As the clinical use and application of an opportunistic CT HU measurement continues to grow, standardization of measurement is critical. This study recommends a 200-mm2 circular ROI placed at the vertebral body centroid and be used for a clinical opportunistic CT HU measurement. With this approach, L1 values are approximately 15 HU lower than L4, and sagittal measurements are ~ 10 HU lower than axial. Importantly, non-radiologist physicians can reproducibly measure HU with this approach.

Many prior studies have correlated opportunistic CT with DXA for osteoporosis screening, and interest in the clinical use of osteoporosis screening using opportunistic CT imaging is increasing. Despite increased clinical use, guidance of the HU measurement technique is lacking [6, 7, 14]. Various methods for HU measurements have been reported in prior studies, including a single oval ROI over the vertebral trabecular bone on axial CT between T12 and L5 [6], the L1 trabecular space measurement of the single axial CT image between the mid-vertebral level and superior endplate [7], and a single ovoid of a 100–300-mm2 ROI placed at the L1 trabecular bone on axial imaging [14]. As the clinical use of opportunistic CT increases, this lack of guidance regarding optimal ROI size and placement could reduce the reliability of opportunistic CT to accurately identify those at most risk. Furthermore, opportunistic CT has been used to assess treatment changes over time [15, 16]; however, the precision of repeat CT for this purpose has previously been unknown. This lack of guidance may also discourage non-radiologists from utilizing this technique in clinical practice.

Understanding vertebral body microarchitecture is important when using the opportunistic CT. The cortical shell and endplates have higher bone density but were excluded in this, and the majority of studies were using the opportunistic CT. The trabecular bone structure is anistopic, meaning it has a varying structure according to orientation and location. Vertical trabeculae are thicker with longer struts, whereas horizontals are shorter and thinner and are purposedly to provide cross-linking. There is a variation in structure from left to right which will change if deformity is present. Thus, age-rated changes and bone loss, i.e., osteoporotic change, will alter the trabecular structure, and thus will affect HU. In osteoporosis, the horizontal trabeculae are lost first, and thus in cross-section the vertebral body appears dotted and non-homogenous compared to younger patients. This will account for changes in HU when ROI are oriented vertically or axially. We did not use the coronal views; although, this is certainly possible.

To our knowledge, this is the first study aimed at identifying the optimal ROI size and placement for HU measurement of bone using the opportunistic CT. Our study incorporated various ROI sizes (100, 200, 300 mm2, and maximum) and positions on both axial and sagittal views to determine optimal ROI size and location for HU measurement. HU measurements in this study did not differ importantly by ROI size; however, 200 mm2 is recommended as larger sizes may include the vertebral cortex in small patients which can alter HU. Axial HU measurements were found to be higher than sagittal measurements by approximately 6–10 HU. This is consistent with prior data indicating areas of higher density on sagittal measurements [17]. Anterior, middle, and posterior locations of ROI placement did not affect HU measurements in our study, but an image plane (left/right or cranial/caudal) did alter HU by up to 62 HU, likely reflecting anisotropic vertebral microarchitecture [10]. Vertebral rotation due to positioning or spinal deformity makes location of the ROI more important, but the PACs software (multiplanar reconstruction) can allow correction but requires users to have more experience in using this technology. HU measurement differences of this magnitude (i.e., up to 62 HU) can greatly affect the clinical usefulness of this technique as mean HU decreases with age and are estimated at a rate of only 2.5 HU per year, and differences in clinically meaningful HU thresholds for osteoporosis risk may only differ by up to 50 HU [7].

The protocol to assess precision followed ISCD recommendations for DXA; although, we did not use volunteers who had multiple imaging studies performed for the purpose of precision assessment, but rather patients who had two scans based on clinical needs. The CVs were two- to threefold larger than that of DXA [18] despite CT calibration and technique being closely monitored at our institution. This variation appeared to be in the inability to exactly match the ROI location on two different scans where even a difference of 2.5 mm (one slice) could change HU by as much as 10. Although we attempted to use the centroid of the vertebral to locate our ROI, no references are available to determine that this point is at the same location on the two CTs except visually. Another source of variability is differences in machines/models that have been shown by Engelke to affect HU by similar amounts to what we observed [19]. However, that report used phantoms rather than patients as in our study. The CT scanners at our facilities, although of different models, are from a single manufacturer and are assured daily for calibration and quality. The variation in HU between L1 and L4 may be partially explained by a study of a spine clinic population where L4 degenerative changes are more prevalent and likely resulting in higher HU.

Normal bone structure is anisotropic, and further variation may occur from an osteoporosis-related bone loss or other anomalies. In regards to this within vertebral variability, a result from our study that may warrant further investigation is the small, but greater than previously reported, higher mean L1 HU (~ 6–12 HU) observed on the axial compared with the sagittal image [9]. This higher HU value likely reflects inclusion of a slightly denser stripe of the trabecular bone parallel to the endplates at the mid-vertebral level noted on sagittal CT images [9, 20]. An approach would be to avoid this location, resulting in lower HU; however, this would come at the expense of a decreased clinical reproducibility, but at least it is when these measurements are performed manually. Moreover, a precedent exists for avoiding the lowest BMD region as measured by DXA, i.e., Ward’s region, in part due to concerns about reproducibility [21]. Finally, the small HU difference in this study compared with prior data is likely of no clinical significance, as opportunistic CT does not serve as a definitive diagnostic assessment.

Our study demonstrates that non-radiologists can reliably identify the vertebral body centroid on axial and sagittal CT views, which measure HU on axial and sagittal CT imaging. As use of opportunistic CT increases clinically, it may be beneficial for non-radiologists to be comfortable measuring HU on CT scans obtained for other purposes. This would allow non-radiologists to estimate bone loss and fracture risk for patients in their practice. This could be beneficial in a variety of clinical contexts including orthopedics, oncology, and other clinical scenarios where patients are at high risk for bone loss or fracture and undergo routine CT imaging for other purposes [22]. Not at the level of an individual clinician, but rather at the healthcare system level, artificial intelligence may lead to a routine HU measurement of bone and muscle to identify not only fracture but also falls risk, thereby facilitating widespread adoption of the opportunistic CT [23,24,25,26]. It seems reasonable that even such approaches will be dependent on the agreement of what ROI(s) should be measured [27].

There are limitations to this study. First, our study was conducted at one institution in two small cohorts of patients undergoing CT imaging for other indications. Specific health data and bone loss risk were not evaluated which limits generalizability of this cohort to other populations. All CTs were performed using a single manufacturer; although, models varied. While HU will vary by small amounts based on the manufacturer [28], HU measurement reliability is independent of the manufacturer [29]. It must be emphasized that all studies in this investigation were non-contrast and used only a 120-kV tube voltage; HU measurements will be affected by contrast and tube voltage differences. A linear inverse relationship exists between tube voltage and HU [30]; although, the reliability of HU measurement is independent of this variable. Similarly, the contrast will increase HU [3]; although, we do not believe that reliability will be affected. Other factors that have small effects on HU such as the patient table height in the scanner and prone vs. supine position were not evaluated. The slice thickness and number of slices obtained for each vertebral body could affect precision as thinner sections would be expected to have less variation in HU than when fewer sections are available.

In conclusion, this study provides guidance on optimal ROI size and placement for HU measurement of bone and demonstrates that non-radiologists can reliably identify vertebral centroid and accurately measure HU on axial and sagittal images. Based on results from this study, we recommend a 200-mm2 circular ROI placed at the vertebral centroid on L1 axial imaging for clinical use. As clinical use of opportunistic CT further progresses, this guidance may help optimize clinical usefulness.