Research on Image Classification Method for Mammography Based on KAN

Breast cancer is one of the most common malignant tumors among women worldwide. Early diagnosis and timely treatment are crucial for significantly reducing patient mortality. Therefore, early detection and accurate diagnosis of the disease are particularly important. Mammography is a commonly used technique for detecting breast cancer. By analyzing the images of masses, it is possible to determine whether they are benign or malignant. However, traditional breast image classification methods suffer from complex feature extraction, low classification efficiency, and sensitivity to adversarial samples, which limit their effectiveness in clinical applications. This paper introduces the application of a hybrid model combining Convolutional Neural Networks (CNNs) and Kolmogorov-Arnold Networks (KANs) for the classification of mammographic images. The model leverages the powerful feature extraction capabilities of CNNs and integrates the approximation and interpretability of KAN spline functions into the convolutional layers of CNNs. This approach efficiently fuses different types of features and reduces the number of parameters by replacing the linear weight matrix with learnable spline curves, thereby improving the accuracy of image classification. The model was validated on the standard DDSM dataset. According to the experimental results, the hybrid CNN-KAN model achieved an accuracy rate of over 98 % for mammographic image classification while reducing the number of model parameters by half. This demonstrates that the hybrid model has significant advantages in practical clinical applications and plays an important role in the early diagnosis of breast cancer