Deep Learning-based Detection Methods for Covert Communications in E-Health Transmissions

Despite the numerous advantages offered by e-health applications, healthcare data of patients are at huge risk, being targeted by several different types of digital menaces and attacks. Covert communications (or channels) are dramatic threats in e-health transmissions. A covert communication in fact establishes a secret path hidden within legitimate network traffic, and can be used for leaking patients' medical information, identity, personal data, and medical history. This paper focuses on covert timing channels, where the covert data are transmitted by modulating the inter-packets delays of overt transmitted data, proposing a deep learning-based method for the effective detection of such a dangerous threat. More in details, we first collect all the inter-arrival packet delays affected by a timing covert communication, and convert them into colored images (namely, into spectrograms). Then, we exploit image-based deep learning framework to implement an effective detection of such hidden communications. The results, obtained from simulated covert data traffic also in comparison with existing deep-learning methods, demonstrate the efficiency of the proposed method to detect and prevent this form of malicious e-health information transfer.