(For USM Staff/Student Only)

EngLib USM > Ω School of Electrical & Electronic Engineering >

Development of image segmentation and feature extraction technique for cervical cell images

Development of image segmentation and feature extraction technique for cervical cell images / Tan Yi Hui
Palitan Pap adalah satu teknik konvensional yang digunakan untuk memeriksa peringkat sel pangkal rahim. Walau bagaimanapun, kesilapan teknikal seperti teknik persampelan yang kurang sesuai, alat persampelan yang tidak sesuai, dan kesilapan manusia seperti keletihan mata, keletihan, dan lain-lain berkemungkinan akan berlaku disebabkan oleh pemeriksaan manual. Oleh itu, peruasan imej dan pengekstrakan ciri untuk imej sel pangkal rahim dibangunkan untuk membantu ahli patologi semasa proses diagnosis. Peruasan imej adalah satu proses yang membahagikan imej kepada beberapa kawasan berdasarkan ciri tekstur yang sama. Teknik peruasan berdasarkan tiga algoritma pengelompokan, iaitu Purata-K (KM), Penyesuaian Purata-K (AKM), dan Purata-C Fuzi (FCM) telah dicadangkan dan yang terbaik akan dipilih untuk bergabung dengan operasi morfologi untuk membahagikan sel pangkal rahim kepada tiga kawasan, iaitu nukleus, sitoplasma dan latar belakang. Analisis kualitatif membuktikan bahawa peruasan berdasarkan algoritma pengelompokan dengan operasi morfologi mampu meruaskan sel pangkal rahim kepada tiga kawasan tersebut. Hasil uji kaji menunjukkan bahawa daripada tiga teknik peruasan tersebut, AKM menunjukkan prestasi yang terbaik kerana ia mampu meruaskan ketiga-tiga kategori sel pangkal rahim lebih baik daripada KM dan FCM. Ia meruaskan imej lebih baik daripada KM dan FCM dengan menunjukkan kesilapan pada kawasan salah berkelompok yang lebih rendah. Pengekstrakan ciri dilakukan untuk mengeluarkan ciri-ciri yang mengandungi maklumat penting daripada imej yang telah diruaskan. Lima ciri-ciri, iaitu saiz nucleus, saiz sitoplasma, tahap kelabu nucleus, tahap kelabu sitoplasma dan nisbah nukleositoplasma diekstrak daripada imej yang telah diruas. Analisis kuantitatif dijalankan bagi menilai prestasi ciri-ciri yang telah diekstrak. Ciri-ciri yang diekstrak akan dinilai bagi menentukan keupayaannya untuk dilatih sebagai atribut kepada Sokongan Mesin Vektor (SVM), untuk mengelaskan sel pangkal rahim kepada tiga kategori iaitu, Normal, Gred- Rendah Squamous Intraepithelial Luka (LSIL), dan Gred-Tinggi Squamous Intraepithelial Luka (HSIL). Berdasarkan keputusan klasifikasi, jenis atribut dengan ketepatan yang lebih daripada 90% akan digabung dan diuji untuk mengelaskan sel pangkal rahim. Keputusan ini menunjukkan ketepatan yang tinggi untuk tiga jenis atribut iaitu, saiz nucleus, saiz sitoplasma, dan tahap kelabu sitoplasma. Atribut gabuangan yang digunakam untuk proses pengelasan menunjukkan hasil yang baik dengan ketepatan yang tinggi. Penemuan ini menunjukkan bahawa peruasan berdasarkan algoritma pengelompokan dengan operasi morfologi dan pengekstrakan ciri mempunyai potensi yang tinggi untuk diaplikasikan pada masa depan. _______________________________________________________________________________________________________ Pap smear is a conventional technique used to examine cervical cell stages. However, technical error such as poor sampling technique, inappropriate sampling tools, and human error such as eye fatigue, tiredness, etc. tend to occur due to manual inspection. Thus, image segmentation and feature extraction technique for cervical cell images is developed in this study to assist pathologist during diagnosis process. Image segmentation is a process which divides an image into region of interest (ROI) with homogenous texture. A segmentation technique based on three clustering algorithms, namely K-Means (KM), Adaptive K-Means (AKM), and Fuzzy C-Means (FCM) are proposed and the best will be chosen to be combined with morphological operations in order to segment cervical cell into three regions, namely nucleus, cytoplasm and background. Qualitative analysis proves that segmentation based on clustering algorithms with morphological operations is able to segment cervical cell into three ROI. Experimental results show that out of the three segmentation techniques, AKM performed the best as it is able to segment the three categories of cervical cell better than KM and FCM. It performs better than KM and FCM with lower error of wrongly clustered regions. From the segmented images, feature extraction is employed to extract features with significant information. Five features, namely size of nucleus, size of cytoplasm, average gray level of nucleus, average gray level of cytoplasm and nucleocytoplasmic ratio is extracted from the segmented image. Quantitative analysis is employed to evaluate the performance of feature extraction process. Each of the extracted features will be evaluated by determining its capability to be used as attributes for Support Vector Machines (SVM), to classify cervical cell into Normal, Low-Grade Squamous Intraepithelial Lesion (LSIL), and High-Grade Squamous Intraepithelial Lesion (HSIL). From the result of classification, the types of attribute with accuracy which more than 90% are combined and tested to classify the type of cervical cell. The results show high accuracy for three attributes, namely the size of nucleus, size of cytoplasm, and cytoplasm gray level. These three attributes are combined and tested. The combined attributes which are used for classification process show a good result with high accuracy. These findings suggest that segmentation based on clustering algorithm with morphological operations and feature extraction has high potential for future work.
Contributor(s):
Tan Yi Hui - Author
Primary Item Type:
Final Year Project
Identifiers:
Accession Number : 875006336
Language:
English
Subject Keywords:
Pap smear; conventional technique; cervical cell
First presented to the public:
6/1/2016
Original Publication Date:
6/5/2018
Previously Published By:
Universiti Sains Malaysia
Place Of Publication:
School of Electrical & Electronic Engineering
Citation:
Extents:
Number of Pages - 137
License Grantor / Date Granted:
  / ( View License )
Date Deposited
2018-06-05 15:11:03.308
Date Last Updated
2019-01-07 11:24:32.9118
Submitter:
Mohd Jasnizam Mohd Salleh

All Versions

Thumbnail Name Version Created Date
Development of image segmentation and feature extraction technique for cervical cell images1 2018-06-05 15:11:03.308