Prediction of Breast Cancer from Histopathological Images using Fusion of Deep Learning Models

Millions of women and men worldwide face breast cancer (BCr), an increasing disease in modern society. Although standard cancer screening methods can be effective, they are often costly and time-consuming tedious tasks. Traditional BCr detection methods are limited by tumor heterogeneity, pattern complexity, data types, dataset overfitting, and feature engineering, making a single algorithm insufficient for accurate diagnosis. Creating a hybrid model aims to overcome standard model restrictions. Our goal is to improve accuracy and reduce calculation time compared to current models. Deep learning enables complex models to accurately characterize BCr in BreaKHis Datasets, providing an effective early detection tool. CNN and LSTM algorithms use BreaKHis datasets to discover and classify breast cancer by analyzing distinguishing features. We compared the proposed networks to CNN, Resnet101, and InceptionResNetV2 with LSTM. All models undergo testing using Adam, utilizing regression characteristics to detect breast cancer in BreaKHis datasets. The hybrid Resnet101 with LSTM architecture and Adam optimizer achieved 0.969 accuracy at 40x magnification. Results indicate that the hybrid Resnet101 with LSTM model and Adam optimization may accurately classify BCr, potentially improving patient outcomes and early detection.