Table 5.

Leave-one-patient-out cross-validation F1-score, recall, and precision for Sheffield and Piracicaba datasets.

			F1-score	Recall (Sensitivity)	Precision (PPV)
Inception ResNet-v2	Sheffield Dataset	Patches	77.9%	74.1%	90.2%
		Patient	87.2%	99.3%	81.1%
		RoI	87.0%	77.8%	97.6%
	Piracicaba Dataset	Patches	69.8%	66.9%	85.8%
		Patient	94.2%	97.9%	92.5%
		RoI	89.4%	82.1%	98.2%
ResNet-101	Sheffield Dataset	Patches	71.1%	72.0%	79.0%
		Patient	84.0%	98.0%	78.2%
		RoI	80.2%	73.0%	87.0%
	Piracicaba Dataset	Patches	85.0%	67.3%	89.1%
		Patient	94.5%	98.5%	94.9%
		RoI	96.3%	84.8%	98.4%
Inception-v3	Sheffield Dataset	Patches	69.9%	71.3%	78.4%
		Patient	85.1%	95.8%	80.8%
		RoI	78.3%	74.2%	82.7%
	Piracicaba Dataset	Patches	71.5%	83.0%	93.0%
		Patient	94.3%	91.0%	96.0%
		RoI	91.0%	92.6%	97.9%
VGG-16	Sheffield Dataset	Patches	71.2%	72.4%	80.8%
		Patient	79.9%	90.4%	79.3%
		RoI	76.2%	70.7%	81.0%
	Piracicaba Dataset	Patches	83.1%	78.3%	95.7%
		Patient	97.9%	100.0%	95.4%
		RoI	94.6%	90.2%	100.0%