Shearlet transform and convolutional neural network for histopathology images in breast cancer classification

Breast cancer stands out as one of the global health threats, as it may cause death if improperly treated. Thus, detecting the illness at the early stage through precise diagnosis is important to prevent progression of tumors with effective treatments through medical imaging. Traditionally, manual diagnostic processes rely on the input data representation and expert knowledge, which consume much time and are prone to human error due to heavy workloads and fatigue. Recently, deep learning has shown distinguishing results in medical imaging analysis for image classification and detection. Nevertheless, the increasing demand to enhance the performance of image classification is becoming more prominent. In this study, a hybrid method of deep learning is proposed by combining Shearlet transform and convolutional neural network (CNN) for breast cancer histopathology image classification. First, the histopathology images are decomposed using Shearlet transform for Shearlet coefficients. Then, the CNN approach is used to classify the images into benign and malignant with minimal pre-processing procedure. The ability of Shearlet transform to address singularities helps to increase the quality of images. The proposed hybrid model improves the performance of the original basic CNN model. Results from the experiment show that the proposed hybrid model achieves an accuracy of 75%, an F1-score of 85% for malignant tumor, and a misclassification rate of 0.25%. This result shows that the use of Shearlet transform as the first feature extraction layer in the CNN architecture provides better feature extraction, consequently leading to improved accuracy for image classification.