Evaluation of lung involvement in COVID-19 pneumonia based on ultrasound images

Table 1 Method comparison and ablation experiment on different datasets

Method	S (%)				S&M (%)				S&M&P (%)
Method	Acc	PP	Se	Sp	Acc	PP	Se	Sp	Acc	PP	Se	Sp
Method comparison
VGG [20]	93.15	73.07	74.62	96.09	89.87	76.36	73.05	94.06	88.81	73.5	67.37	93.19
ResNet [21]	93.19	73.56	75.55	96.06	90.63	77.77	74.94	94.51	89.41	74.56	68.78	93.56
ResNeXt (baseline) [22]	93.15	77.82	74.62	96.31	91.7	79.98	77.57	94.94	90.46	75.92	70.9	94.04
SENet-50 [23]	92.5	72.16	73.33	95.65	93.64	83.16	83.74	96.28	92.37	79.71	75.93	95.33
SKNet-50 [24]	92.8	86.32	76.97	95.79	92.12	80.45	78.63	95.4	91.03	77.21	72.55	94.55
Ablation experiment
w/o. CRFA and w/o. fusion module	93.15	77.82	74.62	96.31	91.7	79.98	77.57	94.94	90.46	75.92	70.9	94.04
w/o. CRFA module	93.19	78.09	75.55	96.34	92.12	81.21	78.63	95.17	90.93	77.14	71.94	94.3
MCRFNet	97.73	85.72	88.06	98.69	96.25	87.2	83.59	97.74	94.39	82.28	76.27	96.44

S Stork dataset accuracy, M Mindray dataset accuracy, P Philips dataset accuracy, Acc accuracy, PP precision, Se sensitivity, Sp specificity

ISSN: 1475-925X