Deep Learning Approach for Breast Cancer Diagnosis from Microscopy Biopsy Images
sara Noor Eldin, Jana Khaled Hamdy, Ganna Tamer Adnan, Maysoon Hossam, Noha ElMasry, Ammar Mohammed
Abstract
Breast Cancer is the abnormal growth of cells in the breast. According to the World Health Organisation, Breast Cancer is the most common type of cancer that women are diagnosed with worldwide. Breast cancer tests vary from Mammograms to CTs and Ultrasounds; however, the only way to tell for sure if the suspicious lesions found in the breast are cancerous or not is by performing a biopsy test. The main contribution of this paper is proposing a deep learning approach to diagnose breast cancer from biopsy microscopy images. Several types of Deep Convolution nets such as Vgg16, Alexnet, Inception, Resnet50, Resnet101, and Densenet169 are used. The three best models that achieved the highest accuracy without data preprocessing techniques are Densenet169, Resnet50, and Resnet101 with accuracy 62%, 68%, and 85% respectively. Additionally, the paper shows the impact of various data preprocessing techniques on the performance of the best models. The experimental results indicate that data augmentation and segmentation increase the accuracy of the best models by 20%, 17%, and 6% respectively. To further boost the accuracy of the models, the paper aggregates the best models using an ensemble learning technique. The results reveal that the best-achieved accuracy is 92.5%.