Deep learning parametric response mapping from inspiratory chest CT scans: a new approach for small airway disease screening

Table 2 Evaluation metrics of predicted images obtained through different methods (n = 76)

Methods	Under different methods				Predicted expiratory CT		Predicted PRM
Methods	GAN	Conditional generator	PRM generator	Learnable threshold	SSIM	RMSE (HU)	Dice (PRM^Normal)	Dice (PRM^fSAD)	Dice (PRM^Emph)
SegmentNet with inspiratory			√		…	…	0.84	0.27	0.23
Add GAN	√		√		0.80	88.10	0.85	0.42	0.45
Add Conditional generator	√	√	√		0.86	80.13	0.86	0.48	0.54
Threshod PRM results	√	√			0.86	80.13	0.86	0.45	0.50
Add threshod segmentNet	√	√	√	√	0.86	80.13	0.85	0.51	0.63

GAN: generative adversarial network; PRM: parametric response mapping; fSAD: functional small airway disease; PRM^Normal: the volume percentage of normal area in PRM; PRM^fSAD: the volume percentage of fSAD in PRM; PRM^Emph: the volume percentage of emphysema in PRM; SSIM: structural similarity index; RMSE: root mean squared error

ISSN: 1465-993X