Table 3.

A comprehensive comparison of different models used for the detection of Pneumonia from CXR images. These algorithms have been compared on basis of no. of epochs at which model converged, CapsNet accuracy (CN Acc$t_r$, CN A${\mathrm{cc}}_{valid}$, CN Acc$T_e$), CapsNet loss (CN Loss$t_r$, CN Loss$v_a$), Decoder accuracy (D Acc$t_r$, D Acc$v_a$), Decoder loss (D Loss$t_r$, D Loss$v_a$), and total loss (sum of all losses) (Loss$t_r$, Loss$t_r$). Here, CapsNet in the first row is the Simple CapsNet model discussed in this work.

Metrics	CapsNet	ICC	ECC	E3CC	E4CC	E8CC	E16CC
epochs	162	100	148	213	300	182	190
CN Acc$t_r$	90.25%	94.29%	95.70%	71.78%	95.53%	95.36%	94.19%
CN Acc$v_a$	93.75%	95.22%	95.41%	77.69%	96.29%	94.53%	95.30%
Acc$t_e$	93.96%	95.33%	95.90%	81.54%	96.36%	94.19%	95.67%
CN Loss$t_r$	0.089	0.048	0.043	0.186	0.049	0.050	0.057
CN Loss$v_a$	0.086	0.045	0.062	0.1979	0.045	0.063	0.054
D Acc$t_r$	6.63%	33.32%	33.33%	33.39%	33.33%	33.35%	33.32%
D Acc$v_a$	4.91%	33.34%	33.32%	33.25%	33.38%	33.36%	33.29%
D Loss$t_r$	0.017	0.022	0.021	0.023	0.015	0.015	0.017
D Loss$v_a$	0.019	0.023	0.022	0.028	0.016	0.019	0.021
Loss$t_r$	0.106	0.070	0.065	0.2094	0.064	0.065	0.074
Loss$v_a$	0.105	0.068	0.087	0.226	0.061	0.082	0.075