Breast Cancer Diagnosis: A Transfer Learning-based System for Detection of Invasive Ductal Carcinoma (IDC)

Abstract

Breast cancer (BC) is a form of cancer that originates in the breast. BC cells typically form a tumor readily detectable on an X-ray or can be inspected as a lump. Even though improvements in screening, treatment, and monitoring have made patient survival rates higher, BC is still the common type of cancer in women and the main reason they die from cancer. Invasive ductal carcinoma (IDC) makes up about 85% of all cases that have been studied. It is the most aggressive type of BC. The purpose of this reserach is to find IDC early on so that it can be treated quickly. In the present study, we attempted to apply deep learning principles to detect breast-invasive ductal carcinoma via transfer learning. This study implemented six transfer learning approaches on two widely used datasets: Agios Pavlos and Databiox. The approaches utilized were InceptionReNetV2, DenseNet-121, DenseNet201, VGG19, ResNet-50, and ResNet-101v2. The models' efficacy in identifying IDC demonstrates their potential utility in assisting pathologists in illness diagnosis. The experimental results that the authors have attained concerning the accuracy: DenseNet121 with 97.13%, DenseNet201 with 96.34%, VGG19 with 95.65%, ResNet-50 with 94.90%, ResNet-101v2 with 93.53%, and InceptionReNetV2 with 93.20%. It is evident from the experimental findings that DenseNet121 provides the highest level of accuracy for cancer detection, while InceptionReNetV2 provides the lowest level of accuracy.