Ensemble learning approach for detecting breast invasive ductal carcinoma from histopathological images

Invasive ductal carcinoma is a type of breast cancer that is one of the most frequent and aggressive forms of breast malignancy, necessitating accurate and timely diagnosis for effective treatment. Though considered the gold standard, traditional histopathological diagnosis is subject to inter-observer and intra-observer variability, potentially impacting patient outcomes. This study proposed an ensemble learning approach for classifying invasive ductal carcinoma to address these challenges. The proposed method combines the strengths of multiple deep-learning models to enhance diagnostic accuracy and robustness. We employed a diverse set of pre-trained convolutional neural networks, viz, ResNet50, Xception, MobileNetV2, VGG16, and VGG19, each trained on histopathological images of breast histology slides. These five different deep learning models were compared in this work, and the resulting inference results are also shown. Ensemble and a fine-tuning approach to transfer learning were also used to extract the best results. These models were evaluated using evaluation metrics like accuracy to see which one does the job best. The proposed weighted average ensemble algorithm achieved 97.27 % accuracy. Among all models, the ResNet50 model outperforms the other models in identifying invasive ductal carcinoma. Therefore, ResNet50 is the preferred model when accuracy is the top concern for a particular resolution image, and the weighted average ensemble approach enhances the performance of the proposed work. Our results indicate that the proposed ensemble approach decreases variability in diagnoses and advancements in accuracy. This method holds promise for enhancing the precision of breast cancer diagnostics, potentially leading to better patient management and outcomes.