TABLE I.

Parameter Setting for sLTP Learning

Parameters	Setting
ROI size	= 25 mm3, to approximate the size of secondary pulmonary lobules
β1: random shift (for ROI sampling)	∈ [0, 25] mm
β2: sample density (for ROI sampling)	= 3 samples per stack
# of textons: (for texture features)	= 40, targeting 10 textons per standard emphysema subtype and normal tissue class, according to [13]
Texton size	3×3×3 pixels, according to [18]
# of lung sub-regions (for spatial features)	= 36, according to binning of (r, θ, ϕ) in Section II-C3
NL T P: # of LTPs in initial set	= 100, as suggested in [18], for sufficient diversity of the patterns and being able to discover rare emphysema types