Tumor Segmentation in Contrast-Enhanced Magnetic Resonance Imaging for Nasopharyngeal Carcinoma: Deep Learning with Convolutional Neural Network

Table 3

Comparison of segmentation performance among our convolutional neural network model and other models. N.A.: not available.

		Dice similarity coefficient		Corresponding ratio		Percent match
Study	Algorithm	Mean ± SD	Range	Mean ± SD	Range	Mean ± SD	Range
Current study	Convolutional neural network	0.89±0.05	0.80-0.95	0.84±0.06	0.71-0.92	0.90±0.04	0.83-0.96
Zhou at al. [7]	Support vector machine	N.A	N.A	0.72±0.06	0.58~0.85	0.79±0.07	0.65-0.91
Huang et al. [14]	Hidden Markov random field	N.A	N.A	0.72	0.44-0.91	0.85	0.64-1.00
Wang et al. [15]	Deep Convolutional Neural Networks	N.A	-0.80	N.A	N.A	N.A	-0.90