Table 4.
Performance of the CNN model integrating images with clinical parameters inputting for classifying osteoporosis, assessed on the training, validation, and test cohorts.
| Datasets | Image projection | AUC (95% CI) | Sensitivity (%) | Specificity (%) | PPV (%) | NPV (%) |
|---|---|---|---|---|---|---|
| Training | AP | 0.981 (0.976–0.984) |
86.20 (84.42–87.80) |
95.75 (95.00–96.40) |
90.91 (89.35–92.26) |
93.36 (92.47–94.16) |
| LAT | 0.963 (0.957–0.968) |
88.67 (87.02–90.13) |
90.37 (89.31–91.34) |
81.95 (80.07–83.69) |
94.18 (93.30–94.95) |
|
| AP and LAT | 0.996 (0.993–0.998) |
99.94 (99.61–100) |
99.97 (99.81–100) |
99.94 (99.61–100) |
99.97 (99.81–100) |
|
| Validation | AP | 0.922 (0.897–0.941) |
73.21 (66.83–78.79) |
92.57 (89.46–94.85) |
84.54 (78.50–89.17) |
86.18 (82.48–89.21) |
| LAT | 0.926 (0.901–0.944) |
81.70 (75.87–86.41) |
86.88 (83.10–89.94) |
77.54 (71.58–82.59) |
89.54 (85.98–92.31) |
|
| AP and LAT | 0.928 (0.904–0.947) |
75.00 (68.70–80.42) |
92.08 (88.89–94.44) |
84.00 (78.01–88.65) |
86.92 (83.26–89.89) |
|
| Test cohort 1 | AP | 0.928 (0.904–0.946) |
73.93 (67.37–79.61) |
90.17 (86.80–92.77) |
79.19 (72.71–84.50) |
87.24 (83.63–90.17) |
| LAT | 0.930 (0.907–0.949) |
81.52 (75.47–86.38) |
88.73 (85.20–91.52) |
78.54 (72.39–83.66) |
90.46 (87.09–93.05) |
|
| AP and LAT | 0.943 (0.921–0.960) |
75.36 (68.87–80.90) |
91.61 (88.42–94.01) |
81.96 (75.66–86.96) |
88.02 (84.50–90.85) |
|
| Test cohort 2 | AP | 0.912 (0.887–0.933) |
68.57 (61.76–74.69) |
92.34 (89.26–94.63) |
81.82 (75.15–87.06) |
85.40 (81.72–88.45) |
| LAT | 0.905 (0.878–0.926) |
70.00 (63.24–76.01) |
88.76 (85.24–91.54) |
75.77 (69.01–81.50) |
85.48 (81.73–88.59) |
|
| AP and LAT | 0.915 (0.889–0.935) |
69.05 (62.25–75.13) |
92.58 (89.53–94.83) |
82.39 (75.77–87.55) |
85.62 (81.96–88.65) |
AP, anteroposterior; LAT, lateral; AUC, area under the curve; PPV, positive predictive value; NPV, negative predictive value.