ParaPET: non-invasive deep learning method for direct parametric brain PET reconstruction using histoimages

EJNMMI Research

Table 2 Comparison of kinetic parameters in grey and white matter estimated using DL and the NLS, NLS-C and Patlak methods

Kinetic parameter	Deep learning		NLS		NLS-C		Patlak
Kinetic parameter	GM	WM	GM	WM	GM	WM	GM	WM
K1	0.043–0.087	0.014–0.020	0.168–0.250	0.083–0.136	0.174–0.280	0.0913–0.151	–	–
k2	0.049–0.095	0.012–0.019	0.086–0.171	0.048–0.169	0.095–0.201	0.052–0.176	–	–
k3	0.043–0.059	0.016–0.020	0.026–0.040	0.016–0.023	0.047–0.090	0.024–0.031	–	–
K_i	0.021–0.033	0.008–0.010	0.030–0.110	0.034–0.067	0.055–0.130	0.041–0.077	0.021–0.046	0.008–0.016
p values	–	–	0.16	0.06	0.17	0.09	–	–

Values provided are based on 95% confidence across all subjects. The p values were calculated to establish significant mean differences between K₁, k₂, k₃ between the DL and NLS results (with and without constraints)