Medical information security in the era of artificial intelligence
Abstract
In recent years, biometric technologies, such as iris, facial, and finger vein recognition, have reached consumers and are being increasingly applied. However, it remains unknown whether these highly specific biometric technologies are as safe as declared by their manufacturers. As three-dimensional (3D) reconstruction based on medical imaging and 3D printing are being developed, these biometric technologies may face severe challenges.
Introduction
Biometrics is an identification technology that uses human biological traits and comprises fingerprint, face, iris, and finger vein recognition [1]. Most people believe that biometric technology, which has a high degree of uniqueness,is very advanced and secure. In fact, if some medical imaging materials typically used by physicians were made available, many biometric systems would be easy to crack.
The insecurity of fingerprinting identification is well known.Models for copying fingerprints to fake time attendance systems in universities and companies have even evolved into a small-scale industry [2]. Hence, iPhone X began using Face ID to replace Touch ID. Given the iPhone’s tremendous consumer influence, facial recognition is likely to become a popular biometric technology in the coming years. However, whether facial recognition is as safe as we believe it is remains unknown. Facial three-dimensional (3D) reconstruction at various levels of precision can be accomplished via several methods, such as traditional computed tomography (CT), magnetic resonance imaging (MRI), cone-beam computed tomography (CBCT, used in dental scanning), and medical 3D scanner[3-5]. We are interested in learning whether reconstructing faces using these techniques can be 3D printed and used to crack facial recognition systems.
Hypothesis
Will medical imaging development threaten biometric technology?Accurately reconstructing the human face can be accomplished by using multiview imaging or medical 3D scanner and subsequently be printed on high-precision 3D printer. These 3D models, with some modifications, may pose serious threats to current facial recognition technologies. This is especially true for low-end facial recognition systems, which can be easily cracked by regular medical imaging data (MRI or CT head scanning data).
Currently, many medical fields, such as stomatology, are becoming increasingly digitalized, such that massive imaging data are stored digitally and can be used for 3D reconstruction. However, if the servers accommodating these medical data are hacked or leaked, the resulting security problems will present great challenges.
Evaluating the Hypothesis
Three-dimensional reconstruction technology has long been used in stomatology to cater to various orthodontic, repair, and cosmetic surgery needs [6,7]. In recent years, CBCT (used for dental scanning of hard tissue) and various medical 3D scanners (used for intraoral or extraoral scanning) have enabled remarkably precise 3D imaging. Even multiview imaging, which has long been used in maxillofacial surgery and has a low hardware requirement, can be assisted by software to conveniently generate precise 3D models.This is also due to advances in camera imaging quality and image reconstruction algorithms.
Facial 3D reconstruction is mainly used in stomatological operations, such as oral and maxillofacial plastic surgery. After being registered to CT and MRI data, a 3D image can be established to simulate surgery and assess postoperative outcomes[8]. In addition, 3D reconstruction can be subjected to many analyses,including finite elemental analysis[9]. In this study, we examined multiview imaging, which is the simplest 3D imaging technology approach and only requires a camera or cell phone to take photos or videos from different angles.The images or videos can then be used to rebuild a facial model using software such as Agisoft PhotoScan (a demo version is available for download on the company’s website) or Autodesk ReCap Photo. Many large medical institutions now have high-precision 3D-imaging facilities, professional reconstruction software, and 3D printers, which can be used to crack facial recognition systems.
Discussion
To date, there have been reports of 3D printing being used to crack facial recognition systems. Even iPhone X, the currently popular system of three-dimensional facial recognition, has reportedly been hacked. For example, the Vietnamese security firm, Bkav (www.bkav.com), proclaimed that they deceived iPhone X’s facial ID recognition using a 3D-printed face mask glued to a set of specifically manufactured 2D eye images. However, this report received little attention, as the effort required to deceive the facial ID was deemed impractical in real life.
The question becomes how likely will this conception become a reality. We argue that the risk is real and the risk is gradually rising due to the increased digitalization of medical data[10]. In the field of medical imaging, the application of artificial intelligence (AI) technologies is being heavily funded. AI technologies in both the training phase and the application phase involve large amounts of medical imaging data. In some countries, medical Big Data are easily accessible due to inadequate privacy laws. After obtaining these data, some technology companies may not strictly handle confidential patient information. Therefore, both basic personal information and biological information of a patient could be obtained and used to hack biometric identification technology. We can even imagine some possible channels for leaking this information. For example, with the cooling of the AI boom, some closed AI companies may resell these data or choose to ignore privacy regulations after acquiring medical data, or an individual may hack the in-cloud medical data storage servers.Some physicians remain unaware of the importance of patient data confidentiality, and unfortunately, they may unintentionally share these data.
It is time for us to reflect on whether the “uniqueness” of these biometric technologies is really as safe as manufacturers have advertised. In fact, biometrics security may even be worse than the password system security of which we have grown weary. For example, if a password is hacked and exposed, we can simply reset the password, but what should we do if our biological information secured in medical Big Data are hacked and exposed?
A multipronged strategy is needed to prevent such incidences.Medical institutions need proper data management systems and should use caution in cooperating with tech companies; physicians should be aware of the importance and risk of medical data and protect patients’ privacy; and the state should step up legislative efforts on medical data, thereby legally clarifying the ownership and right of use for medical data.