Reliable Patient Data Securing for Medical Image Segmentation with Different CNN Architectures

Image segmentation is an essential task in the field of medical image analysis such as the segmentation of carcinogenic foci. However, since these images contain the patient's confidential information, there is a growing concern for protecting them from unauthorized access. With the widespread use of digital medical imaging systems, medical images are now stored in a centralized information system that is accessible to multiple users. In order to prevent unauthorized access and hacking attempts, this paper proposes a segmentation technique of carcinogenic foci in medical images combined with an encryption algorithm for data relating to this segmentation operation. The suggested method is divided into two major steps. For the segmentation step, a comparative study is conducted between some architectures of convolutional neural networks such as Unet and AlexNet. In fact, the architecture, which provides better segmentation results based on several criteria, is adopted for this work. A suitable algorithm is used for the encryption step to ensure the confidentiality of patient information such as patient name, calculated cancer area, exact disease area and direction of spread. The proposed technique ensures that the medical images are segmented and encrypted prior to being uploaded to the information system. This would allow only authorized personnel to access the medical images, thereby safeguarding the confidentiality and privacy of patient information. The obtained result on real data is interesting and proves the efficiency of the proposed method.