Study of the automated breast tumor extraction using 3D ultrasound imaging: The usefulness of depth-width ratio and surface-volume index

Abstract

We applied quantitative parameters in three-dimensional ultrasonic images to distinguish benign from malignant breast tumors in 29 benign cases including 8 cysts and 21 fibroadenomas, and 32 malignant cases including 23 ductal carcinomas, 2 special types of carcinoma, 1 malignant lymphoma and 6 others. This procedure involved simultaneously acquiring video data from real-time ultrasonic images and recording the original position and orientation of the probe. Both sets of data were fed directly into a desktop computer. Fuzzy reasoning and relaxation techniques were use to automatedly extract the shape of the tumor and render it in three dimensions. We then evaluated three parameters: 2D-D/W, the so-called depth-width ratio measured in B-mode images: 3D-D/W; and the S/V index ([surface area](3)/36π [volume](2)) calculated from the three-dimensional volume extracted with this system. All three parameters were significantly higher in the malignant group (averages: 0.81, 0.64, and 11.3, respectively) than in the benign group (averages: 0.62, 0.47, and 3.78, respectively). All three parameters were thus found to be useful in differentiating the two groups.

Quantitative evaluation of diagnostic information around the contours in ultrasound images

“…[7][8][9]26,27 In the present study, we calculated the degree of circularity, the degree of irregularity, and the depth-width ratio as indicators of typical morphological characteristics. Among these indicators, the degree of circularity corresponds to the deformity index.…”

“…Average interclass variance near the border ( ) A 15 ¥ 15 pixel window is placed on the obtained border, and the variance s 2 B between the two regions (classes) is calculated and the average along the circumference obtained. Interclass variance is calculated using the following equation: (9) where N 1 , ,N 2 , and denote the pixel number of each class and the mean density. N and P -indicate the total pixel number for each window and the mean density within the window, respectively.…”