Table 5.

The training computation cost of different 2D and 3D CNN models on PET image.

Model	Batch	MADs (G)	e-Time(s)
3D-CNN [14]	8	96.02	56.27 ± 1.45*
2D-CNN [17]	16	0.19 × 10−3	2.03 ± 0.03
Jo et al. [15]	16	34.39	36.18 ± 0.05
Hybrid [19]	16	1.86	2.16 ± 0.04
Max.	16	18.15	14.33 ± 0.12
Avg.	16	18.15	14.02 ± 0.63
ARP [24]	16	18.15	17.14 ± 0.02
LWP (current study)	16	18.53	16.69 ± 0.05

Asterisk (*) indicates the largest value of the epoch time (e-Time).