Table 4.

Mean and standard deviation (SD) image quality scores (1=uninterpretable; 2=poor; 3=adequate; 4=good; 5=excellent) and the proportion of high-quality images (scores 4–5) from the four readers.

Reader		Ground-truth	Method A	Method B	Method C	Method D	Low-count
1	Mean±SD	4.78±0.58	3.35±0.98	4.28±0.75	4.48±0.51	4.45±0.64	3.65±1.03
	High-quality images (%)	92.5	45	82.5	100	92.5	60
2	Mean±SD	3.40±0.55	2.70±0.72	3.65±0.89	3.73±0.72	3.60±0.67	2.65±0.66
	High-quality images (%)	42.5	15	52.5	57.5	60	10
3	Mean±SD	2.93±0.69	2.90±0.38	4.05±0.39	3.58±0.68	3.35±0.77	1.15±0.36
	High-quality images (%)	20	2.5	95	62.5	42.5	0
4	Mean±SD	2.38±0.49	2.65±0.77	3.83±0.78	4.08±0.76	3.90±0.74	1.08±0.26
	High-quality images (%)	0	17.5	65	80	72.5	0

Network training methods: A-direct application of pre-trained network; B-transfer learning starting with pre-trained network; C-training new network from scratch; D-training new network with combined datasets