Table 2.

Feature sets

2A. Functions and Conventions in Threshold Relaxation Pseudocode
GetBoundingBox(PEi): X → XB	Restricts the volume X to only the volume in the bounding box, XB about PEi.
BW\P	Binary mask of all 1s except at locations, P
Dr(BW\P)	Euclidean distance from P to all points in BW
μ(E|B)	Mean of E where B is true
σ(E|B)	Standard deviation of E where B is true
DP(B, P)	Pathlength distance from point P to all points in mask B, where all paths are restricted to lie within B
S = Dijkstra(B, PEi, PCi)	The Dijkstra shortest path from PEi to PCi constrained to mask, B

2B. The features, ΦBS, summarizing each bridging strand
Strand intensity statistics	Confidence	C
Strand geometry	Vertex count	ν
	Point to point length	1
	Curve length	χ
	Tortuosity	τ = χ/1
Strand neighborhood	Profile distance slope	P
Incident strand geometry	Originating strand length	λ

2C. The features, ΦDCS, summarizing each deletion candidate strand
Strand intensity statistics	Mean z	zμ = (1/n) Σk z(k)
	Minimum z (i.e. Confidence)	zmin min(z)
	Middle z drop	zd = (z(1) + z(n))/2 − zm
Strand topology	Endpoint	γ = (deg(S(1)) = 1) ∨ (deg(S (n)) = 1)
	Number of incident strands	nc = deg(S(1)) + deg(S(n))
Strand geometry	Vertex count	ν
	Point to point length	1
	Curve length	χ
	Tortuosity	τ = χ/1
Incident strand intensity Statistics	Median neighbor strand intensity	ρmedian = median(ρ)
	Minimum neighbor strand intensity	ρmin = min(ρ)
	Variance of neighbor strand intensity	ρσ = (1/nc) Σk (ρ(k) − ρμ)2
Incident strand geometry Statistics	Maximum neighbor strand length	Λmax = max(Λ)
	Mean neighbor strand Length	Λμ = (1/nc)Σk Λ (k)
	Minimum neighbor strand length	Λmin = min(Λ)
	Variance of neighbor strand length	Λσ = (1/nc) Σk (Λ(k) − Λμ)2