Proposed CAD method segments lung which contains an essential step. creating massive training soft computed method show an abnormal detect on soft computed image. So, an unwanted feature suppression being done from x ray. When suppressing unwanted feature, abnormal candidate are being suppressed. Fig 2 shows creating soft tissue technique with massive training soft computing method.
Massive training soft computing method (MTSCM) is non linear filter learns features from existing CAD method.
fb(u",v) =NN (au",v) ----------------------------------------------(1)
Where au",v = {g(u-i",v-j) |i",j є Rs } is vector to MTSCM which is extracted feature from x ray and fb(u",v) represents estimate for a teaching region.
MTSCM has trained by using subregions through teaching single pixels. The training set of pairs of sub region and teaching pixel represented by {a(u",v), T(i",j)| u",v є RT }
= { (a1 ",T1) (a2",T2) (aN",TN) }--------------------------(2)
Where T(i",j) is a teaching image, and N is number of vectors in RT.
For a single MTSCM, rib suppression containing various frequencies was difficult due to limited capability. Since we need to overcome such capability, multiresolution decomposition/composition techniques were applied. First, lower resolution region GL(u",v) was obtained from an original higher resolution region GH(2u",2v) by performing reduction in clarity i.e., lower clarity region are replaced by a four pixel mean value represented by
GL(u",v) = (1/4) ∑ GH(2u, 2v)---------------------------(3)
u",v є R22
where R22 represent 2 x 2 region. The replaced lower
clarity region has enlarged to middle clarity level by pixel substitution i.e., a pixel having lower clarity region is replaced by four pixels having middle clarity level, as follows:
GU(u",v)=GL(u/2",v/2)-------------------------------------------(4)
Then, we subtract middle clarity level from the higher clarity level to get high resolution difference region, represented by
DH(u",v) = GH(u",v) – GU(u",v) --------------------------------(5)
These procedures are done repeatedly producing lower resolution region. Thus, multiresolution region having various frequencies obtained by using machine learning technique.
GH(u",v)=GU(u",v)+DH(u",v) ------------------------------------(6)
Therefore, we choose multiresolution region by processing independently instead of processing original high resolution regions directly. When learning x ray image and the dual energy image, the multiresolution MTSCM produces teaching image. teaching image fb(u",v) produced by trained MTSCM with lung masking image n(u",v) and weighting parameter wc is subtracted from sub region of original x ray g(u",v) to create soft computed image.
f(u",v)= g(u",v) – wc x fb(u",v) x n(u",v)------------------------(7)
where f(u",v) denotes soft computed image having different types of rib contrast using weighting parameter wc. It segments abnormal detect on soft computed image by using clustering watershed method.
We extract sixty features from abnormal detect from x ray and soft computed image and proceed feature selection method. A non linear SVM with a Gaussian kernel was employed for the classification of abnormal detect into nodules or non-nodules.
