Skip to main content

Table 4 The areas of concern of level of coincidence between DLM and radiologists

From: Deep learning models of ultrasonography significantly improved the differential diagnosis performance for superficial soft-tissue masses: a retrospective multicenter study

Level of coincidence

Complete (100%)

Most (50–99%)

Partial (1–49%)

Not at all (0)

Radiologist-1

 Lipomyoma

36.4% (8/22)

54.5% (12/22)

9.1% (2/22)

0 (0/22)

 Hemangioma

25.0% (5/20)

60.0% (12/20)

10.0% (2/20)

5.0% (1/20)

 Neurinoma

13.8% (4/29)

72.4% (21/29)

20.7% (6/29)

0 (0/29)

 Epidermal cyst

20.7% (6/29)

58.6% (17/29)

10.3% (3/29)

3.5% (1/29)

 Calcifying Epithelioma

55.0% (11/20)

30.0% (6/20)

15.0% (3/20)

0 (0/20)

 Sarcoma

30.0% (3/10)

50.0% (5/10)

10.0% (1/10)

10.0% (1/10)

 Total

28.4% (37/130)

56.2% (73/130)

13.1% (17/130)

2.3% (3/130)

Radiologist-2

 Lipomyoma

27.3% (6/22)

68.2% (15/22)

4.6% (1/22)

0 (0/22)

 Hemangioma

25.0% (5/20)

35.0% (7/20)

25.0% (5/20)

15.0% (3/20)

 Neurinoma

6.9% (2/29)

72.4% (21/29)

27.6% (8/29)

3.5% (1/29)

 Epidermal cyst

13.8% (4/29)

55.2% (16/29)

13.8% (4/29)

6.9% (2/29)

 Calcifying epithelioma

50.0% (10/20)

30.0% (6/20)

15.0% (3/20)

5.0% (1/20)

 Sarcoma

30.0% (3/10)

30.0% (3/10)

20.0% (2/10)

20.0% (2/10)

 Total

23.1% (30/130)

52.3% (68/130)

17.7% (23/130)

6.9% (9/130)