection and characterization of fractures that involves the use of a height compass approach (red and green lines, left). The geometric arrangement of the compass-like layout consists of a central circular sector surrounded by two ring-shaped finite thickness concentric bands. The sagittal CT image (left) shows vertebral column segmentation and partitioning. The adjacent image (right) shows the stacked-height compass of the entire vertebral column. The circular images show the height compasses for a grade-3 wedge fracture at L1 (middle) and preserved vertebral height at T12 (top) and L5 (bottom). The patient underwent spine DXA less than 2 months earlier that was interpreted as normal (T-score, −0.5). (b) Sagittal CT image in an 85-year-old woman shows a severe T11 compression deformity (yellow rectangle), which was detected with a different automated deep-learning algorithm. (Fig 5b courtesy of Einav Blumenfeld, Zebra Medical Systems.)

Figure 6a.

Quantification of abdominal aortic calcification with a semiautomated CT tool. (a) Coronal CT images in two different asymptomatic adults show segmentation of aortic calcium (blue shading) with a semiautomated software tool originally devised for coronary artery calcium scoring. Abdominal aortic calcium is segmented in an ROI that is specified by the user. (b) Superimposed receiver operating characteristic curves for cardiovascular events occurring within a 2-year period after CT show a corresponding area under the curve of 0.82 for abdominal aortic calcification (AAC) and 0.64 for the FRS. Adding the FRS to the AAC did not improve predictive performance (area under the curve, 0.82). (Fig 6b reprinted, with permission, from reference .)

Figure 6b.

Quantification of abdominal aortic calcification with a semiautomated CT tool. (a) Coronal CT images in two different asymptomatic adults show segmentation of aortic calcium (blue shading) with a semiautomated software tool originally devised for coronary artery calcium scoring. Abdominal aortic calcium is segmented in an ROI that is specified by the user. (b) Superimposed receiver operating characteristic curves for cardiovascular events occurring within a 2-year period after CT show a corresponding area under the curve of 0.82 for abdominal aortic calcification (AAC) and 0.64 for the FRS. Adding the FRS to the AAC did not improve predictive performance (area under the curve, 0.82). (Fig 6b reprinted, with permission, from reference .)

Figure 7.

Change in the automated abdominal aortic calcium level over time in an asymptomatic 60-year-old man undergoing noncontrast CT for colonographic screening. Axial image from index CT examination (left) shows moderate hard plaque involving the abdominal aorta, which corresponds to an automated Agatston score of 1514. Axial CT images acquired 5 years later (middle and right) show an interval increase in hard plaque (red shading, right) that corresponds to an automated Agatston score of 5070. The patient experienced a myocardial infarction 4 years after the second CT examination and developed congestive heart failure 3 years after that.

Figure 8.

Bar graph shows automated population-based assessment of abdominal aortic calcium scoring at noncontrast CT and progressive increases in mean abdominal aortic calcium scores from a generally healthy screening population of nearly 10 000 adults. Agatston scores in women lag by nearly a decade behind those for men. More than 30% of all individuals had an Agatston score of zero, more than 50% had a score less than 100, and nearly 70% had a score less than 300, whereas more than 30% had a score greater than 300. Of note, greater than 90% of this screening cohort was younger than 70 years of age, which explains the lower frequency of higher Agatston scores. (Reprinted, with permission, from reference .)

Figure 9.

Kaplan-Meier time-to-death plots by quartile for, A, automated CT-based aortic calcium scores and, B, FRSs in a cohort of asymptomatic adults. Note that the separation for the highest-risk quartile (Q4, corresponding to an Agatston score > 492) is greater for automated aortic calcium scoring alone than that with the multivariate FRS. This difference is also reflected in the larger hazard ratio for aortic calcification than that for the FRS (4.5 vs 2.8). (Adapted and reprinted under a CC BY license from reference .)

Figure 10.

Combining automated CT tools for prediction of future adverse events in an asymptomatic 52-year-old man undergoing CT colonography (CTC) for colorectal cancer screening. A, Coronal (left) and axial (middle and right) screening images show calcification (red shading, left), the visceral (blue shading, middle) to subcutaneous (red shading, middle) fat ratio, and liver attenuation (green shading). This patient had a body mass index of 27.3 and an FRS of 5% (low risk). However, several CT-based metabolic markers were indicative of underlying disease (values and percentiles provided). The patient had an acute myocardial infarction 3 years after this initial CT examination and died 12 years after the initial CT examination at the age of 64 years. B, Axial (left), coronal (middle), and sagittal (right) contrast-enhanced CT images acquired for minor trauma 7 months before death were interpreted as negative for abnormalities but show substantial progression of vascular calcification, visceral fat, and hepatic steatosis. (Adapted and reprinted under a CC BY license from reference .)

Figure 11.

Automated body composition tools for metabolic syndrome in an asymptomatic 60-year-old woman. Coronal noncontrast (left) and axial (right) CT images show liver attenuation (green shading), aortic calcium (red shading, left), and abdominal segmentation and quantification of visceral (blue shading, right) and subcutaneous (red shading, right) fat. In this case, automated mean liver attenuation was 34 HU (95th percentile), which indicates moderate to severe steatosis. The automated Agatston score was 2781 (92nd percentile), and the automated L1-level visceral fat area was 326 cm² (95th percentile for women, 91st percentile overall). This patient had a myocardial infarction 3 years after the CT examination and died within 1 year after that.

Figure 12.

Non-gated coronary calcium scoring. Noncontrast CT image obtained without gating in a 71-year-old man shows automated segmentation of left coronary artery calcification (orange shading) with a validated deep-learning algorithm. The Agatston score was 611. Although most abdominal CT examinations do not include imaging the heart in its entirety, even partial coverage showing a calcium score greater than 400 signifies a high risk for a future adverse event. (Case courtesy of Einav Blumenfeld, Zebra Medical Systems.)

Figure 13.

Visceral and subcutaneous fat segmentation and quantification at CT in two asymptomatic obese adults with similar body mass index levels but different body habitus. Axial CT image (left), which was acquired with the use of the automated fat tool in a 67-year-old man, shows a relative abundance of visceral fat (blue shading) compared with subcutaneous fat (red shading), whereas the other axial CT image (right), for which an earlier semiautomated fat tool was used at the umbilical level in a 52-year-old woman, shows a relative abundance of subcutaneous fat (orange shading) in comparison to visceral fat (light blue shading). These findings show the difference between so-called apple- and pear-shaped body habitus, respectively. This difference can be quantified with the visceral to subcutaneous fat ratio. The increased visceral to subcutaneous fat ratio for the patient on the left portends a higher cardiovascular risk.

Figure 14.

Density plots comparing the measurement of subcutaneous, visceral, and total adipose tissue according to sex. These density plots are derived from a large asymptomatic adult population and reflect automated CT-based measures at the L1 level. Note the relative distribution of subcutaneous and visceral fat between women and men, with men having more visceral fat than women, on average. The data on the y axis are the relative frequency of adipose tissue in the population studied. (Reprinted, with permission, from reference .)

Figure 15a.

Kaplan-Meier time-to-death plots by quartile for automated CT-based body mass index (BMI) (a) and visceral to subcutaneous fat ratio (b) in a cohort of asymptomatic adults. Note how the separation among all four quartiles for visceral to subcutaneous fat ratio is substantially better than that for BMI, in which there is little difference among quartiles. This difference is also reflected in the larger hazard ratio (HR) between the highest-risk quartile (Q4) and the other three quartiles for the visceral to subcutaneous fat ratio compared with those for body mass index (2.3 vs 1.4). (Adapted and reprinted under a CC BY license from reference .)

Figure 15b.

Kaplan-Meier time-to-death plots by quartile for automated CT-based body mass index (BMI) (a) and visceral to subcutaneous fat ratio (b) in a cohort of asymptomatic adults. Note how the separation among all four quartiles for visceral to subcutaneous fat ratio is substantially better than that for BMI, in which there is little difference among quartiles. This difference is also reflected in the larger hazard ratio (HR) between the highest-risk quartile (Q4) and the other three quartiles for the visceral to subcutaneous fat ratio compared with those for body mass index (2.3 vs 1.4). (Adapted and reprinted under a CC BY license from reference .)

Figure 16a.

Automated muscle segmentation at abdominal CT. (a) Axial CT image at the L3 level in a 50-year-old man shows the automatically segmented muscle (red shading). (b) Graph of automated CT-based muscle area and attenuation differences according to subject age shows that muscle attenuation (red line) decreases at a greater rate with aging compared with the cross-sectional area (blue line). (c) Graph shows that this trend exists for both men and women. After age 70, both muscle attenuation and area values plateau more in women than in men. (Reprinted, with permission, from reference .)

Figure 16b.

Automated muscle segmentation at abdominal CT. (a) Axial CT image at the L3 level in a 50-year-old man shows the automatically segmented muscle (red shading). (b) Graph of automated CT-based muscle area and attenuation differences according to subject age shows that muscle attenuation (red line) decreases at a greater rate with aging compared with the cross-sectional area (blue line). (c) Graph shows that this trend exists for both men and women. After age 70, both muscle attenuation and area values plateau more in women than in men. (Reprinted, with permission, from reference .)

Figure 16c.

Automated muscle segmentation at abdominal CT. (a) Axial CT image at the L3 level in a 50-year-old man shows the automatically segmented muscle (red shading). (b) Graph of automated CT-based muscle area and attenuation differences according to subject age shows that muscle attenuation (red line) decreases at a greater rate with aging compared with the cross-sectional area (blue line). (c) Graph shows that this trend exists for both men and women. After age 70, both muscle attenuation and area values plateau more in women than in men. (Reprinted, with permission, from reference .)

Figure 17.

Sarcopenia for prediction of future hip fractures. ROC curves (top) for prediction of hip fractures over a 2-year period show that automated CT-based measurement of muscle attenuation in Hounsfield units (HU) alone (top left, AUC = 0.75) surpasses the multivariate FRAX score (top right, AUC = 0.73). When bone and muscle attenuation are combined (not shown), the performance is further improved (AUC = 0.76). Axial CT image (lower left) shows automated L3-level muscle segmentation in a 54-year-old asymptomatic woman. Mean muscle attenuation was −1.2 HU (98th percentile). Axial CT image (lower right) shows aortic calcium (red shading) with an automated Agatston score of 4283 (96th percentile). The patient had a myocardial infarction 3 months later, had a left hip fracture 1 year after that, and died 3 years later at age 58. (ROC curves reprinted, with permission, from reference .)

Figure 18a.

Automated versus manual liver fat quantification at noncontrast CT. (a) CT images at the same level show manual (left) and automated (right) methods for measuring liver attenuation (yellow circle, left; green shading, right) in an asymptomatic 60-year-old man with severe steatosis. (b) Scatterplot shows that the results of the manual ROI technique agree well with those of the automated volumetric approach. (Adapted and reprinted, with permission, from reference .)

Figure 18b.

Automated versus manual liver fat quantification at noncontrast CT. (a) CT images at the same level show manual (left) and automated (right) methods for measuring liver attenuation (yellow circle, left; green shading, right) in an asymptomatic 60-year-old man with severe steatosis. (b) Scatterplot shows that the results of the manual ROI technique agree well with those of the automated volumetric approach. (Adapted and reprinted, with permission, from reference .)

Figure 19.

Changes in liver fat content over time. Axial noncontrast CT images acquired over a decade in an asymptomatic man who was 51 years of age at the time of initial imaging show substantial variation in liver attenuation in the ROI (yellow circle) corresponding to a liver fat fraction change of greater than 20%. Note the excellent agreement between the automated and manual measures of liver attenuation. Also note how little body mass index (BMI) has changed over time. (Reprinted, with permission, from reference .)

Figure 20.

Semiautomated measurement of the liver surface nodularity score for detection of hepatic fibrosis and cirrhosis. Axial contrast-enhanced CT image (top left) of the liver in a patient with a hepatitis C virus infection who had not previously been evaluated with biopsy or noninvasive methods for liver fibrosis shows a cirrhotic liver with a nodular surface and trace ascites. The blue box shows the area that is magnified (magnification, ×2.5) and appears in the upper right image, which shows how the liver is manually painted with the green brush stroke. Axial contrast-enhanced CT image (bottom left) shows automatic detection of the liver. The yellow box indicates the area that is magnified in the corresponding bottom right image. (Magnification, ×30.) The green line represents the automatically detected liver surface, and the red line is a smoothed polynomial line (spline) fit to represent a smooth liver surface. The distances between the detected edge and spline are measured on a pixel-by-pixel basis, with approximately 100 measurements per section (bottom right) and 1000 measurements for the recommended 10 sections (not shown). The liver surface nodularity score in this case was 4.8, which is considered severe cirrhosis. (Case courtesy of Andrew D. Smith, MD.)

Figure 21.

Automated hepatosplenic volumetry for improved assessment of hepatomegaly and splenomegaly. A, Coronal CT image in a 59-year-old woman shows a bulbous-appearing liver, but the maximal craniocaudal length was less than 20 cm. However, the automated liver volume was 2573 cm³, which is well above the upper limit of normal, derived as two standard deviations higher than the mean, as we describe in a currently unpublished study of an asymptomatic cohort. The segmented liver is shown on the inset image. B, Similarly, coronal CT image in a 45-year-old man shows a long-axis splenic measurement of greater than 12 cm, but the automated splenic volume of 277 cm³ is well within the upper limit of the normal range and was established in a similar way. The segmented spleen is shown on the inset image. Note the normal crescentic shape of the spleen.