快猫短视频

EVERY MOMENT of every day, your heart thumps along reassuringly. It鈥檚 one of
the most familiar rhythms of life. For nearly a century, doctors have used the
spiky tracings of an electrocardiogram to look for signs of disease. Heart
specialists have become expert at spotting abnormal patterns in an ECG and
linking them to particular heart conditions. Yet there鈥檚 a lot more to these
commonplace recordings than even the trained eye of a cardiologist can see.

Far from being a steady, metronomic drone, heart rate ebbs and peaks
irregularly, even chaotically鈥攁nd it鈥檚 in these variations that the hidden
information lurks. Cardiologist Ary Goldberger of Harvard Medical School wanted
to unearth these buried nuggets. But he knew that to do so he鈥檇 need to draw on
a completely different kind of expertise.

鈥淭he idea of looking at signals and finding hidden information using
complexity theory and so forth is novel for clinicians,鈥� Goldberger says. But
for physicists and mathematicians in the discipline known as signal processing,
it鈥檚 old hat. And it鈥檚 not just academics and scientists who have the right
tricks up their sleeve. 鈥淚t might turn out that people who are good at financial
forecasting have something important to offer to time-series interpretation in
medicine,鈥� he says.

Goldberger needed to find these people and draw them into his search. So last
year, in conjunction with the annual Computers in Cardiology conference, he
mounted a contest. The challenge: to diagnose sleep apnoea using nothing but a
standard ECG.

People with sleep apnoea periodically stop breathing as they sleep. This puts
them at risk of serious illnesses, including heart disease and stroke. It鈥檚 a
big public health problem, says Goldberger. It鈥檚 also expensive and inconvenient
to diagnose. You have to spend the night in a sleep laboratory, and patients who
are treated then have to be brought back to see if the treatment is working.

A cheaper, easier way of screening people for sleep apnoea and then following
them up would be a big advance鈥攁nd ECGs seem to fit that bill. The
recorders comprise only two electrodes and a memory unit small enough to tuck in
a pocket, and are now unobtrusive enough to wear overnight. 鈥淭he possibility of
a link between sleep apnoea and heart rate has long been in the back of people鈥檚
minds,鈥� says Alan Murray, professor of cardiovascular physics at the University
of Newcastle upon Tyne. 鈥淏ut until recently ECGs weren鈥檛 thought to contain
enough information to get at it.鈥�

The raw data of Goldberger鈥檚 apnoea challenge was made up of 35 eight-hour
ECG records made in a sleep lab. Some subjects suffered from sleep apnoea, and
their records came with extra information showing exactly when and for how long
they stopped breathing. The rest of the subjects had normal breathing
patterns.

Goldberger and his colleagues posted all the data in this 鈥渢raining鈥� set on
the Web and invited competitors to analyse them. The aim was to develop a
computer algorithm that could pick out which patients suffered from apnoea. Then
the competitors had to put their algorithm through its paces by screening a
second, unlabelled 鈥渢est鈥� set of ECG records and identifying the people with
sleep apnoea. The prize for the winner was a modest $500, plus a rather
narrow sort of glory.

鈥淭his was an opportunity for people with powerful signal-processing tools to
apply them to real clinical problems,鈥� says Goldberger. 鈥淲e weren鈥檛 sure anyone
would respond.鈥� But they did. The challenge attracted 15 groups, three of which
normally worked completely outside medicine. Four entrants鈥攖wo American
groups, one British and one Irish鈥攃orrectly identified all the apnoea
sufferers except a few borderline cases that doctors would quibble over,
too.

鈥淲e looked at about 150 different things,鈥� says Connor Heneghan, an
electrical engineer at University College Dublin, who was a member of the
successful Irish team. Two sorts of information turned out be useful: the
variability of heart rate over timescales of a minute or more, and the way the
chest wall moves as you breathe. Chest movements alter heart rate slightly,
probably because of changes in mechanical pressure. When you stop breathing for
a moment, this effect vanishes. 鈥淵ou can pick a bit of that up in the ECG,鈥� says
Heneghan.

Cycles upon cycles

Several other groups鈥攊ncluding Murray鈥檚, which misidentified only two
of 30 patients鈥攁lso used algorithms that searched for changes in heart
rate. Because the rate varies naturally with each breath, further changes caused
by apnoea are imposed on this underlying pattern: cycles upon cycles. Still
other entrants looked for distinctive features in the waveform of the traces.
But whatever their approach, researchers learned quickly from each other. 鈥淭his
has proved that if you put the data sets out there, investigators will come and
bring great innovation with them,鈥� says Goldberger.

This year Golberger ran a second competition with a more difficult task: to
identify which of a group of subjects were prone to a condition called
paroxysmal atrial fibrillation (PAF) in which the heart鈥檚 upper chambers undergo
sudden bursts of rapid and irregular beating. This reduces the heart鈥檚 pumping
efficiency by as much as a third and increases the risk of stroke.

Contestants were again given a training set of recordings from healthy people
and PAF sufferers between attacks, then asked to screen people in an unlabelled
test set to identify those prone to PAF. A second challenge provided recordings
from known PAF sufferers and asked contestants to identify which recordings
presaged an imminent attack. The top scores were 82 per cent accuracy for
screening and 79 per cent for prediction.

The ECG challenges and the data that underlies them are part of a larger
effort to open up the mountains of physiological data gathered by doctors and
biologists to the widest possible audience. Geneticists have done this for years
by posting genetic data鈥攕equences of everything from single genes to whole
genomes鈥攐n the Web where other researchers have free access.

Goldberger and his colleagues have set up a similar clearing house of data:
not only ECGs but brainwaves, breathing patterns, oxygen concentrations, blood
pressures and much else contributed by enthusiasts around the world: 26 data
sets or 50 gigabytes at the last count. It鈥檚 part of a scheme called PhysioNet,
which is run from MIT by ECG signal-processing expert George Moody. PhysioNet is
backed by the National Center for Research Resources, part of the US National
Institutes of Health. The hope is that interested groups will use the data to
develop new ways of extracting hidden information.

Such analyses could have many uses, says Eugene Stanley, a physicist at
Boston University who is another of the scheme鈥檚 prime movers. Screening to
identify disease is the most obvious. Only a little more ambitious would be a
real-time monitor that would alert its wearer or their doctor that something was
amiss, and that they should take special precautions or seek medical care.

A wearable monitor might eventually do even more. For people whose innate
heart rhythm is unreliable, surgeons already routinely implant a defibrillator
that kicks the heart into line if it goes haywire. But this is only reacting to
a problem once it starts. A device incorporating a predictive signal-processing
algorithm might be able to take gentle action to prevent a problem rather than
more extreme measures to fix it.

So how far ahead might a smart monitor anticipate impending disaster?
Goldberger talks of minutes and hours, and while it might be possible to extend
that a little, no one is suggesting being able to see months or years ahead.
Instead, improvements are likely to focus on acquiring specific pointers to
individuals at risk. For example, doctors already know that, on average, people
whose heart rate doesn鈥檛 change much from moment to moment are at greater risk
of dying suddenly from heart problems
(快猫短视频, 3 January 1998, p 20).
鈥淲hat we really want to know is if this can be picked up in individuals,鈥�
says Murray.

All sorts of states and conditions, from depth of anaesthesia to depression,
might eventually turn out to be identifiable from analyses as simple as
variations in heart rate. Other types of physiological records could also be
fruitful (see 鈥淥n the blink鈥�). The body is full of unexploited
physiological signals, says Murray. The challenge is to decode the message they
are sending us.

Heart rate is not the only body rhythm that has a tale to tell about
a person鈥檚 health. The brain too has well-recognised rhythms, and many
researchers have studied electroencephalograms of people with epilepsy in the
hope of predicting seizures. Paul Gailey of the Fetzer Institute in Kalamazoo,
Michigan, is now analysing the complex behaviour of these traces. 鈥淚t would be
of great benefit to patients if they had even a few minutes鈥� warning. It may be
stretching things a bit, but people with epilepsy might even be able to drive,鈥�
he says.

Walking rhythms contain information, too. Jeff Hausdorff, a biomedical
engineer at Harvard Medical School and member of the PhysioNet team, fits people
with an ankle-worn recording device wired to pressure-sensing insoles
(快猫短视频, 22 September, p 23).
These signal the time at which the foot strikes
the ground and the forces that develop as it does so. Hausdorff鈥檚 aim is to
identify old people who are at risk of falling. 鈥淲e measured the fluctuations in
fairly simple things like stride interval. People who had a more variable gait
were more likely to fall,鈥� he says.

More speculatively, Stafford Lightman of the Research Centre for
Neuroendocrinology at Bristol University is measuring changes in chemical
signals. He suspects that susceptibility to all sorts of illnesses, from
cardiovascular disease to cancer, may be influenced by patterns of hormone
production that are programmed into the brain during childhood. 鈥淚n the past,鈥�
says Lightman, 鈥減eople would simply measure a hormone level and say it鈥檚 normal
or it isn鈥檛. But what鈥檚 really important is the organisation of the
蝉颈驳苍补濒濒颈苍驳.鈥�

For now, he鈥檚 focusing on fluctuations in the body鈥檚 output of hormones in
response to stress. To do this routinely will require an implanted device as
unobtrusive as an ECG recorder that can take frequent blood samples without
inhibiting normal activity. Lightman has developed such a device for rats, but
it is not yet ready for testing in humans, so his subjects come to his lab to
have samples drawn the old-fashioned way.