Abstract:
Tuberculosis is an infectious disease and becomes a major threat all over the world but still
diagnosis of tuberculosis is a challenging task. In literature, chest radiographs are consid-
ered as most commonly used medical image in under developed countries for the diagnosis of
TB. Di erent methods have been proposed, some systems are too slow and some are expen-
sive especially not a ordable in under developed areas. Our paper presents a methodology in
which segmentation technique is applied to get lung eld region and dominant features are
extracted for detection of TB.In segmentation technique, lung model is calculated according
to similarity measure and kernel graph cuts segmentation technique is used to partition the
image, through the kernel mapping of image data, into di erent graph cut regions. Kernel
function is used to transform image data to make piecewise constant model of graph cuts be-
comes applicable. Evaluation of deviation of transformed data is done by piecewise constant
model and smoothness cost. Minimization of energy function contains graph cut iterations
of image partitioning and region parameters are evaluated by using kernel function. After
segmentation, di erent combinations of features are extracted based on intensities, shape
and texture of chest radiograph and given to classi er for the detection of TB. To reduce
dimensionality and increase e ciency, dominant features are computed and given to classi-
er for the detection of tuberculosis (TB). The performance of our methodology is evaluated
on three publically available standard datasets by using parameters such as accuracy, speci-
city and sensitivity. The improvement performance is demonstrated by comparing result
with recently proposed and published methods. The results showed that proposed method
have high accuracy as compared to previous methods. This research will assist radiologists
in saving their time.
