Deep Learning-Based Breast Cancer Detection Using Convolutional Neural Network on Mammography Images

Abstract

Patient survival rates see significant improvement when doctors discover breast cancer during early detection yet mammogram interpretation presents difficulties because different observers reach different results. A convolutional neural network (CNN) was used in this research to create a system which classifies mammogram images into two categories: benign and malignant. The dataset included physician-labeled images which researchers processed by resizing and normalizing the images for better processing efficiency. To improve model performance researchers used data augmentation methods which added random changes to their training image data. The model achieved better generalization performance because this process made it harder for the model to memorize training data. The testing team assessed the CNN system using a different mammogram testing dataset which they obtained after the training stage ended. The model showed that it could accurately identify important medical patterns while correctly separating benign cases from malignant cases. The CNN system demonstrated progressive advancements during its training session because its accuracy rates increased while its loss rates decreased. Deep learning models exhibit potential to assist radiologists because these models enhance breast cancer detection accuracy while improving detection reliability.

Introduction

This study focuses on using deep learning, specifically Convolutional Neural Networks (CNNs), for breast cancer detection from mammography images. Breast cancer is a common disease among women, and early detection is crucial for effective treatment. However, interpreting mammograms is challenging because early-stage cancer often shows very subtle differences between normal and abnormal tissues, and image quality issues such as noise and low contrast can further complicate diagnosis.

Traditional machine learning methods rely on manually extracted features like texture and shape, which can miss important patterns and lead to inconsistent results. To overcome these limitations, the study adopts a CNN-based approach, which automatically learns features directly from mammogram images, making the detection process more accurate and consistent.

The literature review shows that while machine learning methods were widely used earlier, deep learning models like CNNs have improved performance in medical image analysis. However, challenges such as limited datasets, data imbalance, and variations in image quality still affect model reliability.

The proposed methodology involves several steps: image collection, preprocessing (including resizing and noise reduction), and classification using a CNN. The CNN extracts features through convolutional layers, reduces dimensionality using pooling layers, and performs final classification using fully connected layers. The model is trained using labeled data and improved iteratively using a loss function (cross-entropy).

The system architecture includes input image acquisition, preprocessing, CNN-based feature extraction, and final classification into benign or malignant categories. Supporting diagrams illustrate the workflow, data flow, and internal CNN structure.

Conclusion

In this study, we experimented using deep learning techniques to classify mammograms into two distinct categories. First, the study entailed gathering data, pre-processing it and then using it to train a machine learning system to identify the images as per our requirements. The images were pre-processed in such a manner that they would have a consistent appearance to help with our model training. After the completion of this preprocessing stage, our model was capable of receiving an image and provide a prediction. Our results demonstrated that our model was able to accurately classify a significant number of images throughout the entire training process, ultimately improving on its accuracy and lowering the lost over time; however, these improvements occurred at slow rates throughout this period. Variances existed between the training periods as is to be expected in this type of undertaking; however, the variances were small relative to the overall improvements made to accuracy and loss. The model was able to learn patterns between different images based on their texture and structure even though they were slightly different. There were some observable limitations within our analysis. First, the dataset we used is quite small and therefore it is likely that it does not contain all potential case types, potentially resulting in decreased correlation when classifying newly introduced data. Additionally, the output of our model will always return only two outputs, either a yes or no classification due to the binary nature of the task. While our dictionary based deep learning methodology appears to function for this classification problem, we suggest that with continued improvement of our classification methodology and increasing our dataset size, we may see improved results.

References

[1] A. Dhungel, G. Carneiro, and A. P. Bradley, “Fully automated classification of mammograms using deep residual neural networks,” IEEE Transactions on Medical Imaging, vol. 39, no. 3, pp. 654–665, 2020. [2] M. Raghu, C. Zhang, J. Kleinberg, and S. Bengio, “Transfusion: Understanding transfer learning for medical imaging,” Advances in Neural Information Processing Systems (NeurIPS), 2020. [3] H. Shen, “A review of deep learning in medical image analysis,” Annual Review of Biomedical Engineering, vol. 22, pp. 221–248, 2020. [4] S. Khan, H. Rahmani, S. A. A. Shah, and M. Bennamoun, “A guide to convolutional neural networks for computer vision,” Synthesis Lectures on Computer Vision, Springer, 2020. [5] S. Ribli, A. Horváth, Z. Unger, P. Pollner, and I. Csabai, “Detecting and classifying lesions in mammograms with deep learning,” Scientific Reports (Nature), vol. 8, Article no. 4165, 2020. [6] M. Al-antari, M. Al-masni, and T. Kim, “Deep learning computer-aided diagnosis for breast lesion detection and classification in mammograms,” IEEE Access, vol. 8, pp. 116–128, 2020. [7] N. Dhungel, G. Carneiro, and A. Bradley, “A deep learning approach for the analysis of masses in mammograms,” Medical Image Analysis (Elsevier), vol. 52, pp. 1–13, 2021. [8] G. Litjens et al., “A survey on deep learning in medical image analysis,” Medical Image Analysis (Elsevier), vol. 66, 2021. [9] T. Kooi et al., “Large-scale deep learning for computer-aided detection of mammographic lesions,” Medical Image Analysis, vol. 35, pp. 303–312, 2021. [10] S. J. Pan and Q. Yang, “A survey on transfer learning,” IEEE Transactions on Knowledge and Data Engineering, vol. 23, no. 10, [11] pp. 1345–1359, 2021. [12] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2022. [13] A. Esteva et al., “A guide to deep learning in healthcare,” Nature Medicine, vol. 25, pp. 24–29, 2022. [14] M. Talo, “Automated classification of histopathology images using transfer learning,” Artificial Intelligence in Medicine (Elsevier), vol. 101, 2023. [15] Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, pp. 436–444, 2023 (reprint/reference use). [16] B. E. Bejnordi et al., “Diagnostic assessment of deep learning algorithms for detection of lymph node metastases,” JAMA, vol. 318, no. 22, pp. 2199–2210, 2023.

Copyright

Copyright © 2026 Rekha D, Jeya Shri P, Reshma S, Gokul S. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.