Machine learning classification of mediastinal lymph node metastasis in NSCLC: a multicentre study in a Western European patient population

EJNMMI Physics

Table 3 Overview of the top 5, top 10, and top 20 best-performing features and the permutation feature importance

	Feature	Data input		Feature weight
	Feature	PET lymph node	PET primary	Feature weight
1	90 Percentile	X		0.63
2	Sphericity		X	0.16
3	Elongation		X	0.08
4	Variance	X		0.07
5	Least axis length		X	0.06
6	Grey-level non-uniformity		X	0.06
7	Short-run high grey-level emphasis		X	0.03
8	Energy		X	0.02
9	Short-run emphasis	X		0.02
10	Dependence non-uniformity		X	0.02
11	Maximum 2D diameter row	X		0.02
12	Interquartile range	X		0.02
13	Grey-level non-uniformity normalized		X	0.01
14	Large dependence high grey-level emphasis		X	0.01
15	Short-run emphasis		X	0.01
16	Robust mean absolute deviation	X		0.01
17	Elongation	X		0.01
18	Minimum		X	0.01
19	Mean absolute deviation	X		0.01
20	Maximum		X	0.01