Page 31 - 2018

Ph.D.
(Computer Applications)
ANALYTICAL STUDY TO IMPROVE ACCURACY AND PRECISION TO
DETECT ABNORMAL TISSUES OF BRAIN MR IMAGES USING
CLASSIFIER AND CLUSTERING METHODS OF IMAGE SEGMENTATION
Ph.D. Scholar : Patel Shivang Manibhai
Research Supervisor : Dr. J. N. Dharwa

Regi. No.: 13146041001
Abstract :
Image analysis by segmentation techniques has wide range of utilities. The research on
image analysis includes industrial inspection, object tracking from multiple sequential
images, traffic control systems, geographical classification from satellite images,
geomorphological mapping at land surfaces, crop analysis in agriculture, treatment
planning in medical science and many others.

Segmentation of an image is very significant in medical field by improving treatment
possibilities like early diagnosis of abnormalities in brain. Manual segmentation for
detection of abnormal tissues from large amount of brain MR images are difficult and
time-consuming task.

For this research study, brain MR images are taken due to its quality parameters for the
purpose of detecting abnormalities compared to other techniques like X-Ray, CT Scan etc.
This study reviews the trade-off between various approaches for analyzing the
abnormalities from brain Magnetic Resonance Images using different methods of image
segmentation.

In addition, the proposed model has given a roadmap to improve accuracy and precision
to detect the region of interest. From multiple methods of segmentation, segmentation
using hybrid clustering approach followed by deep learning based classifier method
improves the level of precision to detect abnormal tissues from brain MR images. The
result leads to plan precise treatment for the medical practitioners.
This thesis presents an efficient approach of segmentation based on hybrid clustering
called KFCM in which K-means technique is integrated with Fuzzy C-means technique
and the acquired results are further taken as input in the Deep Learning based classifier
technique called DLBC to increase the level of accuracy in finding out ROI i.e. abnormal
tissues from brain MR images.
The proposed approach called HCADLC is applied on multiple brain MR images datasets
acquired from publicly available authenticate online sources (like Brainweb, BRATS 2013-
17 dataset etc.) and various imaging centers of Ahmedabad region. The performance of
the proposed approach is evaluated by comparing gathered outcomes with ground truth

12