Design and Implementation - Segmentation of Hand Bone for Bone Age Assessment - page 89

Fig. 3.20 Two arbitrary blocks in segmented bone image. ( a ) This bone image contains extra
regions that should be eliminated. The range of pixels intensity within the red rectangular boxes
(spongy bone pixels) is overlapping with range of pixels intensity within the blue rectangular
boxes (soft-tissue pixels). The former group of pixel needs to be maintained, and the latter group
of pixels need to be eliminated ( b ) this bone image has been over-segmented and the regions that
undergo this artifacts are regions in spongy bone as expected, which are pixels within the red rec-
tangle. Those missing pixels have to be restored
3.5 Quality Assurance Process
Despite previous schemes, not all pixels are perfectly classified via ACR k-mean
clustering in quadruple division framework is correct due to the inherent nature
of hand bone radiograph that the pixel intensity range of soft-tissue region and
the spongy bone is overlapping. Therefore, most of the errors are due to incor-
rect classification of soft-tissue region and spongy bone. For instance, the pixels
that belong to soft-tissue region have been wrongly classified as spongy bone and
hence they have been included in the segmented hand bone image; the pixels that
belong to the spongy bone have been wrongly classified as pixels in soft-tissue
region and hence have been excluded in the segmented hand bone image. The
extra pixels in the former case need to be eliminated and the wrongly excluded
pixels in the latter case need to be restored to obtain an accurate segmented hand
bone image. It is not trivial at all to detect the wrongly classified pixels and per-
form the restoration and elimination. Therefore, we proposed a restoration and
elimination system by using the Bounded-area Restoration and Non-bounded Area
Elimination (BARNAE) with some carefully chosen and designed features. The
detail of the features chosen and the illustration of the Restoration and elimination