Steganography-based voice hiding in medical images of COVID-19 patients

A novel image steganography technique in order to hide the ciphered voice data has been suggested in this work. The doctor's voice comments belonging to a coronavirus disease 2019 (COVID-19) patient are hidden in a medical image in order to protect the patient information. The introduced steganography technique is based on chaos theory. Firstly, the voice comments of the doctor are converted to an image and secondly, they are ciphered utilizing the suggested encryption algorithm based on a chaotic system. Then, they are embedded into the cover medical image. A lung angiography dual-energy computed tomography (CT) scan of a COVID-19 patient is used as a cover object. Numerical and security analyses of steganography method have been performed in MATLAB environment. The similarity metrics are calculated for R, G, B components of cover image and stego image as visual quality analysis metrics to examine the performance of the introduced steganography procedure. For a 512 x 512 pixel cover image, SSIM values are obtained as 0.8337, 0.7926, and 0.9273 for R, G, B components, respectively. Moreover, security analyses which are differential attack, histogram, information entropy, correlation of neighboring pixels and the initial condition sensitivity are carried out. The information entropy is calculated as 7.9993 bits utilizing the suggested steganography scheme. The mean value of the ten UACI and NPCR values are obtained as 33.5688% and 99.8069%, respectively. The results of security analysis have revealed that the presented steganography procedure is able to resist statistical attacks and the chaotic system-based steganography scheme shows the characteristics of the sensitive dependence on the initial condition and the secret key. The proposed steganography method which is based on a chaotic system has superior performance in terms of being robust against differential attack and hiding encrypted voice comments of the doctor. Moreover, the introduced algorithm is also resistant against exhaustive, known plaintext, and chosen plaintext attacks.