Table 5.

ICC comparison of trabecular parameter regression results.

Trabecular parameter	Radiomic-feature-based model	CNN regression model	Distillation-learning-based CNN regression model
BV/TV	0.548 ± 0.011	0.670 ± 0.053	0.804 ± 0.037
Tb.Th	0.407 ± 0.032	0.488 ± 0.038	0.773 ± 0.042
Tb.Sp	0.356 ± 0.017	0.523 ± 0.054	0.711 ± 0.063
Tb.N	0.326 ± 0.036	0.502 ± 0.085	0.622 ± 0.133
Overall	0.409 ± 0.026	0.545 ± 0.059	0.742 ± 0.046

The compared regression methods include the radiomic-feature-based model, CNN regression model, and the proposed distillation-learning-based CNN regression model. The trabecular parameters that were regressed are BV/TV, Tb.Th, Tb.Sp, and Tb.N. BV/TV, bone volume/total volume; Tb.Th, trabecular thickness; Tb.Sp, trabecular separation; Tb.N, trabecular number.