br used for preoperative localization before biopsy Similarl
511
used for preoperative localization before biopsy. Similarly, the
512
distance between the centroid of the suspicious lesions on
513
456
suspicious lesions and to eliminate false positives using
length of minor axis of the lesion, M1, measured in terms of
515
457
reduced annular search region. Secondly, in order to expose
pixels/Euclidean distance as shown in Fig. 6(b). The arc (green
516
458
the maximum portion of the breast for imaging, the nipple
line) with radius, mr1, is drawn with nipple tip, Nc, on CC view
517
459
during SQ109 is pulled away from the chest wall equally
as centre is shown in Fig. 6(b).
460
in CC and MLO views [28]. This results in an approximately
An arc pair (yellow band) around this central arc is drawn
519
461
constant distance between centre of nipple and the centroid of
with width DW which is calculated using Eqs. (6) and (7). The
520
462
suspicious lesion on both the views. This distance is useful
arc intersects the breast border at two points, the centre of
521
463
during mapping the corresponding lesions on ipsilateral
which is used to define breast midline (maroon line) from
522
464
views. The centre of nipple and centroid are marked under
nipple tip to centre of chest wall as shown in Fig. 6(c). This
523
465
the supervision of the radiologists involved in this study. A
midline divides the CC view in two sections known as Outer
524
466
scatter plot of these radial distances on the radiologist-
and Inner quadrants, as shown in Fig. 6(c). The location of the
525
467
identified objects on the ipsilateral views is shown in Fig. 5. It is
lesion in a breast quadrant is used for preoperative localization
526
468
revealed that there is a high correlation of the radial distances
before biopsy.
469
of the correlated lesions on the ipsilateral views. However, the
Thus the annular regions for all the suspicious lesions are
528
470
angular coordinates are found with a significant correlation.
drawn to limit the number of lesions to be paired. The square
529
471
Consequently, the radial distance is selected as the metric for
templates on each of the paired lesions are selected for
530
472
determining the correlation of a lesion on one view with the
extracting the features shown in Fig. 7(a) and (b).
473
same physical lesion on the other view.
The template is the square region centred on the centroid
532
The suspicious lesions are found in a wide variety of shapes
of the selected suspicious lesion belonging to either view. To
533
475
and size. Moreover, the radial distance of a physical lesion on
calculate the similarity features, the templates of the
534
476
CC view is different than that of on MLO view. Thus, both these
suspicious lesions on both views must be of equal size.
535
477
facts are considered for deciding the width of annular region
The size of the templates is equal to minimum of the minor
536
478
for optimal search. The first coefficient is calculated by
axis lengths the respective lesions. The square template
537
479
minimizing mean square error between the radial distance
within minor and major axis ensures the maximum of the
538
480
of ith true lesion on CC, Ri
, with its jth true distance on a MLO
common portion of the lesions is selected for extracting
539
Rj
cc
as given in Eq. (5).
similarity features. We conducted a small experiment on few
540
mlo
templates and calculated score by applying
similarity
541
features and traditional pixel by pixel template matching
542
¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffið
effective instead of pixel by pixel score as like in traditional
544
Ry
485 axis of the lesion on CC view, Lc. The width of annular region,
An important task in two view diagnosis is to find the same
547
abnormal lesion on both CC and MLO views. This is achieved
548
using similarity features of all possible pairs of lesions, one
549
Please cite stigma article in press as: Sapate S, et al. Breast cancer diagnosis using abnormalities on ipsilateral views of digital mammograms.
Fig. 6 – Breast coordinate system (a) CC view measuring R1cc, R2cc; (b) MLO view measuring R1mlo, R2mlo annular region within yellow arcs, midline (red) showing upper/lower quadrants; (c) CC view annular region within yellow arcs, midline (green) showing outer/inner quadrants,.
Fig. 7 – The lengths of minor axis of the sample lesions (a) on CC view and (b) on MLO view.