Table 7.

Multi-class classification for accuracy measures, using the RSNA dataset (standard-sized test dataset). LB stands for the label balancing usage (PBC in the case of SSDL). PBC results with no statistical significance gains over the non-balanced SSDL implementation are written in italic.

COVID-19/Pneumonia/Normal	SSDL	LB	$n_l=30$		$n_l=50$		$n_l=70$		$n_l=90$
			$\overline{x}$	$s$	$\overline{x}$	$s$	$\overline{x}$	$s$	$\overline{x}$	$s$
10%/45%/45%	No	No	0.561	0.052	0.575	0.062	0.588	0.056	0.603	0.039
	Yes	No	0.587	0.048	0.565	0.045	0.595	0.035	0.571	0.056
		Yes	0.736	0.054	0.748	0.043	0.757	0.045	0.744	0.027
20%/40%/40%	No	No	0.607	0.054	0.636	0.042	0.667	0.036	0.711	0.036
	Yes	No	0.67	0.063	0.691	0.043	0.695	0.048	0.7	0.044
		Yes	0.733	0.052	0.747	0.048	0.758	0.052	0.752	0.024
30%/35%/35%	No	No	0.656	0.048	0.666	0.057	0.698	0.041	0.707	0.039
	Yes	No	0.727	0.654	0.766	0.047	0.76	0.032	0.738	0.04
		Yes	0.732	0.043	0.752	0.04	0.778	0.072	0.727	0.038