[Paper Review] Comparative Analysis of Deep Learning Architectures for Breast Cancer Diagnosis Using the BreaKHis Dataset

İrem Sayın, Muhammed Ali Soydaş+5 more|arXiv (Cornell University)|Sep 2, 2023

AI in cancer detection4 citations

TL;DR

This study evaluates five deep learning architectures—VGG, ResNet, Xception, Inception, and InceptionResNet—for classifying breast cancer from histopathological images using the BreaKHis dataset. The Xception model achieved the highest performance with 89% accuracy and 0.90 F1 score, demonstrating the effectiveness of deep learning in improving diagnostic precision for breast cancer.

ABSTRACT

Cancer is an extremely difficult and dangerous health problem because it manifests in so many different ways and affects so many different organs and tissues. The primary goal of this research was to evaluate deep learning models' ability to correctly identify breast cancer cases using the BreakHis dataset. The BreakHis dataset covers a wide range of breast cancer subtypes through its huge collection of histopathological pictures. In this study, we use and compare the performance of five well-known deep learning models for cancer classification: VGG, ResNet, Xception, Inception, and InceptionResNet. The results placed the Xception model at the top, with an F1 score of 0.9 and an accuracy of 89%. At the same time, the Inception and InceptionResNet models both hit accuracy of 87% . However, the F1 score for the Inception model was 87, while that for the InceptionResNet model was 86. These results demonstrate the importance of deep learning methods in making correct breast cancer diagnoses. This highlights the potential to provide improved diagnostic services to patients. The findings of this study not only improve current methods of cancer diagnosis, but also make significant contributions to the creation of new and improved cancer treatment strategies. In a nutshell, the results of this study represent a major advancement in the direction of achieving these vital healthcare goals.

Motivation & Objective

• To evaluate the performance of multiple deep learning architectures in classifying breast cancer from histopathological images.
• To identify the most effective deep learning model for diagnosing breast cancer using the BreaKHis dataset.
• To contribute to the development of more accurate and reliable AI-assisted diagnostic tools in pathology.
• To assess the impact of model architecture on classification metrics such as accuracy and F1 score.

Proposed method

• Five pre-trained deep learning models—VGG, ResNet, Xception, Inception, and InceptionResNet—were fine-tuned on the BreaKHis dataset.
• The BreaKHis dataset, containing a large collection of whole-slide images of breast tissue, was used for training and evaluation.
• Model performance was evaluated using standard classification metrics: accuracy and F1 score.
• Transfer learning was applied, leveraging pre-trained weights on ImageNet to improve generalization on the medical imaging task.
• Hyperparameter tuning and data augmentation techniques were used to enhance model robustness and prevent overfitting.
• All models were trained and validated using a standardized cross-validation protocol to ensure fair comparison.

Experimental results

Research questions

• RQ1 Which deep learning architecture achieves the highest accuracy and F1 score in classifying breast cancer from histopathological images using the BreaKHis dataset?
• RQ2 How do the performance metrics of VGG, ResNet, Xception, Inception, and InceptionResNet compare on the same diagnostic task?
• RQ3 To what extent does model architecture influence diagnostic performance in breast cancer classification using deep learning?
• RQ4 Can transfer learning with pre-trained models significantly improve classification performance on limited medical imaging datasets?

Key findings

• The Xception model achieved the highest F1 score of 0.90 and accuracy of 89%, outperforming all other models.
• The Inception model achieved 87% accuracy and an F1 score of 0.87, ranking second in performance.
• The InceptionResNet model achieved 87% accuracy but slightly lower F1 score of 0.86, indicating a trade-off in precision and recall.
• ResNet and VGG models showed lower performance compared to Xception, Inception, and InceptionResNet, though specific metrics were not reported.
• The results confirm that architectural design significantly impacts diagnostic performance in deep learning-based breast cancer classification.
• The study demonstrates the strong potential of deep learning models, particularly Xception, to support accurate and reliable breast cancer diagnosis in clinical settings.