Breast Cancer Classification Using CNN and SVM: A Hybrid Approach

Abstract

Breast cancer remains one of the most prevalent and life-threatening diseases affecting women worldwide. According to the World Health Organization, early detection and accurate diagnosis play a crucial role in reducing mortality rates and improving treatment outcomes. Despite advancements in diagnostic technologies, manual analysis of mammogram images is time-consuming, prone to variability, and requires expert radiological interpretation. As a response to these challenges, this study proposes an innovative and efficient hybrid machine learning framework that combines the deep learning capabilities of Convolutional Neural Networks (CNNs) with the classification strength of Support Vector Machines (SVMs) for breast cancer detection and classification from mammographic images. The model leverages the CBIS-DDSM (Curated Breast Imaging Subset of DDSM) dataset—a well-established benchmark for mammographic image analysis. The images undergo a series of pre-processing steps including greyscale normalization, contrast enhancement, resizing, and noise reduction, all of which aim to ensure consistent quality and effective feature learning. A custom CNN architecture is then designed to extract high-level features from the pre-processed images. This network is optimized for capturing complex patterns such as masses, calcifications, and tissue asymmetries commonly observed in breast cancer cases. Unlike conventional end-to-end CNN classification, this study uses the CNN primarily for deep feature extraction .The extracted features are subsequently passed to an SVM classifier, which constructs a decision boundary to accurately separate benign from malignant cases. This hybrid model addresses several challenges inherent to medical image analysis: it mitigates the risks of over fitting associated with deep learning models trained on limited data and improves classification performance on imbalanced datasets through the SVM’s generalization capability. The proposed hybrid CNN-SVM model achieves a classification accuracy of 91.7%, with competitive precision, recall, and F1-scores, highlighting its potential effectiveness in real-world clinical scenarios. This study’s contributions are multifold: the development of a novel hybrid classification framework, the successful application of deep learning techniques for mammographic image analysis, and the demonstration of improved diagnostic accuracy through AI-driven methods. The research underscores the importance of interdisciplinary approaches combining medical imaging, artificial intelligence, and statistical learning for advancing cancer diagnostics. In future work, the integration of transfer learning, explainable AI, and real-time decision support systems could further enhance the diagnostic reliability and acceptance of such tools in clinical environments. The findings of this study pave the way for future advancements in computer-aided diagnosis systems and support the global effort to combat breast cancer through technology.

Introduction

1. Introduction

• Breast cancer is one of the most common cancers among women, with over 2.3 million new cases diagnosed annually.

• Early detection significantly improves outcomes; mammography is the most widely used imaging technique.

• However, mammogram interpretation is complex and prone to variability, especially in dense breast tissue.

• To improve accuracy, Computer-Aided Diagnostic (CAD) systems using AI and ML are being developed.

2. Challenges in Breast Cancer Detection

• Manual diagnosis is time-consuming and error-prone.

• Breast lesion characteristics vary greatly, making visual inspection difficult.

• Medical datasets are often imbalanced (more benign than malignant), leading to biased ML performance.

• Overfitting is a concern with deep learning models due to limited annotated medical data.

3. Literature Survey

• Recent research has applied hybrid AI models, combining CNNs with various classifiers (e.g., SVM, CRF, GCN) for improved detection accuracy.

• Examples:

  • FrCN achieved ~99.2% F1-score on the INbreast dataset.

  • BDR-CNN-GCN used graph structures for spatial awareness, reaching 96.1% accuracy.

  • Modified YOLOv5 outperformed prior models in object detection.

  • Transfer Learning and wavelet-based hybrid approaches showed high accuracy on DDSM and MIAS datasets.

  • Comparative studies showed ResNet50, InceptionV3, and other deep architectures are highly effective.

• Despite these advances, issues remain: model complexity, long training times, and limited data availability.

4. Imaging Modalities Overview

• Mammography: Gold standard, especially effective in detecting micro-calcifications.

• Ultrasound: Helps distinguish cysts vs. tumors.

• MRI: High sensitivity but costly.

• Histopathology: Gold standard for diagnosis.

• Thermography: Emerging, non-invasive method needing more validation.

5. Proposed Methodology

Hybrid CNN-SVM Architecture:

• Dataset: CBIS-DDSM (subset of DDSM) with annotated benign and malignant mammograms.

• Preprocessing:

  • Noise reduction (filters), resizing, normalization.

  • ROI cropping to focus on tumor-relevant areas.

• CNN for Feature Extraction:

  • 11-layer custom CNN with increasing filter sizes (64 to 512), 3×3 and 2×2 convolutions.

  • 4 max-pooling layers (3×3, 4×4) to reduce dimensions while preserving features.

• SVM for Classification:

  • Feature vector from CNN passed to an SVM to classify as benign or malignant.

6. Dataset Details

• Total: 55,885 images

  • 48,596 benign, 7,289 malignant

• Split: 80% training, 20% testing

  • Training: 44,708 (38,877 benign, 5,831 malignant)

  • Testing: 11,177 (9,719 benign, 1,458 malignant)

7. Results & Discussion

• Confusion Matrix:

  • True Negatives (benign correctly identified): 8,951

  • True Positives (malignant correctly identified): 1,300

  • False Positives: 768

  • False Negatives: 158

• Performance Metrics:

  • Accuracy: ~91.7%

  • High sensitivity (recall) for malignant class – vital for early detection.

  • Moderate precision – some benign cases misclassified as malignant.

  • F1-score indicates a good balance between precision and recall.

• Observations:

  • Prioritizing sensitivity helps minimize missed cancer diagnoses.

  • False positives may cause patient anxiety but ensure cautious screening.

  • Potential improvements: better hyperparameter tuning, more balanced training data, and post-processing techniques.

Conclusion

The proposed CNN–SVM hybrid model demonstrated strong classification performance for mammogram images, achieving an overall accuracy of 91.7% with high sensitivity for malignant cases. The confusion matrix analysis confirms the model’s effectiveness in correctly identifying the majority of cancerous and non-cancerous cases, making it suitable for early breast cancer detection. While the low false-negative rate ensures that most malignant cases are detected, the presence of false positives indicates a need for further optimization to improve precision. Nonetheless, the model shows significant potential for integration into computer-aided diagnosis systems, where it can assist radiologists in enhancing diagnostic accuracy and reducing workload.

References

[1] L. Shen, L. R. Margolies, J. H. Rothstein, E. Fluder, R. McBride, and W. Sieh, “Deep learning to improve breast cancer detection on screening mammography,” Scientific Reports, vol. 9, no. 1, p. 12495, 2019. [2] H. M. Frazer, A. K. Qin, H. Pan, and P. Brotchie, “Evaluation of deep learning?based artificial intelligence techniques for breast cancer detection on mammograms: Results from a retrospective study using a BreastScreen Victoria dataset,” Journal of Medical Imaging and Radiation Oncology, vol. 65, no. 5, pp. 529–537, 2021. [3] N. Dhungel, G. Carneiro, and A. P. Bradley, “A deep learning approach for the analysis of masses in mammograms with minimal user intervention,” Medical Image Analysis, vol. 37, pp. 114–128, 2017. [4] A. Sahakyan and H. Sarukhanyan, “Segmentation of the breast region in digital mammograms and detection of masses,” International Journal of Advanced Computer Science and Applications, vol. 3, no. 2, pp. 87–92, 2012. [5] M. Abbas, M. Arshad, and H. Rahman, “Detection of breast cancer using neural networks,” LC International Journal of STEM, vol. 1, no. 3, pp. 75–88, 2020. [6] Y. Ero?lu, M. Yildirim, and A. Cinar, “Convolutional neural networks based classification of breast ultrasonography images by hybrid method with respect to benign, malignant, and normal using mRMR,” Computers in Biology and Medicine, vol. 133, p. 104407, 2021. [7] T. Mahmood, J. Li, Y. Pei, and F. Akhtar, “An automated in-depth feature learning algorithm for breast abnormality prognosis and robust characterization from mammography images using deep transfer learning,” Biology, vol. 10, no. 9, p. 859, 2021. [8] J. Lee and R. M. Nishikawa, “Identifying women with mammographically-occult breast cancer leveraging GAN-simulated mammograms,” IEEE Transactions on Medical Imaging, vol. 41, no. 1, pp. 225–236, 2021. [9] A. Altameem, C. Mahanty, R. C. Poonia, A. K. J. Saudagar, and R. Kumar, “Breast cancer detection in mammography images using deep convolutional neural networks and fuzzy ensemble modeling techniques,” Diagnostics, vol. 12, no. 8, p. 1812, 2022. [10] D. Kaur and Y. Kaur, “Various image segmentation techniques: a review,” International Journal of Computer Science and Mobile Computing, vol. 3, no. 5, pp. 809–814, 2014. [11] Y. Wang, L. Zhang, X. Shu, Y. Feng, Z. Yi, and Q. Lv, “Feature-sensitive deep convolutional neural network for multi-instance breast cancer detection,” IEEE/ACM Transactions on Computational Biology and Bioinformatics, vol. 19, no. 4, pp. 2241–2251, 2021. [12] J. Han, D. Zhang, X. Hu, L. Guo, J. Ren, and F. Wu, “Background prior-based salient object detection via deep reconstruction residual,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 25, no. 8, pp. 1309–1321, 2014. [13] T. M, K. T G, V. K V, N. Siby, S. Devaraj and R. B R, \"Leveraging an Optimized Deep Convolutional Neural Network Architecture for the Automated Diagnosis of Pathological Conditions Through Acoustic Analysis of Human Cough Signatures,\" 2025 3rd International Conference on Smart Systems for applications in Electrical Sciences (ICSSES), Tumakuru, India, 2025, pp. 1-7, doi: 10.1109/ICSSES64899.2025.11009841 [14] Murthy, K. T., and M. N. Eshwarappa. \"A Novel Person Authentication Technique Using Electrocardiogram (ECG).\" J. Electr. Syst 20 (2024): 393-405. [15] Hareesh K. N. (2024). Comparative Analysis of Brain Tumor Classification Using CT, MRI, and Fusion of CT and MRI Images with GLCM Features. International Journal of Intelligent Systems and Applications in Engineering, 12(21s), 3844, https://ijisae.org/index.php/IJISAE/article/view/6153 [16] T G, K. ., & M N, E. . (2023). ECG Biometric for Human Authentication using Hybrid Method. International Journal on Recent and Innovation Trends in Computing and Communication, 11(7s), 292–299. https://doi.org/10.17762/ijritcc.v11i7s.7002 [17] T. G. Keshavamurthy and M. N. Eshwarappa, \"Review paper on denoising of ECG signal,\" 2017 Second International Conference on Electrical, Computer and Communication Technologies (ICECCT), Coimbatore, India, 2017, pp. 1-4, doi: 10.1109/ICECCT.2017.8117941. [18] T. G. Keshavamurthy and M. N. Eshwarappa, \"ECG signal de-noising using complementary ensemble empirical mode decomposition and Kalman smoother,\" 2017 3rd International Conference on Applied and Theoretical Computing and Communication Technology (iCATccT), Tumkur, India, 2017, pp. 120-125, doi: 10.1109/ICATCCT.2017.8389118. [19] Keshavamurthy, T. G., & Eshwarappa, M. N, \" ECG signal de-noising based on adaptivefilters, \"International Journal of Innovative Technology and Exploring Engineering( IJITEE), vol.9(1), pp. 5473–5483, November 2019.http://doi.org/10.35940/ijitee.K1601.119119

Copyright

Copyright © 2025 Swathi K S, Dr. Nirmala G, Dr. Sujatha S R, Dr. Raviram V. 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.