RATIONALE AND OBJECTIVE: To evaluate the ultrasonographic features of breast masses using a computerized scheme and to correlate the feature values with radiologists' grading. MATERIALS AND METHODS: One hundred and seventy-five breast ultrasound images (one to five images per subject) from 61 women (age 17-89 years, mean 43 years) were studied. Thirty-eight of the 157 images were from 11 women with malignant lesions, and the remaining 137 were from 50 patients with benign lesions. Two breast imaging radiologists participated in an observer performance study and were asked to grade, on a scale of 3, shape (1: regular, 3: very irregular), border (1: sharp, 3: ill-defined), internal texture (1: homogeneous, 3: very heterogeneous), width/depth ratio (1: flat, 3: tall), posterior enhancement (1: strong, 3: none), and lateral shadowing (1: strong, 3: none). The computerized scheme analyzed the breast region within a region of interest that was placed by a radiologist and quantified the following parameters: shape (jag count, disperse, convex hull depth, and lobulation count), border (acutance, average maximum ascending gradient, and sigmoid curve fitting), texture (edge density, co-occurrence matrix, and fractal dimension), width-depth ratio, posterior enhancement, and lateral shadowing. Correlations between the radiologists and the computerized scheme for assessing parameters in corresponding categories were computed. RESULTS: Good agreement was seen in posterior enhancement (P < .001, r = 0.45), lateral shadowing (P < .001, r = 0.38), width-depth ratio (P < .001, r = 0.33), and shape features (all P < .001): jag count (r = 0.38), disperseness (r = 0.55), and convex hull depth (r = 0.44). The remaining parameters demonstrated a poor or weak correlation (r < 0.30). CONCLUSION: The radiologists and the computerized scheme correlated best in analysis of shape features and posterior enhancement. We have yet to determine the significance of these features for the implementation of a computer-aided diagnosis program for characterizing breast ultrasound masses.
RATIONALE AND OBJECTIVE: To evaluate the ultrasonographic features of breast masses using a computerized scheme and to correlate the feature values with radiologists' grading. MATERIALS AND METHODS: One hundred and seventy-five breast ultrasound images (one to five images per subject) from 61 women (age 17-89 years, mean 43 years) were studied. Thirty-eight of the 157 images were from 11 women with malignant lesions, and the remaining 137 were from 50 patients with benign lesions. Two breast imaging radiologists participated in an observer performance study and were asked to grade, on a scale of 3, shape (1: regular, 3: very irregular), border (1: sharp, 3: ill-defined), internal texture (1: homogeneous, 3: very heterogeneous), width/depth ratio (1: flat, 3: tall), posterior enhancement (1: strong, 3: none), and lateral shadowing (1: strong, 3: none). The computerized scheme analyzed the breast region within a region of interest that was placed by a radiologist and quantified the following parameters: shape (jag count, disperse, convex hull depth, and lobulation count), border (acutance, average maximum ascending gradient, and sigmoid curve fitting), texture (edge density, co-occurrence matrix, and fractal dimension), width-depth ratio, posterior enhancement, and lateral shadowing. Correlations between the radiologists and the computerized scheme for assessing parameters in corresponding categories were computed. RESULTS: Good agreement was seen in posterior enhancement (P < .001, r = 0.45), lateral shadowing (P < .001, r = 0.38), width-depth ratio (P < .001, r = 0.33), and shape features (all P < .001): jag count (r = 0.38), disperseness (r = 0.55), and convex hull depth (r = 0.44). The remaining parameters demonstrated a poor or weak correlation (r < 0.30). CONCLUSION: The radiologists and the computerized scheme correlated best in analysis of shape features and posterior enhancement. We have yet to determine the significance of these features for the implementation of a computer-aided diagnosis program for characterizing breast ultrasound masses.