Value-added Opportunistic CT Screening: State of the Art

Abstract

Opportunistic CT screening leverages robust imaging data embedded within abdominal and thoracic scans that are generally unrelated to the specific clinical indication and have heretofore gone largely unused. This incidental imaging information may prove beneficial to patients in terms of wellness, prevention, risk profiling, and presymptomatic detection of relevant disease. The growing interest in CT-based opportunistic screening relates to a confluence of factors: the objective and generalizable nature of CT-based body composition measures, the emergence of fully automated explainable AI solutions, the sheer volume of body CT scans performed, and the increasing emphasis on precision medicine and value-added initiatives. With a systematic approach to body composition and other useful CT markers, initial evidence suggests that their ability to help radiologists assess biologic age and predict future adverse cardiometabolic events rivals even the best available clinical reference standards. Emerging data suggest that standalone "intended" CT screening over an unorganized opportunistic approach may be justified, especially when combined with established cancer screening. This review will discuss the current status of opportunistic CT screening, including specific body composition markers and the various disease processes that may be impacted. The remaining hurdles to widespread clinical adoption include generalization to more diverse patient populations, disparate technical settings, and reimbursement.

Graphical abstract

Figure 1:

Case examples of fully automated CT-based body composition measures from six different older adult patients (over 60 years of age). Noncontrast (top row) and postcontrast (bottom row) CT images at the L1 vertebral level demonstrate artificial intelligence–based segmentation of skeletal muscle, visceral fat, subcutaneous fat, aortic calcium, liver, and spleen. The green region of interest is intended for assessment of trabecular bone mineral density. These visual correlates allow for rapid quality assurance for correct tissue segmentation, although multisection review would be needed to fully confirm some measures. In addition, some adjustments are indicated to correct for the effect of intravenous contrast media on certain measures. Use of the L1 level for bone, fat, and muscle assessment allow for use of both abdominal and chest CT examinations. Note that automated algorithms for segmenting the pancreas, kidneys, and other structures exist but are not depicted here.

Figure 2:

Opportunistic detection of extracolonic pathologic findings at CT colonography (CTC) screening. Top row: Supine image from CTC screening in an asymptomatic 63-year-old man (left) shows an unsuspected 5.5-cm left adrenal mass (arrow), which measured 27 HU (indeterminate). Subsequent contrast-enhanced CT scan (not shown) with adrenal protocol showed little or no washout (<10%). On opposed-phase chemical shift MRI scan (middle), the adrenal mass (arrow) fails to show signal dropout. Subsequent CT-guided core needle biopsy (right, arrow) was suggestive of adrenocortical carcinoma, which was confirmed after adrenalectomy with negative margins. The patient is doing well without evidence of disease over 13 years later. Bottom row: Supine image from CTC screening in an asymptomatic 52-year-old man (left) shows an unsuspected infrarenal abdominal aortic aneurysm (arrow) that measured up to 6.8 cm in anteroposterior diameter, as well as a horseshoe kidney that was previously unknown. Subsequent work-up included CT angiography (middle) within a week and aortoiliac stent graft repair (right) within 1 month of the acquisition of the CTC image. The patient is alive and well 14 years later.

Figure 3:

Unsuspected osteoporosis in an asymptomatic 74-year-old woman with subsequent hip fracture and death. Top row: Sagittal images from CT colonography (CTC) screening in 2005 (left) and 2010 (middle) show progressive bone loss, with L1 trabecular attenuation values of 85 HU and 63 HU, respectively. Sagittal image from subsequent CT in 2012 (right) shows further bone loss at L1 (48 HU), as well as multiple new vertebral compression fractures, most notably at L5 (arrow). Bottom row: Both dual-energy x-ray absorptiometry (not shown) and quantitative CT (left) T-scores were falsely negative for osteoporosis. Composite body composition image (middle, similar to Fig 1) from 2012 shows advanced osteosarcopenia. The patient experienced an intertrochanteric femoral fracture as shown on anteroposterior left hip radiograph (right, arrow) later that year and died in 2016.

Figure 4:

Unsuspected osteoporosis and sarcopenia (osteosarcopenic obesity) in a 58-year-old woman with subsequent hip fracture. Top row: Sagittal (left) and L3-level transverse (middle) images from CT examination for unexplained abdominal pain show a prevalent L1 compression fracture and associated low bone mineral density (BMD) (L2 region of interest [ROI]), compatible with complicated osteoporosis. The same transverse L3-level image (right) shows superimposition of automated BMD (green) and muscle (red) segmentations, with mean muscle attenuation of 2 HU, comparing well with the manual paraspinal ROI (4 HU). Bottom row: Retrospective quantitative CT image (left) shows an osteoporotic femoral neck T-score of –2.9, but central dual-energy x-ray absorptiometry examination 2 years later (not shown) was falsely negative for osteoporosis and also missed the L1 compression. The patient presented 3 months later with hip pain. Initial pelvic radiograph (middle) was negative, but an MRI examination later that day (right) revealed a trochanteric fracture with extension into the metaphysis (arrows), which required internal fixation. Note also generalized sarcopenia at MRI.

Figure 5:

Automated tool for detection of vertebral compression fractures. Sagittal image from chest CT shows automated spine segmentation tracking vertebral height (colored lines) obtained with a dedicated artificial intelligence tool (AI Genant, IRA analysis). Compression fractures with 25%–40% height loss are depicted in yellow and more severe compressions (>40%) in red. (Image courtesy of Alexey Petraikin, MD, PhD, Moscow Laboratory of Innovation Technologies. See https://mosmed.ai/en/ for general information about the Moscow project “Artificial intelligence in radiology.”)

Figure 6:

Myosteatosis with subsequent hip fracture in an 87-year-old woman. Top row: Composite L1-level CT image (left) and L3-level CT image with (middle) and without (right) automated muscle overlay show sarcopenia, with intermuscular adipose tissue most notably involving the paraspinal musculature. Automated mean muscle attenuation measured –10 HU, similar to the manual measurement (–12 HU). See Figure 1 for the color key. Bottom row: The patient experienced a femoral neck fracture 3 years later as shown on anteroposterior radiograph of pelvis (left, arrow) and died 3 years after that. L3-level CT images (middle and right) from another patient using a different automated algorithm show separate segmentation of skeletal muscle (pink) and intermuscular adipose tissue (green). Results for myosteatosis (in Hounsfield units) and myopenia (area) are dependent on how a specific muscle algorithm handles intermuscular adipose tissue. (Images from the second patient courtesy of Akshay Chaudhari, PhD, and Robert D. Boutin, MD, Stanford University.)

Figure 7:

Differences in visceral and subcutaneous fat distribution (apple vs pear body habitus). Midabdominal CT image with automated body composition overlays in an 81-year-old man (left) shows an abundance of visceral fat relative to subcutaneous fat, corresponding to a high visceral to subcutaneous fat ratio, as well as the so-called apple-shaped body habitus. In contrast, CT image from a 34-year-old woman (right) shows a disproportionate amount of subcutaneous fat (pear-shaped). The patient on the right had a higher body mass index, or BMI (35 kg/m²), than the patient on the left (31 kg/m²), who later experienced a myocardial infarction and subsequently died of metabolic syndrome–related issues. See Figure 1 for the color key.

Figure 8:

Automated quantification of aortic atherosclerotic calcification. Top row: Transverse (left) and coronal (right) noncontrast CT images in an 89-year-old woman show extensive aortoiliac calcification, which has been automatically segmented (bright yellow). The patient experienced a myocardial infarction 3 years later and died within a year after that. Note also the abundant visceral fat. Bottom row: Transverse (left) and coronal (right) postcontrast CT images in an 80-year-old woman also show extensive calcified aortic plaque. This patient also experienced a myocardial infarction and subsequently died. Note how the automated algorithm correctly segmented the aortic calcification despite luminal contrast enhancement. For both patients, the aortic Agatston score was markedly elevated. See Figure 1 for the color key.

Figure 9:

Automated artificial intelligence (AI) algorithm for opportunistic screening of cardiomegaly. Panel of images from contrast-enhanced CT in a 58-year-old man undergoing surveillance for previously resected right renal cell carcinoma. The AI algorithm performed two-dimensional segmentation of the whole heart (lower left, green), inner chest (upper middle, blue), outer chest (lower middle, red), and left ventricle (lower right, yellow). Original full-view (upper left) and ×2 magnified (upper right) images are included for comparison. Area of the inner and outer chest, as well as patient age and sex, are used to standardize the whole-heart and left ventricular measurements to best identify cardiomegaly. (Images courtesy of Andrew D. Smith, MD, University of Alabama at Birmingham.)

Figure 10:

Automated tools for assessing liver fibrosis, cirrhosis, and portal hypertension. Top row: Images from fully automated Couinaud segmentation of the liver in a 45-year-old man (left) and 60-year-old man (right), both with hepatitis C virus, who had fibrosis scores of F0 and F4, respectively, at liver biopsy. Note the relative compensation of the left lateral segments (II and III) in the cirrhotic patient, compatible with segmental redistribution. This is also reflected in more quantitative terms by the elevated liver segmental volume ratio (LSVR), which objectively compares Couinaud segments I–III over IV–VIII. Bottom row: Upper abdominal transverse CT images in two patients with compensated cirrhosis (49-year-old woman with alcoholic cirrhosis on left, 50-year-old man with hepatitis C virus cirrhosis on right) again show composite depictions of the automated body composition tools, which demonstrate splenomegaly. Automated splenic volumes (1092 mL and 985 mL, respectively) matched semiautomated derivation within 10% in both patients. As with liver segmental volume ratio, splenic volume correlates with pathologic stage of liver fibrosis, and the two measures are actually complementary. For the bottom row, see Figure 1 for the color key. (Top row images courtesy of Sungwon Lee, MD, PhD.)

Figure 11:

Sarcopenic obesity in two patients with colorectal cancer. CT images in a 79-year-old man (left) and 69-year-old woman (right) with automated body composition tools applied show abundance of visceral fat and myosteatosis (L3-level muscle attenuation was <10 HU for both patients). However, the amount of subcutaneous fat differs substantially between the two patients. Conflicting data exist on whether abdominal fat is protective or detrimental, and may depend on the specific compartment (visceral vs subcutaneous). See Figure 1 for the color key.