Diagnosis of Thyroid Nodules Based on Image Enhancement and Deep Neural Networks

Table 1

Performance comparison of DNN with different network structures.

Number of network layers	Network structure	Accuracy	Precision	Recall	F1-score
Three layers	2500-50-1	0.843137	0.863158	0.964706	0.911111
	2500-60-1	0.882353	0.892473	0.976471	0.932584
	2500-80-1	0.892157	0.893617	0.988235	0.938547
	2500-100-1	0.901961	0.912088	0.976471	0.943182
	2500-200-1	0.892157	0.893617	0.988235	0.938547
	2500-300-1	0.882353	0.901099	0.964706	0.931818
Four layers	2500-10-2-1	0.823529	0.874045	0.917647	0.896552
	2500-20-2-1	0.833334	0.869565	0.941176	0.903955
	2500-30-2-1	0.862745	0.881720	0.964706	0.921348
	2500-40-2-1	0.901961	0.894737	1	0.944444
	2500-60-2-1	0.823529	0.876404	0.917647	0.896552
	2500-40-10-1	0.872549	0.8912304	0.964706	0.926554
	2500-40-20-1	0.852941	0.880435	0.952941	0.915254
Five layers	2500-40-5-2-1	0.843137	0.8555670	0.976471	0.912088
	2500-100-5-2-1	0.882353	0.901099	0.964706	0.931818
	2500-200-10-2-1	0.872549	0.882979	0.976471	0.927374
	2500-200-20-2-1	0.803921	0.857143	0.917647	0.886364
	2500-200-50-2-1	0.843137	0.896552	0.917647	0.906976
	2500-500-50-2-1	0.852941	0.880435	0.952941	0.915254