Quasi-Cross Bilateral Dual Domain Fourier Transforms Capsule Network with Sea Horse Optimization for Breast Cancer Identification

Abstract

Mammography images are necessary for early diagnosis and prompt treatment and better patient outcomes. Despite the development of the deep learning algorithms, there still are barriers to accurately and consistently identify breast cancer. This paper thus presents a new framework which is the Quasi-Cross Bilateral Dual Domain Fourier Transforms Capsule Network with Sea Horse Optimization (QCB-DDFT-CNet-SHO) to address these problems. The proposed approach is tested on the DDSM (Digital Database of Screening Mammography) data set and is based on some pre-processing done using Quasi-Cross Bilateral Filtering (QCBF) to reduce noise and artefacts in the image. Precise segmentation is performed on malignant patches using Dual Domain Attention with GAN (DDA-GAN). Short Period Fourier Transform along with Continuous Wavelet Transform (SPFT-CWT) is used to extract reliable feature and the parameters of the searched polygonal waterfall are represented through Pixel and Window mechanisms. The Elastic Decision Gate Graph Capsule Network (EDGGCNet) classifies by performing classification. Finally, robustness and efficiency in model performance is improved using the Sea Horse Optimization (SHO) technique. The proposed QCB-DDFT-CNet-SHO framework performs extremely well on the DDSM dataset with a 99.8% recall rate and a 99.9% accuracy rate. These findings demonstrate how well it works to improve the diagnosis of breast cancer from mammography pictures, outperforming current techniques and having a great deal of clinical application potential.