This study utilized a scan data set of 18 people with dual energy X-ray absorptiometry, CT, and QCT data at Inha University Hospital. This study adhered to the principles of the Declaration of Helsinki and was approved by the Human Research Ethics Committee (IRB) of Inha University Hospital. All research procedures were in strict accordance with ethical standards, including protecting the privacy, confidentiality and rights of participants. In this study, Clari-QCT software was used to evaluate whether a report equivalent to QCT could be output using only CT data. Outcomes were assessed using several indices, including intraclass correlation coefficient (ICC).

Clari-QCT is a program that takes abdomen CT images as input, automatically distinguishes between L1 and L2, and then sets the volume of interest in three slices (the middle slice and one slice each above and below it) to calculate bone density. Upon completion of the calculation, it outputs the bone mineral density (BMD), T-score, and Z-score of L1 and L2 mean BMD in the analyzed slices in a report format. This allows for a comparative analysis with the actual QCT image values (Fig. 1).

To objectively evaluate the efficacy of Clari-QCT, patients were classified into normal, mild, and severe groups based on the presence and degree of spinal deformity. The normal group consisted of patients with anatomically normal spinal alignment and no deformities. The mild deformity group included patients with minimal spinal curvature changes that did not significantly impact daily life, typically requiring non-surgical treatments or regular monitoring. The severe deformity group comprised patients with significant spinal curvature or deformities leading to functional impairments, where surgical intervention might be necessary.[5]

We employ Python to calculate the ICC, which serves as a measure of reliability or agreement between two methods of measurement. The dataset consists of BMD values, both actual and predicted, across eighteen subjects. The ICC calculation assesses the actual versus predicted BMD values. The output, rounded to three decimal places for clarity, provides crucial statistics including the ICC value, F-statistic, degrees of freedom, P-value, and the 95% confidence interval, a conducted evaluation. The criteria for evaluating ICC were Regier 2012, depending on the characteristics and purpose of the study.

To analyze how other variables affect report generation, variables such as age, height, and weight and predicted L1 BMD values were visualized and calculated using Pearson correlation.

The study participants ranged in age from 43 to 91, with an average age of 64. The mean BMI was 24.88. The average BMD measured by QCT was 94.7 g/cm³, while the average BMD measured by Clari-QCT was 122.5 g/cm³. The difference in average BMD between QCT and Clari-QCT for individual patients ranged from a minimum of 4.9 to a maximum of 61.8 (Table 1).

ICC analysis in all population was performed for average BMD values were calculated. ICC1 (single rater absolute), 0.597; ICC2 (single random rater), 0.64; ICC3 (single fixed rater), 0.81; ICC1k (mean rater absolute), 0.748; ICC2k (mean random rater), 0.78; ICC3k (mean fixed evaluator), 0.895 (Table 2).

As a result of analyzing predicted bone density (L1, L2 mean BMD), age, height, and weight using the Pearson correlation coefficient, a very weak negative correlation (r=−0.03) was observed between age and predicted bone density, indicating that the predicted BMD slightly decreases as age increases, although this correlation is not statistically significant. Comparing height and predicted BMD showed a weak negative correlation (r=−0.11). Comparison of body weight and predicted BMD also showed a weak negative correlation (r=−0.18) (Fig. 2).

The quality of Clari-QCT’s report generation using CT images showed excellent similarity overall. ICC analysis performed consistency of BMD measurements. Average ICC values indicate moderate to significant agreement between raters across different scenarios. Specifically, ICC1, ICC2, and ICC3 values ranged from 0.597 to 0.81, indicating a significant level of agreement between individual evaluators. Additionally, the range of ICC1k, ICC2k, and ICC3k values, which represents the average agreement between evaluators, is 0.748 to 0.895. These results can be interpreted to mean that Clari-QCT’s BMD measurement has a similar degree of agreement with the actual QCT measurement when multiple evaluators are considered simultaneously.[6]

We conducted the following analysis using Pearson correlation analysis performed on predicted BMD and various demographic factors (age, height, weight). First, a weak negative correlation (r=−0.03) was observed when comparing age and predicted BMD. This suggests a slight decrease in estimated BMD with increasing age. This correlation was not statistically significant, indicating that the observed association may have occurred by chance. Comparing height and predicted BMD also showed a negative correlation (r=−0.11), This indicates that height and expected BMD values are not related. Importantly, this correlation was not statistically significant, suggesting a reliable association between height and predicted BMD. Lastly, between body weight and estimated bone density relationship, a weak negative correlation (r=−0.18) was observed. This means that higher body weight is not associated with a slight increase in predicted BMD.

For CT scans, there is flexibility to image patients with abnormal spinal structures, but they do not provide as accurate bone information compared to QCT.[7] Therefore, QCT is required for assessing osteoporosis and accurately evaluating bone health in patients with spinal deformities. [2] However, direct QCT imaging is challenging for such patients due to the following reasons: a specific posture is required for precise measurements.[8] In QCT, the exact position and angle of the bones are crucial, so the patient must remain in a fixed position for an extended period during the scan.[9]

The spinal deformity levels were classified into normal (N=9), mild (N=4), and severe (N=5) groups, and the differences between actual QCT BMD and Clari-QCT BMD were calculated for each group. The mean BMD errors for the normal, mild, and severe groups were 25.71, 55.93, and 25.3, respectively. The analysis revealed that the mild deformity group showed the largest error, while the normal and severe deformity groups had similar levels of error. The relatively lower errors in the normal and severe groups suggest that the Clari-QCT software is able to estimate BMD with reasonable accuracy in patients with normal spinal structures, and may also recognize and adjust for significant spinal deformities in the severe group. In contrast, the larger error observed in the mild group indicates that Clari- QCT may struggle to handle subtle structural deformities, likely due to its inability to fully account for the effects of minor changes in bone position or angle.

We conducted a comparison between Mindways QCT Pro [10] and Clari-QCT, two software tools designed to assess BMD using different application approaches. QCT Pro performs high-resolution three-dimensional analysis that includes detailed segmentation of bone structure, allowing for highly reliable BMD measurements. However, this level of precision requires specialized imaging, which limits its immediate accessibility. In contrast, Clari-QCT’s AI-based methodology takes standard CT images as input and uses machine learning to calculate BMD in an automated manner. Targeting the L1 and L2 vertebrae, Clari-QCT generates T-score and Z-score similar to those derived from QCT, which are essential for osteoporosis assessment. Due to this difference in approach, Clari-QCT offers greater versatility in settings where traditional QCT may not be available, particularly for patients with conditions like spinal deformities, where additional scans may be challenging, making Clari- QCT a more broadly applicable option. Also, Clari-QCT and Opportunistic CT are similar in that they both utilize conventional CT scans; however, their application methods differ. Clari-QCT is primarily developed as a specialized tool for evaluating BMD by converting CT reports to resemble QCT, processing data in a manner similar to traditional QCT, and thus differs somewhat from opportunistic screening. In contrast, Opportunistic CT takes advantage of existing CT images obtained for health checkups or other purposes to opportunistically assess osteoporosis risk. This approach does not provide precise bone density measurements.[11] Therefore, Clari-QCT itself serves as a specialized and precise evaluation tool particularly suited for patients with spinal deformities, whereas Opportunistic CT reuses existing data without focusing on high-precision evaluation of specific regions.

This study has several limitations. The study’s small sample size may have been inadequate to detect meaningful correlations. Additionally, the study population may have lacked diversity in terms of demographic characteristics such as age range, ethnicity, and socioeconomic status. Using a more heterogeneous sample allows results to be more effectively generalized to a broader population. To establish Clari-QCT as a reliable tool, it is necessary to acknowledge that discrepancies were observed between absolute BMD values obtained from CT and Clari-QCT. Further research should be conducted to define osteoporosis thresholds based on Clari-QCT measurements.