BehindTheMedspeak: Knee analysis for personal identification

According to an April 2, 2009 Economist article, knees are unique enough to provide a means of personal identification.

The story  follows.

•••••••••••••••••••••••

Knobbly ID

Once the art of disguise would rely on a false moustache, a wig and probably a touch of make-up to foil an identification check. Today you need to beat an electronic fingerprint-analyser or a retinal scanner. It can be done with contact lenses designed to mislead eye scanners and some stick-on false fingerprints. But a new method of identification may prove much harder to evade with fake body parts. Knees, it turns out, also have unique characteristics. An X-ray of a person’s knee can show if he limps or was injured in a sporting accident, even if it happened years ago. Hospital records often contain lots of pictures of knees.

Lior Shamir, a geneticist at the National Institutes of Health in Maryland, has developed a knee-analysing mathematical algorithm for medical use. Algorithms are used by computers to analyse knee images in order to compare and contrast tiny structures in the joint that might indicate diseases like osteoarthritis. Computers make this work less labour-intensive. Dr Shamir and his colleagues now think his algorithm could identify individuals as well.

To find out, they used the algorithm to explore X-ray images of the general structure of various knees and then to look in greater detail by measuring the texture of the bone by monitoring differences in individual picture points, called pixels. The researchers found that analysing fine details at this level was roughly equivalent to studying fingerprints. The best identification was possible by concentrating on a smaller image of the centre of the joint rather than the entire knee.

The researchers  tested the technique on 425 people, who each had four X-rays taken of their knees. The knee images were then digitised for processing by a computer using the algorithm. Three of the X-rays of each person were matched to the individual they came from. The fourth image of each person was then analysed anonymously and the computer intructed to search for a match.

The team report in the International Journal of Biometrics that the system was able to match a knee to an individual correctly 34% of the time. It was also able to pick the ten closest matches to a particular knee 56% of the time. Although this is far short of the nearly perfect scores of retinal scans and fingerprinting, the work is at an early stage and the researchers expect that with fine tuning the analysis will become more accurate.

They are planning to tweak the algorithm to look at more features of the knee and to try magnetic-resonance imaging (MRI) to produce images of soft tissue in greater detail. The use of MRI would also allow more frequent scans for identification purposes by avoiding the potential risk of X-rays. The possible uses include high-security checks and forensic work. And the time may yet come when, before you board an aeroplane, a security guard will want a good look at your knees.

•••••••••••••••••••••••

The abstract of their paper follows.

•••••••••••••••••••••••

Biometric identification using knee X-rays

Identification of people often makes use of unique features of the face, fingerprints and retina. Beyond this, a similar identifying process can be applied to internal parts of the body that are not visible to the unaided eye. Here we show that knee X-rays can be used for the identification of individual persons. The image analysis method is based on the wnd-charm algorithm, which has been found effective for the diagnosis of clinical conditions of knee joints.  Experimental results show that the rank-10 identification accuracy using a dataset of 425 individuals is ∼56%, and the rank-1 accuracy is ∼34%. The dataset contained knee X-rays taken several years apart from each other, showing that the identifiable features correspond to specific persons, rather than the present clinical condition of the joint.