The ASU-1004 whole breast US scanner and its scanning procedure. (a) A female subject bent down to position her breast in a water coupling bath above the US membrane and the probe is immersed in a water coupling bath. (b) The probe is moved mechanically to project an entire breast within a area in three overlapping passes. Note that the image acquisition was obtained in a prone (nipple down) orientation but the images in Fig. 2 were presented in a supine orientation (nipple up).
Examples of whole breast US. (a) An original image from one pass. (b) A merged whole breast US image.
The grading results of radiologists A, B and C, and the ground standard.
The flowchart of the proposed analysis system. The original images are preprocessed to find the breast region and to divide the breast region into fibroglandular and fatty tissues. Subsequently, the proposed methods are applied for evaluating the density value of each testing case. Finally, each case is categorized according to the BI-RADS density classification.
Finding the breast region. (a) The skin line and the chest line are used to exclude the nonbreast regions. (b) The skin and chest lines are applied to all US images for locating the breast region.
An example of speckle noise removing. (a) The original image. (b) The result after applying the adaptive speckle reduction.
The category determination of the pixel in the threshold-based method. (a) A pixel on the result image would be determined as fibroglandular tissue, denoted as , if . The gray point, , represents fibroglandular tissue and the black point, , represents fatty tissue. (b) An example for the classified result image (right side) produced by applying the threshold-based method. The white region, , indicates areas classified as fibroglandular tissue.
The ground standard distribution of density classifications in the 32 test cases, which include 16 B-II cases, 12 B-III cases, and 4 B-IV cases.
Examples for the classified results, including BI-RADS II, BI-RADS III, and BI-RADS IV, produced by applying the threshold-based method. (a) The classified results. The white region indicates areas classified as fibroglandular tissue by the algorithm. (b) The corresponding mammographic film. Note that the breast morphology in mammography is remarkably different from that of ultrasound, which may be caused by positioning difference and the difference between compressed projection mammograms and uncompressed 3D US projections.
The breast density values of 32 experiment cases calculated by the threshold-based method and the proportion-based method, respectively. Note that the maximum/minimum values of threshold-based and proportion-based methods are 57/10 and 41/22, respectively.
The grading results for the threshold-based method, the proportion-based method, and the ground standard. Note that the grading results of the two methods are both worse for B-IV cases.
The confusion table of grading results for three radiologists. Note that B-I denotes BI-RADS I, B-II denotes BI-RADS II, B-III denotes BI-RADS III, and B-IV denotes BI-RADS IV.
The confusion table of the analysis results for the threshold-based method and the proportion-based method. Note that T denotes the true result, and F denotes the false result.
The MSE values show the concordance degree among reports by three radiologists. Note that A denotes radiologist A, B denotes radiologist B, and C denotes radiologist C.
The MSE values show the concordance degree among method 1, method 2, and G. Note that method 1 denotes the threshold-based method, method 2 denotes the proportion-based method, and G denotes the ground standard.
Article metrics loading...
Full text loading...