Privacy Aware Deep Learning Model for Multi Class Classification in Big Data

Big data and deep learning (DL) are evolving technologies applied extensively in the medical field. Artificial intelligence (AI) technologies have simplified operations such as sharing and retrieving large medical images and swiftly providing disease results in no time. Sharing medical images that are highly sensitive information for every user might give away vulnerable information to the opponents. Privacy is a major concern between a user and a database. In this paper, we propose an Advanced Convolutional Neural Network (ACNN) for selecting the features from large-scale medical data, integrated with privacy-preserved cosine similarity (PPCS), to find similarities between users and all databank images securely. A comparison is made between an ACNN and a PPCS-ACNN based multi-class classification model for diagnosing various lung diseases from Computed Tomography (CT) images. The analysis focuses on the trade-offs between data privacy, diagnostic accuracy, and the efficiency of classification.