Automatic lung segmentation in routine imaging is primarily a data diversity problem, not a methodology problem

European Radiology Experimental

Table 2 Test datasets used to evaluate the performance of lung segmentation algorithms

Abbreviation	Description	Number of volumes	Number of slices-L	Total number of slices
RRT	Routine random test	20	5,788	7,969
LTRC	LTRC	105	44,784	51,211
LCTSC	LCTSC	24	2,063	3,675
VESS12	VESSEL12	20	7,251	8,593
Atel	Atelectasis	2	395	534
Emph	Emphysema	2	688	765
Fibr	Severe fibrosis	4	1,192	1,470
Mass*	Mass	2	220	273
PnTh	Pneumothorax	2	814	937
Trau	Trauma/effusion	3	911	2,225
Norm**	Normal (large field of view)	7	1,180	5,301
	Total	191	65,286	82,953

The number of volumes, the number of slices that showed the lung (slices-L), and the total number of slices (#Slices) are listed. LTRC, LCTSC, and VESS12 are cases from the respective public dataset that were not used for training
*Two cases from the publicly available Lung1 dataset
**Four cases from the publicly available Visceral Anatomy 3 dataset