Details

Autor(en) / Beteiligte

• Khairi, Siti Shaliza Mohd
• Bakar, Mohd Aftar Abu
• Alias, Mohd Almie
• Bakar, Sakhinah Abu
• Liong, Choong-Yeun

Titel

A Preliminary Study of Convolutional Neural Network Architectures for Breast Cancer Image Classification

Ist Teil von

• 2021 IEEE Asia-Pacific Conference on Computer Science and Data Engineering (CSDE), 2021, p.1-5

Ort / Verlag

IEEE

Erscheinungsjahr

2021

Quelle

IEEE Electronic Library Online

Beschreibungen/Notizen

• Breast cancer is one of the most common cancer with high mortality rate worldwide. Classification of breast cancer images is an important clinical issue related to accurate early diagnosis and treatment plan preparation. However, it is still uncertain which model is effective for classifying breast cancer images. For medical image analysis, deep learning models have proved to yield excellent outcomes in classification tasks. Hence, this study compared the performance of the most common deep learning models which is convolutional neural networks for breast cancer classification on the histopathology images. A total of 7,909 images were extracted from BreakHis database that comprised of 2,480 benign and 5,429 malignant samples. The images are of four different magnifications which are 40X, 100X, 200X and 400X. This study focused on comparing the state-of the-art architectures, namely, AlexNet, GoogleNet and ResNet 18 to evaluate the performance of model in classifying the breast cancer images. The models were examined through a multiclass classification analysis in terms of accuracy, sensitivity, specificity and F-Score. The experimental results indicated that ResNet18 was the most effective method with an accuracy of 94.8% with 70 min 31 sec time taken for computation. The research findings are expected to facilitate the radiologist in classifying the breast cancer images and hence planning proper treatment for patients.