Computer-aided diagnosis for contrast-enhanced ultrasound in the liver

Table 1 Performance of CAD scheme for classification in five categories using physicians’ pattern classification n (%)

Lesion	n	Classification with CAD
		HCC			Metastasis	Hemangioma
		w-HCC	m-HCC	p-HCC
HCC	74
w-HCC	23	15 (65.2)	4 (17.4)	4 (17.4)	0 (0.0)	0 (0.0)
m-HCC	36	16 (44.4)	15 (41.7)	5 (13.9)	0 (0.0)	1 (2.7)
p-HCC	15	1 (6.7)	1 (6.7)	12 (80.0)	1 (6.7)	0 (0.0)
Metastasis	33	1 (3.0)	0 (0.0)	1 (3.0)	28 (84.8)	3 (9.1)
Hemangioma	30	0 (0.0)	0 (0.0)	1 (3.3)	1 (3.3)	28 (93.3)

Overall diagnostic accuracy: 98/137 (71.5%). CAD performance was evaluated by a leave-one-case-out methods. Reproduced, with modification, form Sugimoto et al[31], Acad Radiol 2009; 16: 401-411. CAD: Computer-aided diagnosis; HCC: Hepatocellular carcinoma; w-HCC: Well-differentiated HCC; m-HCC: Moderately differentiated HCC; p-HCC: Poorly differentiated HCC.

Table 2 Performance of CAD scheme for classification in three categories using physicians’ subjective pattern classification n (%)

Lesion	n	Classification with CAD
		HCC	Metastasis	Hemangioma
HCC	74	73 (98.6)	1 (1.4)	0 (0.0)
Metastasis	33	2 (6.1)	28 (84.8)	3 (9.1)
Hemangioma	30	1 (3.3)	1 (3.3)	28 (93.3)

Overall diagnostic accuracy: 129/137 (94.2%). CAD performance was evaluated by a leave-one-case out method. Reproduced, with modification, form Sugimoto et al[31], Acad Radiol 2009; 16: 401-411.

Table 3 Image feature values used for CAD input data

Image feature values
Temporal features
Replenishment time (s)
Peak pixel value
Slope factor (β)
Morphologic features
Effective diameter of focal liver lesion
Average size of vessel-like patterns
Area ratio of vessel-like patterns
Gray-level features
Average pixel value with vessel-like patterns
Average pixel value without vessel-like patterns
Standard deviation of pixel value with vessel-like patterns
Standard deviation of pixel value without vessel-like patterns
Average pixel value ratio (focal liver lesion/adjacent liver parenchyma)
Average pixel value ratio (central/peripheral)
Features for hypoechoic region
Average pixel value
No. of hypoechoic regions
Area ratio of hypoechoic region
Difference in pixel value (delay-early)
Change in pixel value (delay-early)/s

Reproduced, with modification, from Shiraishi et al[32], Med Phys 2008; 35: 1734-1746.

Table 4 Performance of CAD scheme for classification in five categories using computerized scheme n (%)

Lesion	n	Classification with CAD
		HCC			Metastasis	Hemangioma
		w-HCC	m-HCC	p-HCC
Total	103
w-HCC	24	19 (79.2)	1 (4.2)	2 (8.3)	2 (8.3)	0 (0.0)
m-HCC	28	5 (17.9)	14 (50.0)	4 (14.3)	3 (10.7)	2 (7.1)
P-HCC	9	1 (11.1)	0 (0.0)	7 (77.8)	1 (11.1)	0 (0.0)
Metastasis	26	2 (7.7)	1 (3.8)	0 (0.0)	23 (88.5)	0 (0.0)
Hemangioma	16	0 (0.0)	0 (0.0)	0 (0.0)	1 (6.3)	15 (93.8)

Overall diagnostic accuracy: 78/103 (75.7%). CAD performance was evaluated by a leave-one-case-out methods.Reproduced, with modification, from Shiraishi et al[32], Med Phys 2008; 35: 1734-1746.