An Experiment to Develop an Enhanced Medical Image Security by using Deep Learning Assisted Crypto Policy
G. Ramkumar, V. Sujatha, B. William Dharma Raja, M. Tamilselvi, A. Balaji, ATA Kishore Kumar
Abstract
Ensuring a high level of safety and security is imperative when transmitting medical images via open access channels. Medical images hold immense importance in various applications, particularly in real-time contexts like telemedicine. Due to the unique characteristics of medical imaging data, determining the appropriate techniques for safeguarding confidential images against unauthorized access becomes challenging. Existing encryption methods primarily cater to textual data, leaving a gap in effectively securing multimedia data such as images. Using DNN leaves these systems open to attack from adversarial samples, which are pictures with slight changes that can trick the model into producing inaccurate predictions. Utilizing the MNIST database, this research demonstrates that Deep Learning algorithms may be used to encrypted data without compromising accuracy. This finding holds significance as it offers a potential solution for preserving data security while benefiting from advanced machine learning capabilities. In addition, the research goes farther by training a model using encrypted information to categories X-ray coronary angiography images, making it a more difficult scenario. The suggested DNN model's remarkable accuracy score of 0.983 highlights its ability to make predictions with an exceptional level of correctness.