FDSRN: Fractal Deep Spiking Neural Network for Breast Cancer Detection Using Mammogram Images

Abstract

Breast cancer identification stands as a pivotal field in medical research and technology, highly focused on early identification and diagnosis of abnormalities within breast tissue. Early detection significantly enhances treatment outcomes and improves patient survival rates. By recognizing the critical significance of breast cancer identification, an innovative model known as Fractal Deep Spiking Neural Network (FDSRN) has been introduced. In this model, the mammogram images are initially chosen as input for the pre-processed phase. The preprocessing of mammogram images is accomplished by utilizing a wiener filter. After pre-processing, cancer region segmentation is implemented by utilizing U-NeXt. Then, image augmentation, such as random erasing, shifting and rotation is performed. After accomplishing image augmentation, feature extraction is applied to extract features like Gradient Binary Patterns (GBP), Binary Robust Independent Elementary Features (BRIEF) and Gray level co-occurrence matrix (GLCM). Lastly, breast cancer identification is conducted by utilizing developed FDSRN, which is the incorporation of FractalNet and Deep Spiking Neural Network (DSNN). The FDSRN employed for breast cancer detection has shown outstanding performance, achieving an accuracy of 90.205%, sensitivity of 90.710%, and specificity of 90.943%. The MIAS and DDSM datasets are used for the experimentation.