Enhanced Deep Learning-Based CAD System for Breast Cancer Diagnosis from Histopathology Images: Whole Slide Approach
Omnia Salah Elassal, Rawan Ramdan Gomaa, Salma Ayman Elsayed, Ahmed Hesham Mohammed, Hesham Ali Ahmed · 2023
Breast cancer is a major cause of cancer-related death globally, particularly in women. Early diagnosis is crucial for successful treatment, requiring proper analysis of histological images. However, the large data volume and image complexity make this task time-consuming, prompting the need for automated detection tools. The paper proposes a methodology that combines multiple deep learning models (VGG16, VGG19, Xception, Inception V3, and InceptionResnet V2) to improve accuracy in identifying cancerous and noncancerous regions in histopathology images, assisting doctors in earlier detection and effective treatment. The study demonstrates breast cancer histology image classification using deep learning. The evaluation on the ICIAR 2018 Grand Challenge dataset showcases superior performance compared to state-of-the-art methods. The proposed method achieves a high-test accuracy of 96% for the two-class problem, surpassing individual deep learning models like Xception. For the four-class problem, it outperforms Xception and InceptionResNet V2 with a test accuracy of 95%. Notably, precision and recall metrics significantly improve after applying the method in both cases. Furthermore, the evaluation on whole slide images demonstrates an accuracy of 92%, highlighting the model's proficiency in classifying breast cancer histology images at a macroscopic level. Robustness is confirmed through testing on an independent dataset from Alborg lab, with accuracies ranging from 0.82 to 0.91 across different architectures. Overall, the proposed methodology offers promising advancements in breast cancer diagnosis automation, supporting earlier detection and more effective treatment of the disease.