Archives

  • 2019-07
  • 2019-08
  • 2019-09
  • 2019-10
  • 2019-11
  • 2020-03
  • 2020-07
  • 2020-08
  • 2021-03
  • br S br similarity features of the selected pairs are

    2020-08-18

    423
    S
    (2) similarity features of the selected pairs are calculated, (3) 424
    f ðxÞ ¼ yiaiKðxi; xÞ þ b (4) correspondence score of every lesion is generated and (4) 425
    single view features and two view correspondence score are 426
    X
    training set, ai is the parameter for optimizing the margin, malignant tumour using SVM classifier. The detailed descrip- 428
    Kðxi; xÞ is the kernel function and b is the regularization pa- tion of each of these steps is as follows. 429
    rameter for minimizing training error [42]. In this study, radial
    basis function (RBF) with a variety of combinations of ðai; bÞ is 4.3.1. Localizing suspicious lesions 430
    obtained once at fixed interval values of ai and then with fixed Due to the compression of deformable breast during imaging 431
    interval values of b. The problem of SVM related to trade-off and variable landmarks, it is difficult to pinpoint exact location 432
    between the large margin and few misclassifications is solved of the physical lesion on ipsilateral views detected in Section 433
    here by optimized combination of ðai; bÞ parameters that 4.2. If two or more nonpalpable suspicious lesions are detected 434
    achieves highest accuracy [43,44]. The feature vectors of the on mammogram, the use of preoperative localization may 435
    segmented lesions are split into two 555-60-2 namely benign become painful for the patient. In this section, a technique to 436
    mass and malignant tumour. Both these classes are then determine precise radiologic localization on both the views is 437
    divided into five groups or folds randomly and evenly. Every proposed to help the radiologists in locating the lesion being 438
    time the classifier trained with 4 folds and the remaining fold biopsied. The suspicious lesion on either view can be matched 439
    is applied to test the classifier. Thus every fold is used as a test to the same physical lesion, if visible, on the other view. The 440
    set only once. This cross-validation step is repeated five times. proposed technique deals with locating the suspicious lesions 441
    Please cite this article in press as: Sapate S, et al. Breast cancer diagnosis using abnormalities on ipsilateral views of digital mammograms. Biocybern Biomed Eng (2019), https://doi.org/10.1016/j.bbe.2019.04.008
    Fig. 4 – Schematic flow of proposed fusion-based two views CAD system.
    442 on CC and MLO views of the same breast and forming the mammograms: chest wall on CC or pectoral muscle on MLO 449 443 candidate pairs of lesions representing the single physical view and the nipple centre. The chest wall in CC view may not 450 444 abnormality. be parallel and pectoral muscle may not be perfectly a straight 451 445 There are two popular methods in the literature: straight line which may lead to chaos in determining straight-line 452 446 line-based method and arc-based method [9,27] for locating distances. On the other hand, arc-based method uses only 453 447 the lesions and confirming their geometric relations. The nipple and centroid of the suspicious lesion and is compara- 454 448 straight line-based method requires two landmarks on the tively easy; hence is incorporated in lungfish work to locate the 455
    Please cite this article in press as: Sapate S, et al. Breast cancer diagnosis using abnormalities on ipsilateral views of digital mammograms.
    The annular width is dynamically selected and is limited
    within the breast parenchyma. The limited width reduces the 494
    search region and eliminates all the false positive candidates 495
    from set of suspicious lesions on the ipsilateral views. The 496
    matching process is depicted in Fig. 6. It starts with a breast 497
    coordinate system which has its origin on the central outer tip 498
    of the nipple. The distance between the centroid of the 499
    suspicious lesions C1 and nipple tip Nc, is defined as cr1, and 500
    DC1 is the length of minor axis of the lesion, C1, measured in 501
    radius, cr1, is drawn with nipple tip, Nm, on MLO view as centre 503
    central arc is drawn with width DW which is calculated using 505
    points, the centre of which is used to define breast midline 507
    Fig. 5 – The radial distances of the detected lesions on CC (maroon line) passing from nipple tip to the centre of chest 508
    wall as shown in Fig. 6(b). This midline divides the MLO view in 509
    view versus that of the MLO view from the nipple centre.
    two sections known as Upper and Lower quadrants, as shown 510
    in Fig. 6(b). The location of the lesion in a breast quadrant is