Pengekstrakan ciri-ciri sel pangkal rahim berdasarkan imej thinprep® bagi sistem diagnostik kanser pangkal rahim
Loading...
Date
2007-04
Authors
Mustafa, Nazahah
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
ThinPrep® monolayer cytology was introduced to overcome limitations of conventional
Pap smear test in screening for cervical cancer. The cytology images seen in
ThinPrep® could be improved if unwanted background information and poor contrast
could be eliminated. This study focuses on developing an image processing system for
ThinPrep® image. Automatic region growing (ARG) algorithm is proposed to remove
the background area of the ThinPrep® image and segment the image into nucleus and
cytoplasm regions automatically. The proposed algorithm maintains the original size
and shape of the cervical cell and is capable to segment more than one cervical cell at
the same time. This study also proposes automatic feature extraction (AFE) and
manual feature extraction (MFE) algorithms. These algorithms are able to execute
segmentation and feature extraction functions simultaneously. Besides, the AFE
algorithm is capable to extract more than one cervical cell at the same time. The AFE
and MFE algorithms are used to extract nine features of cervical cell namely size,
average grey level, perimeter, average red colour, average green colour, average blue
colour, average intensity1, average intensity2 and average saturation. The result of
correlation tests shows that these algorithms provide good performance in extracting
features of cervical cell as good as manual extraction by cytotechnologists. Then, these
features are analyzed to verify their suitability as input data to the H2MLP network in
order to detect normal, LSIL and HSIL cells. Combination of seven dominant features,
namely size, average grey level, average red colour, average green colour, average
intensity1, average intensity2 and average saturation are capable to produce good
diagnosis performance with 94.09%. This study also proposes the contrast
enhancement technique on segmented ThinPrep® image (i.e: without the information of
the backgroud area). The linear contrast algorithm is implemented to increase the
contrast of the whole cervical cell region. The non-linear bright and non linear-dark
contrast enhancement algorithms are implemented to increase the contrast of
cytoplasm anrJ nucleus region respectively. While the combination of non-linear bright
and non-linear dark contrast enhancement algorithm is implemented to increase the
contrast of cytoplasm and nucleus region simultaneously. The proposed contrast
enhancement technique produces better results than the conventional techniques. With
aforementioned advantages, the proposed image processing system of the ThinPrep®
image is suitable to be applied in NeuraiPap system in order to increase its
performance.
Description
Keywords
Ciri-ciri sel pangkal rahim berdasarkan imej thinprep , Sistem diagnostik kanser