A Privacy-Preserving Federated Learning Approach for Primary Colorectal Malignancy Detection, Tumor Multiplicity Estimation and Regional Occurrence Mapping

Abstract

It is among the top causes of deaths in the world-Colorectal cancer. As soon as we detect it, we will survive. However, there are a lot of barriers in the medical field around looking at real patient test data. To address this, we propose a privacy preserving solution e.g federated learning. We have been dealing with structured data from SEER and that has aided them in determining three common tasks. (1) Identification of primary malignancies (2) Assessing tumor multiplicity (3) Tissue-wide associations. We have achieved this by using a neural network model. Where FedAvg combines the model which trains it locally and you will get gradient values/parameters of the model after performing an average on it. With the appropriate preprocessing nodes, binary or multiclass predictions can be obtained sequentially for certain tasks. This model achieves a significant and correct improvement in performance against the original baseline whilst maintaining high security levels, as shown by the results M. Enable colon cancer detection without revealing the original data. Multiple original cancer research underlie Al integration and privacy in health.