Accurate and Interpretable Classification of Microspectroscopy Pixels Using Artificial Neural Networks

. Author manuscript; available in PMC: 2018 Apr 1.

Published in final edited form as: Med Image Anal. 2017 Jan 6;37:37–45. doi: 10.1016/j.media.2017.01.001

	Classifier accuracy (Dice)	Extracted rule NUCLEUS		Rule accuracy (Dice)	Rule similarity
Reference rule model		C1	−1 ·x(2855) + 0.5 ·x(2878.9) + 0.5 ·x(2884.8) > 2
		C2	+1·x(2938.8)−1.s2969 > 18
		C3	(−0.33).s(954.6) − (0.33).s(959.1) −0.33.s(963) + 0.5.s(977.3) + 0.5.s(981) > −3
One ANN	0.91	C1	+ (−0.82) ·x(2849.07) + (−2.02) ·x(2855.02) + (−0.79) ·x(2860.98) + (1.36) ·x(2878.87) + (1.30) ·x(2884.85)>6.21	0.48	0.88
		C2	+ (0.19) ·x(2932.82)+ (0.55) ·x(2938.83)+ (0.18) ·x(2944.86)+ (−0.31) ·x(2962.95)+ (−0.19) ·x(2968.99)> 11.09
		C3	+ (−0.80) ·x(954.59) + (−0.93) ·x(959.12) + (−0.78) ·x(963.65) + (0.45) ·x(972.73) + (1.27) ·x(977.27) + (1.27) ·x(981.82)> 2.02
Double ANN	0.95	C1	+ (−2.66) ·x(2855.02) + (1.56) ·x(2878.87) + (1.32) ·x(2884.85)>9.34	0.66	0.98
		C2	+ (1.14) ·x(2938.83) + (−0.61) ·x(2962.95) + (−0.54) ·x(2968.99)> 23.52
		C3	+ (−0.67) ·x(954.59) + (−0.76) ·x(959.12) + (−0.74) ·x(963.65) + (1.03) ·x(977.27) + (1.14) ·x(981.82)>−3.98