Radiation reduction for interventional radiology imaging: a video frame interpolation solution

Insights into Imaging

Table 2 Comparison of PSNR (dB), SSIM [18], and RMSE results by the baseline Average, FLAVR, and four variants of SepConv + + trained with DSA sequences in different modalities. “SepConv + + All,” “SepConv + + A,” “SepConv + + S,” and “SepConv + + F” mean that the SepConv + + is trained with the DSA sequences of all three modalities, angiography modality, subtraction modality, and fluoroscopy modality, respectively.“↑”(or “↓”) means that higher (or lower) is better. The best and second best results are highlighted in bold and italics, respectively

Method		Average	SepConv + + All	Gain	p-value	FLAVR	SepConv + + A	SepConv + + S	SepConv + + F
Angiography	PSNR↑	43.72	44.94	+ 1.22	< 0.01	*44.42*	44.89	43.62	44.48
	SSIM↑	0.9786	0.9840	+ 0.005		*0.9830*	0.9838	0.9818	0.9825
	RMSE↓	1.7636	1.5348	− 0.228		*1.6299*	1.5484	1.7481	1.6130
Subtraction	PSNR↑	33.86	34.84	+ 1.01	< 0.01	*34.34*	33.00	34.86	33.65
	SSIM↑	0.9028	0.9194	+ 0.016		*0.9102*	0.8837	0.9196	0.8942
	RMSE↓	4.0441	3.7446	− 0.300		*3.8873*	4.2735	3.7385	4.1096
Fluoroscopy	PSNR↑	32.81	33.82	+ 1.01	< 0.01	*33.72*	33.62	33.16	33.85
	SSIM↑	0.7491	0.7752	+ 0.026		*0.7736*	0.7691	0.7640	0.7758
	RMSE↓	5.3012	4.8724	− 0.428		*4.9039*	4.9630	5.1257	4.8624