Author/s:
Sema Nizam Abdulghani1,*, Maha Mohammed Attieya2, Ahmed Freidoon Fadhil3, and Seyfettin Sinan Gültekin4
1 Department of Physics, University of Kirkuk
semanizam@uokirkuk.edu.iq
2 Department of Information Technology, University of Kirkuk
mahamohammed@uokirkuk.edu.iq
3 Electrical Engineering Department, University of Kirkuk
ahmedfadhil83@uokirkuk.edu.iq
4 Electrical and Electronics Engineering Department, Konya Technical University
ssgultekin@ktun.edu.tr
DOI: https://doi.org/10.31972/iceit2024.049
Abstract
Breast Cancer is the most common cancer among women worldwide. The most widely used method for diagnosis of this types of cancer is the Histopathological analysis. Many researchers focused on developing computer-aided diagnosis system to support pathologist experience for correct diagnoses. This paper proposes an improved model for CNN using the publically available (BreakHis dataset) to categorize breast cancer histopathological images. The proposed model uses a 2-dimensional discrete wavelet transform (DWT) for feature extraction. Then, the approximate sub-band will be used for training and testing of the CNN instead of the original raw images. It is observed that the use of DWT features attain better results than the use of raw images itself. For instance, proposed MDWT2 CNN method has showed a better performance compared to the previous published of binary classification task with a performance range between 88.9% and 89.9% at image levels. A further investigation was implemented using the step scheduler with increasing mini-batch size for updating Stochastic Gradient Descent (SGD) network parameters at the early stages of the training phase. The results show that the proposed CNN based DWT model achieves the highest accuracy of 90.8% at the patient level and 89.1% at image level, thus outperforms all the previous published approaches. Therefore, the researchers could suggest using the proposed MDWT2 CNN model for future investigation of breast cancer diseases.
Keywords: Breast Cancer; Convolutional Neural Networks; Discrete Wavelet Transform; and Stochastic Gradient Descent.