A Super-Resolution Generative Adversarial Network with Siamese CNN Based on Low Quality for Breast Cancer Identification
Grace Ugochi Nneji, Jingye Cai, Deng Jianhua, Happy Nkanta Monday, Chukwuebuka Joseph Ejiyi, Edidiong Christopher James, Goodness Temofe Mgbejime, Ariyo Oluwasanmi · 2021
Breast cancer is a chronic illness leading to the death of millions of people yearly. Despite the fact that successful identification of benign and malignant images is dependent on radiologists' long-term knowledge, specialists occasionally disagree with their decisions. An automatic system provides an alternative choice for the image diagnosis, thereby helping the expert to make more reliable decisions efficiently, less prone to errors and make diagnosis more scalable. Another issue based with the diagnosis of breast cancer identification is the poor quality of the image which poses a challenge in identification performance. An enhanced super-resolution generative adversarial network has been implemented in this paper to produce super-resolution images of breast cancer from a low-resolution counterpart with higher quality and finer details using an upscale factor of 4. Additionally, siamese convolutional neural network was utilized for the features extraction and classification of breast cancer. The proposed model provides an effective classification performance in terms of accuracy and ROC-AUC scores of 98.87% and 98.76% respectively as compared to other existing approaches.