Enhancing breast cancer detection: a novel deep learning approach using hybrid convolutional neural networks and residual number systems

Document Type : Original Article

Authors

1 1.Department of Computer Engineering, Kerman Branch, Islamic Azad University, Kerman, Iran. B.rezai@kmu.ac.ir

2 Department of Computer Engineering, Kerman Branch, Islamic Azad University, Kerman, Iran

3 Department of Computer Engineering, Graduate University of Advanced Technology, Kerman, Iran

10.30476/jhmi.2024.103601.1232

Abstract

Background: The detection of breast cancer is vital for intervention and treatment as soon as possible. This study attempts to use a new hybrid deep learning approach that is a combination of Convolutional Neural Networks (CNNs) and Residual Number Systems (RNS) to more precisely detect cancer of the breasts.
Methods: INBREAST and MINI-DDSM datasets were employed to evaluate the hybrid model. Precision, recall, F1-score, and accuracy of these were employed to determine effects of the model compared to existing methods.
Results: The hybrid model was found to be 99% accurate in training using INBREAST dataset, and 91.5% in validation using INBREAST dataset, while MINI-DDSM dataset was found to be 98% in training and 95.02% in validation in terms of accuracy. The model was superior in MINI-DDSM dataset compared to existing models such as ZFNET and ResNet18 in precision, recall, and accuracy metrics. INBREAST dataset was hard to manage due to its nature of complexity, hence it was found to produce low precision and recall despite having high overall precision in performance.
Conclusion: This study highlights the potential of the proposed hybrid deep learning approach for breast cancer detection, especially in simpler datasets. Future research should focus on techniques such as data augmentation, transfer learning, and ensemble methods to improve robustness and generalizability across diverse imaging scenarios. The findings contribute to the integration of deep learning in medical diagnostics, aiming for more accurate and efficient breast cancer detection systems.

Keywords

Main Subjects