AbstractData is the key element that runs the modern society. Large amounts of data are being released day by day as a result of many activities. The digital data is transferred through the Internet which may be vulnerable to attacks while transmitting. Especially, the medical data is observed to be of at most importance. Thus, the security of the data should be preserved while communicating through the Internet. Many techniques have been evolved for securing the data, such as Encryption which makes the data not understandable by others and techniques such as Steganography, which hides the data inside other file or document. Traditional methods of Steganography‐like LSB substitution are subjected to risks of manipulation of carrier image and poor embedding capabilities. In light of this, the authors introduce a Convolutional Neural Network‐based image steganography method that operates on DICOM images which conceals the sensitive medical data in the carrier file, thereby increasing the privacy of the sensitive data. The data considered here is brain MRI scan images. The Prep Net, Hiding Net and Reveal Net networks, each with distinct number of convolution layers are used to build the model. The sensitive data is firstly segmented to detect the Region of Interest (ROI) and then the combination of sensitive data and the carrier image are fed to the Prep Net to manipulate sensitive images in such a form that it can easily fit inside the carrier image. The Hiding Net takes the result of Prep Net and embeds the sensitive image inside the carrier image to form a stego image. The Reveal Net takes the result of Hiding Net image and then extracts the sensitive image with minimal loss. The dataset contains medical images of DICOM format taken from open source platform Kaggle that contains 3500 images of 2 classes. The results obtained show that the proposed model achieves 20 percent better output compared to traditional LSB methods. As a result, the potential for hiding the data will be expanded, as well as its security.
Read full abstract