University Links: Home Page | Site Map
Covenant University Repository

CLASSIFICATION OF BREAST CANCER HISTOPATHOLOGY IMAGES USING A CONVOLUTIONAL NEURAL NETWORK MODEL

SIMONYAN, EMMANUEL OLUWATOBI and Covenant University, Theses (2022) CLASSIFICATION OF BREAST CANCER HISTOPATHOLOGY IMAGES USING A CONVOLUTIONAL NEURAL NETWORK MODEL. Masters thesis, Covenant University.

[img] PDF
Download (271kB)

Abstract

Among the different cancers that exist, breast cancer has been identified to account for about 2.26 million new cases among women globally in 2020 according to WHO. The early diagnosis of breast cancer can help reduce the mortality rate. Due to the large volume of breast cancer cases and the limited availability of histopathologists and clinicians, the available ones can be subjective which can lead to misjudgement. An intelligent system that can assist the limited histopathologist is crucial to help in optimal diagnosis. It has been identified that dataset usually came from one source, and a custom CNN was required to train. Therefore, this dissertation aims to employ Convolutional Neural Network models for accurate classification of breast cancer histopathology images curated from different dataset sources. This work utilised two datasets at different magnifications, the BreakHis and the Breast Histopathology dataset. A hybrid dataset was created from these two datasets and divided into 70% and 30% for training and testing. Four pre-trained Convolutional Neural Network (CNN) models (DenseNet201, ResNet50, ResNet101 and MobileNet-v2) were used for the analysis after preprocessing and rescaling. The findings show that DenseNet201 achieved the highest classification accuracy of 88.17%, 87.73%, 92.2% and 91.4% for BreakHis Dataset at 40X, 100X, 200X, 400X magnification factors respectively; 83.67% for Breast Histopathology Dataset at 200X and 85.78% for the Hybrid dataset at 200X. The models were able to classify the images between benign and malignant images, with DenseNet201 giving the best performance in terms of Specificity and Sensitivity at 100%. The implication is that the DenseNet201 model can be used to accurately differentiate between benign and malignant histopathology breast images thus serving as a decision support system in the early diagnosis of breast cancer.

Item Type: Thesis (Masters)
Uncontrolled Keywords: Convolutional Neural Network (CNN), histopathology images, breast cancer, image classification, intelligent system, diagnosis.
Subjects: T Technology > T Technology (General)
T Technology > TK Electrical engineering. Electronics Nuclear engineering
Divisions: Faculty of Engineering, Science and Mathematics > School of Engineering Sciences
Depositing User: AKINWUMI
Date Deposited: 27 Jan 2023 08:46
Last Modified: 27 Jan 2023 08:46
URI: http://eprints.covenantuniversity.edu.ng/id/eprint/16555

Actions (login required)

View Item View Item