Breast Tumor Detection Across Institutions Using Federated Learning and CNN Models

The existing studies propose a new approach for the detection of breast tumors with the help of multicentric Federated Learning (FL) integrated with Convolutional Neural Networks(CNNs) that can overcome the privacy concerns related to dataset sharing. The research assesses the performance of a federated system relative to that of a centralized system in the identification of pictures of breast tumors. The centralized model achieved an accuracy of 89 percent. 4%, precision of 88. 7%, recall of 89. 1%, F1-score of 88. 0. 96, respectively, on the ADE test set have also been achieved. 91. The federated model has an accuracy of 88 percent, as suggested in the guidelines and various literature presented online. 1%, precision of 87. 5%, recall of 88. 0%, F1-score of 87., it achieved an accuracy of 7% and an AUC of 0. 90. However, there is a slight deterioration in the performance measure where accuracy decreases by 1. 3%, precision decreases by 1. 2%, recall decreases by 1. 1%, and F1-score decreases by 1. 2%. The federated approach is helpful since it maintains patient’s privacy across institutions. AUC difference of 0. Equation ‘ 01 tends to show that cell RE recognition capability differs with tumor types and only slightly varies in methods’ capability to differentiate the classifications. The study shows that in comparison with the single institution model, FL does not violate patients’ privacy yet remains effective in tumor detection across institutions. Consistent with previous research, this work highlights federated learning as a solid solution that balances privacy and model performance as compared to centralized approaches.