Enhancing Breast Cancer Classification in Histopathological Images through Federated Learning Framework

In recent decades, the mortality rate of breast cancer in females is rapidly increasing because of unawareness and failed to detect in earlier stages. In existing, several studies are attempted to develop a robust mechanism for detecting breast cancers from the given input samples. However, they are not as much effective because of several limitations and the secured sharing of sensitive medical images is still a challenging problem faced by medical sector. Thus, the proposed study aims to introduce an automated disease diagnosis system using federated learning and deep learning which automates and speed up the process efficiently. The five crucial steps that involved in the proposed study are image acquisition, encryption, optimal key generation, secured data storing and disease classification. Initially, the required input medical images are gathered in the image acquisition stage. Then, to afford more confidentiality, the gathered medical samples are encrypted through an Extended ElGamal Image Encryption (E-EIE) method. Here, the efficiency of encryption process is enhanced by generating the suitable keys in optimal manner with the help of Improved Sand Cat Swarm Optimization (I-SCSO) algorithm. Next, the security of encrypted images are improvised by utilizing federated learning flower (FLF) framework for storage purpose. This framework has the ability to transmit the medical images with higher security. Finally, the stored images are decrypted and performs disease classification by using convolutional capsule twin attention tuna optimal network (C
₂
T
²
Net) model. The available loss in the proposed classifier is reduced by fine-tuning the parameters using chaotic tuna swarm optimization (CTSO) algorithm. For simulation analysis, the proposed study used Python software and the experimental analysis is carried out by using BreakHis Database. The simulation results shows that the proposed study obtained higher performance in terms of accuracy (95.68%), recall (95.6%), precision (95.66%), F-measure (95.63%), specificity (95.6%) and kappa coefficient (95.26%).