Purpose: The aim of this study was to investigate the feasibility of quantitative image analysis to differentiate hepatic nodules on gray-scale sonographic images.
Methods: We retrospectively evaluated 35 nodules from 31 patients with hepatocellular carcinoma (HCC), 60 nodules from 58 patients with liver hemangioma, and 22 nodules from 22 patients with liver metastasis. Gray-scale sonographic images were evaluated with subjective judgment and image analysis using ImageJ software. Reviewers classified the shape of nodules as irregular or round, and the surface of nodules as rough or smooth.
Results: Circularity values were lower in the irregular group than in the round group (median 0.823, 0.892; range 0.641-0.915, 0.784-0.932, respectively; P = 3.21 × 10(-10)). Solidity values were lower in the rough group than in the smooth group (median 0.957, 0.968; range 0.894-0.986, 0.933-0.988, respectively; P = 1.53 × 10(-4)). The HCC group had higher circularity and solidity values than the hemangioma group. The HCC and liver metastasis groups had lower median, mean, modal, and minimum gray values than the hemangioma group. Multivariate analysis showed circularity [standardized odds ratio (OR), 2.077; 95 % confidential interval (CI) = 1.295-3.331; P = 0.002] and minimum gray value (OR 0.482; 95 % CI = 0.956-0.990; P = 0.001) as factors predictive of malignancy. The combination of subjective judgment and image analysis provided 58.3 % sensitivity and 89.5 % specificity with AUC = 0.739, representing an improvement over subjective judgment alone (68.4 % sensitivity, 75.0 % specificity, AUC = 0.701) (P = 0.008).
Conclusion: Quantitative image analysis for ultrasonic images of hepatic nodules may correlate with subjective judgment in predicting malignancy.
Keywords: Hepatocellular carcinoma; Liver hemangioma; Liver metastasis; Quantitative image analysis; Ultrasonography.