A total of 100 opportunistic CT scans were included in the study. Mean age was 76.6 yo ± 10.1 [SD], 72% were women and 89% were contrast-enhanced CT exams. The ground truth identified 52 CT scans as positive for VCF. 

Regarding VCF screening, the comparison of CINA-VCF with the GT yielded a sensitivity of 92.3% [95% CI: 81.5% - 97.9%], a specificity of 91.7% [95% CI: 80.0% - 97.7%] and an accuracy of 92.0% [95% CI: 84.8% - 96.5%]. In terms of discrepancies, 3/4 of false negatives (FNs) and 3/4 of false positives (FPs) were cases with a VHL very close to the positive threshold between grade 1 and 2 (around 25%). The remaining FN and FP were caused by a compression not visible in the midsagittal plane of the vertebral body and by the natural deformation of L5, respectively. In fact, it is well-known in the literature that the L5 natural shape may induce false positive findings [9]. Figure 1 shows an example of a true positive, a false positive and a false negative. 

Fig 1: Example of a true positive (a) with a VCF at L1 (red arrow); a false...

For the vertebral labeling, 1,700 vertebrae (100 CTs x 17 vertebrae each) were analyzed. The overall per-vertebra agreement between the algorithm and the GT was 93.3% [95% CI: 92.0% - 94.5%]. Discrepancies were caused by L5 sacralization and highly deformed vertebrae. An example of vertebral labeling is shown in Figure 2. 

Fig 2: Example of the algorithm’s vertebral labeling. The software names each...