Privacy-Preserving Computation Using CKKS Homomorphic Encryption in Medical AI Systems
Аннотация
Medical AI systems consume an extensive amount of sensitive patient data to provide diagnostic and treatment support while facing substantial security and privacy risks. The CKKS scheme within Homomorphic encryption provides a viable solution through which encrypted data can undergo processing operations without jeopardizing confidentiality about the underlying data contents. Data anonymisation and secure multiparty computation methods fail to fulfil their purpose because they allow re-identifications and produce high communication expenses. This paper develops a Privacy-Preserving Inference Using CKKS Homomorphic Encryption framework that enables secure AI model processing of encrypted medical images without decryption. Patient data gets encrypted through CKKS before it is sent to a remote server for encrypted inference with a modified neural network, then the server returns the encrypted result for user-side decryption. The system establishes confidentiality protection through secure cloud-based medical diagnostic operations. The method maintains accurate diagnostic capabilities while managing an appropriate computational increase, which verifies its operational potential for secure healthcare AI service delivery remotely.