Skip to main content

Table 1 Comparison of vocal cord frame detection results using two augmentation strategies and three confidence levels

From: Automated laryngeal mass detection algorithm for home-based self-screening test based on convolutional neural network

Target

Aug

Conf (%)

TP

TN

FN

FP

Rec

Pre

Acc

F1-score

Vocal cord

No-aug

80

231

0

11

0

0.9545

1.0000

0.9545

0.9767

85

237

0

5

1

0.9793

0.9958

0.9753

0.9875

90

229

0

13

0

0.9502

1.0000

0.9502

0.9745

Single

80

236

0

6

1

0.9752

0.9958

0.9712

0.9854

85

230

0

11

2

0.9544

0.9914

0.9465

0.9725

90

238

0

4

1

0.9835

0.9958

0.794

0.9896

Mixed

80

235

0

7

4

0.9711

0.9833

0.9553

0.9771

85

236

0

6

4

0.9752

0.9833

0.9594

0.9793

90

232

0

10

3

0.9587

0.9872

0.9469

0.9727

Mass

No-aug

80

153

13

96

36

0.6145

0.8095

0.5570

0.6986

85

140

14

106

27

0.5691

0.8383

0.5366

0.6780

90

161

14

85

31

0.6545

0.8385

0.6014

0.7352

Single

80

203

13

45

34

0.8185

0.8565

0.7322

0.8371

85

207

14

43

48

0.8280

0.8118

0.7083

0.8198

90

175

13

73

26

0.7056

0.8706

0.6551

0.7795

Mixed

80

200

12

48

34

0.8065

0.8547

0.7211

0.8299

85

210

13

43

46

0.8300

0.8203

0.7147

0.8251

90

193

14

55

39

0.7782

0.8319

0.6877

0.8042

  1. Aug augmentation, Conf confidence level, TP true-positive, FP false-positive, TN true-negative, FN false-negative, Rec recall, Pre precision, Acc accuracy