. Author manuscript; available in PMC: 2020 Sep 11.

Published in final edited form as: Large Scale Annot Biomed Data Export Label Synth Hardw Aware Learn Med Imaging Comput Assist Interv (2019). 2019 Oct 24;11851:32–41. doi: 10.1007/978-3-030-33642-4_4

Table 2.

Accuracy of 5-group and binary classification produced by random forest (RF), support vector machine (SVM) and SmallNet on different training sets.

Task	Training set	Size	RF	SVM	SmallNet
5-group classification	Raw data	405	19.9%	23.3%	19.2%
	Up-sampling	2500	27.5%	28.7%	23.1%
	Augmentation	1.6M	27.9%	29.3%	44.1%
Binary classification	Raw data	405	54.7%	57.7%	53.7%
	Up-sampling	430	55.6%	58.1%	56.1%
	Augmentation	600K	55.5%	58.2%	73.8%