Integrating Generative AI and Federated Learning for Privacy Preserved Sequence-Based Stomach Adenocarcinoma Detection

Stomach Adenocarcinoma (STAD) significantly impacts global cancer mortality rates. Recent strides in artificial intelligence (AI), machine learning (ML), and deep learning (DL) have primarily harnessed imaging techniques like CT, X-rays, PET, and MRI for cancer detection. Concurrently, the rapidly growing volume of genomic big data presents an untapped reservoir for identifying genetic mutations characteristic of STAD. Our research explores this avenue by examining gene amino acid sequences altered by STAD. We employ physical properties of amino acids, i.e., Electro-Ion Interaction Pseudopotential (EIIP) values and Kidera factors in feature extraction, a novel strategy in STAD diagnostics. To address the issue of class imbalance, we incorporate generative AI to produce additional data samples. Addressing the privacy and security challenges associated with data centralization in healthcare, we propose Fed_ANN11, an artificial neural network (ANN) model developed in a federated environment following an initial deployment as ANN11. Our model demonstrates remarkable accuracy in both simple and federated environments with extracted feature sets, namely, EIIP-based and Kidera factors-based. We found that EIIP-based features eclipse Kidera factors in performance. In a simple setting, the proposed model achieved a testing accuracy of 86% and a training accuracy of 88% for STAD detection using the EIIP-based feature set. In a federated environment, it achieved an accuracy of 0.94% in testing and 0.99% in training for STAD detection using the EIIP-based feature set. Moreover, our proposed model shows significant performance compared to existing state-of-the-art methods. Fed_ANN11 not only excels in diagnostic precision but also upholds stringent big data security and privacy protocols, heralding a paradigm shift in AI’s role in healthcare.