Deep Learning vs Traditional Machine Learning: A Comparative Study of CNN and SVM in Breast Cancer Detection

This article compares the performance of Convolutional Neural Networks (CNN) and Support Vector Machines (SVM) in breast cancer detection through empirical research, evaluating the effectiveness of both methods using multi-dimensional metrics such as accuracy, precision, recall, F1-score, and ROC curves, providing references for technology selection in clinical decision support systems.

Breast cancer is one of the most common malignant tumors among women worldwide, and early diagnosis is crucial for improving survival rates. Traditional diagnosis relies on manual interpretation of mammograms, which has problems such as strong subjectivity and the risk of missed diagnosis. AI technology provides new possibilities for improving diagnostic efficiency and accuracy. As a representative of deep learning, CNN performs excellently in image recognition, while SVM has advantages in small sample and high-dimensional data processing. This study compares the characteristics and applicable scenarios of the two in practical applications.

Core Research Objectives: 1. Implement CNN and SVM models; 2. Compare the two technical routes; 3. Multi-dimensional performance evaluation; 4. Clinical applicability analysis.

Data Preprocessing Process: Image normalization and size adjustment, noise reduction and enhancement, Region of Interest (ROI) extraction, data augmentation (rotating, flipping, etc. to expand the dataset).

CNN Architecture: Includes convolutional layers (automatically extract hierarchical features), pooling layers (reduce dimensionality and enhance translation invariance), fully connected layers (map to classification space). Advantages: automatic feature extraction, end-to-end training.

SVM Architecture: Maximizes sample spacing through optimal hyperplanes, relies on manually extracted features (such as texture, morphological features). Advantages: robust small sample learning, high computational efficiency.

Evaluation Metrics: Accuracy, precision, recall, F1-score, ROC curve and AUC.

Experimental Results:

Model	Accuracy	Precision	Recall	F1-score
CNN	0.9962	0.4413	0.5851	0.5031
SVM	0.9962	0.6275	0.6285	0.6390

Analysis: Both models have the same accuracy; SVM is better in precision, recall, and F1-score; CNN performs better in ROC-AUC.

CNN Advantages: Automatic feature learning, end-to-end training, capturing complex patterns; Limitations: large data requirements, intensive computational resources, poor interpretability.

SVM Advantages: Robust small sample learning, high computational efficiency, better interpretability; Limitations: relies on feature engineering, limited feature extraction capability.

Clinical Considerations: Factors such as data availability, computational resources, interpretability requirements, and misdiagnosis costs affect technology selection.

Current Limitations: Small dataset size, class imbalance, single modality, lack of external validation.

Future Directions: Larger-scale datasets, advanced CNN architectures (ResNet, etc.), multi-modal fusion (X-ray + ultrasound + MRI), improving clinical applicability (interpretability, clinical trials).

This study provides empirical data for AI technology selection in breast cancer detection. Under specific conditions, SVM is still competitive, and technology selection should be based on problem characteristics and constraints. Hybrid strategies (such as CNN feature extraction + SVM classification) may be better. AI has broad prospects in breast cancer diagnosis, and the ultimate goal is to serve patients and improve the level of precision medicine.