(For USM Staff/Student Only)

EngLib USM > Ω School of Electrical & Electronic Engineering >

Classification of cervical cell using deep convolutional neural network

Classification of cervical cell using deep convolutional neural network / Poh Kok Hang
Pengklasifikasian kanser serviks secara automatic bermungkinan berkesan untuk membantu proses diagnosis kanser serviks melalui palitan Pap. Projek ini mencadangkan penggunaan rangkaian neural konvolusi dalam untuk klasifikasi sel serviks bagi mengatasi kelemahan kaedah klasifikasi automatik yang terdahulu. Rangkaian neural konvolusi dalam boleh mengklasifikasikan sel serviks secara langsung tanpa mengambil kira ciri sel individu, berbanding kaedah terdahulu yang memerlukan penggunaan segmentasi sel dan proses pengekstrakan ciri yang merupakan tugas mencabar. Kelebihan dan kekurangan kaedah klasifikasi sel serviks yang dicadangkan sebelum ini akan diberi penekanan dalam kajian ini. Dalam kajian ini, data sel serviks yang didapati daripada hospital tempatan dan dataset Herlev dipilih secara manual dan dilabelkan kepada tiga kategori, iaitu Negatif bagi Malignansi Intraepithelial, Intraepithelial Squamous Lesion Gred Rendah dan Intraepithelial Squamous Lesion Gred Tinggi. Kemudiannya, dataset dipisahkan untuk proses latihan dan ujian. Akhir sekali, model rangkaian neural konvolusi dalam akan dilatih dan dioptimumkan untuk mengklasifikasikan sel-sel serviks. Keputusan menunjukkan bahawa model yang dibina mencapai sensitiviti, spesifisiti dan kejituan klasifikasi masing-masing 93.6%, 94.8% dan 91.4%. Keputusan tersebut telah mengatasi kebanyakan kaedah yang dicadangkan sebelum ini. Tambahan pula, masa yang diambil untuk mengklasifikasikan imej sel serviks tunggal adalah kecil iaitu 2.1 milisaat. Oleh itu, kaedah yang dicadangkan menyediakan cara yang boleh dipercayai untuk pembangunan sistem pemeriksaan automatik yang akan membantu profesional dalam pemeriksaan asas untuk membezakan sel-sel serviks. _______________________________________________________________________________________________________ Automated cervical cancer classification could be effective for assisting the process of cervical cancer diagnosis via Pap smear. This project proposes to implement deep convolutional neural network for the classification of cervical cell to overcome the weaknesses of previous automated classification methods. Deep convolutional neural network can directly classify cervical cells without taking considerations of individual cell features. Whereas, most of the previous methods require the use of cell segmentation and feature extraction processes, which remain a challenging task despite previous extensive researches. This research will further address the advantages and disadvantages of previous proposed cervical cell classification method. In this research, the cervical cell datasets obtained from local hospitals and Herlev dataset are first manually chosen and labelled into three categories, namely Negative for Intraepithelial Malignancy (NILM), Low-grade Squamous Intraepithelial Lesion (LSIL) and High-grade Squamous Intraepithelial Lesion (HSIL). Then, the datasets are separated for training and testing process. Finally, a deep convolutional neural network model is trained and optimized to classify cervical cells. The results show that the developed model has a sensitivity, specificity and classification accuracy of 93.6%, 94.8% and 91.4% respectively which outperforms most of the previous proposed methods. Furthermore, the time taken to classify a single cervical cell image averages to 2.1 milliseconds. Hence, the proposed method provides a reliable method for the development of automated screening systems that will aid professionals in primary screening for abnormal cervical cells.
Contributor(s):
Poh Kok Hang - Author
Primary Item Type:
Final Year Project
Identifiers:
Accession Number : 875008671
Language:
English
Subject Keywords:
cervical; cancer; classification
First presented to the public:
6/1/2019
Original Publication Date:
3/4/2020
Previously Published By:
Universiti Sains Malaysia
Place Of Publication:
School of Electrical & Electronic Engineering
Citation:
Extents:
Number of Pages - 83
License Grantor / Date Granted:
  / ( View License )
Date Deposited
2020-03-04 16:29:26.176
Date Last Updated
2020-12-02 12:09:43.522
Submitter:
Mohd Jasnizam Mohd Salleh

All Versions

Thumbnail Name Version Created Date
Classification of cervical cell using deep convolutional neural network1 2020-03-04 16:29:26.176