Automated quality control in image segmentation: application to the UK Biobank cardiovascular magnetic resonance imaging study

Table 3 Analysis of 4800 Random Forest segmentations with available ground truth

Class	Acc.	TPR	FPR	MAE
				DSC	MSD	RMS	HD
					mm	mm	mm
LVC	0.968	0.997	0.330	0.042	0.906	2.514	11.09
	0.975	0.962	0.011
LVM	0.454	0.956	0.571	0.125	0.963	2.141	11.83
	0.972	0.962	0.012
RVC	0.868	0.957	0.352	0.057	1.140	2.790	15.23
	0.969	0.977	0.040
Av.	0.763	0.970	0.418	0.075	1.003	2.482	12.72
	0.972	0.967	0.032
WH	0.954	0.966	0.148	0.035	1.156	2.762	12.52
	0.978	0.984	0.027

4800 RF segmentation at various depths [5 40] and 500 trees. Manual contours were available through Biobank Application 2964. Classes are LV Cavity (LVC), LV Myocardium (LVM), RV Cavity (RVC), an average over the classes (Av.) and a binary segmentation of the whole heart (WH). First row for each class shows the binary classification accuracy for ‘poor’ and ‘good’ segmentations in the Dice Similarity Coefficient (DSC) ranges [0.0 0.7) and [0.7 1.0] respectively. Second row for each class shows the binary classification accuracy for ‘poor’ and ‘good’ segmentations in the Mean Surface Distance (MSD) ranges [>2.0mm] and [0.0mm 2.0mm] respectively. True-positive and false-positive rates are also shown. We report mean absolute errors (MAE) on the predictions of DSC and additional surface-distance metrics: root-mean-squared surface distance (RMS) and Hausdorff distance (HD)

ISSN: 1532-429X