Deep layer CNN architecture for breast cancer histopathology image detection

• Main Authors: Zainudin, Zanariah, Shamsuddin, Siti Mariyam, Hasan, Shafaatunnur
• Format: Conference or Workshop Item
• Published: 2020
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://eprints.utm.my/id/eprint/89787/; http://dx.doi.org/10.1007/978-3-030-14118-9_5
• Institution: Universiti Teknologi Malaysia

In recent years, there are various improvements in computational image processing methods to assist pathologists in detecting cancer cells. Consequently, deep learning algorithm known as Convolutional Neural Network (CNN) has now become a popular method in the application image detection and analysis using histopathology image (images of tissues and cells). This study presents the histopathology image related to breast cancer cells detection (mitosis and non-mitosis). Mitosis is an important parameter for the prognosis/diagnosis of breast cancer. However, mitosis detection in histopathology image is a challenging problem that needs a deeper investigation. This is because mitosis consists of small objects with a variety of shapes, and is easily confused with some other objects or artefacts present in the image. Hence, this study proposed three types of deep layer CNN architecture which are called 6-layer CNN, 13-layer CNN and 17-layer CNN, respectively in detecting breast cancer cells using histopathology image. The aim of this study is to detect the breast cancer cell which is called mitosis from histopathology image using suitable layer in deep layer CNN with the highest accuracy and True Positive Rate (TPR), and the lowest False Positive Rate (FPR) and loss performances. The result shows a promising performance for deep layer CNN architecture of 17-layer CNN is suitable for this dataset with the highest average accuracy, 84.49% and True Positive Rate (TPR), 80.55%; while the least False Positive Rate (FNR), 11.66% and loss 15.50%.