Breast Cancer Detection In Histopathological Images Using An Ensemble Of VGG-19 And Alexnet

Abstract

Breast cancer as one of primary causes of death among women globally, requires accurate early diagnosis to significantly improve patient survival outcomes. Even though histopathology is the most reliable diagnostic method, manual interpretation poses limitations such as being time-consuming, subjective, and highly dependent on the experience of the examiner. With the advancement in deep learning the convolutional neural networks (CNNs) shows significance in automating cancer detection. An ensemble deep learning architecture that integrates VGG-19 and AlexNet has been proposed for breast cancer classification using histopathological images. After performing independent fine-tuning by both models using transfer learning, the models’ prediction scores are merged using a weighted ensemble method where both models make predictions separately, and then their outputs which are probability scores are combined after the individual models have finished prediction, enabling the system to utilize complementary strengths from each architecture. The proposed ensemble system demonstrates improved performance over individual models, achieving an overall accuracy of 98.88% on the histopathological breast cancer image dataset.  Through heterogeneous CNN models the robustness and classification performance has improved. This implies that ensemble deep learning models may aid pathologists in improving the dependability of computer-assisted breast cancer diagnosis.