making CONNECTIONS
David Hambling looks at recent developments in what may turn out to be the ultimate long haul in science – understanding the complex workings of the human brain.
The human brain is often called the most complex object known to science, with around 100 billion neurons, each connected with up to up to 7,000 others. If we understood how those connections worked, we would know how to repair a brain, how to replicate it, perhaps even how to improve it. But do we have the right tool for the job?
In earlier times, researchers could only poke and prod at the brain, an activity that culminated in some dubious psychosurgery. Later they tickled the brain with electric currents and guessed by the patient’s response what each brain area did. Since 1991 researchers – the well-funded ones – have had functional Magnetic Resonance Imaging (fMRI). This non-intrusive 3D technique sees inside the brain and, by detecting oxygenated blood flow, indicates which parts are working at any given instant.
Researchers can see what happens in the brain when a mathematician struggles with an equation, an arachnophobe sees a spider, or an artist visualises a scene. There are technical complexities, but in essence fMRI shows brain area activation.
fMRI is a remarkable advance. Victorian phrenologists could only guess which ‘organ’ of the brain was responsible for which mental function. They were almost entirely wrong, apart from the ‘faculty for words’, which was located close to Wernicke’s area, now known to control speech (somewhere behind your left ear). Now the brain is divided into over 300 areas, each with a distinct, though not necessarily wellunderstood, function.
How the brain is affected by conditions from Alzheimer’s to autism, and how drug treatments work, has become clearer thanks to fMRI. No documentary about the brain is complete without the colourful computer-generated fMRI graphics. The simplifications and exaggerations of popular science faced with gee-whizz technology led to the powers of fMRI being overstated, followed by an inevitable media backlash.
Craig Bennett of the University of California won an IgNobel Prize in 2012 for an fMRI study on a dead salmon [ FT295:20]. As usual, the imaging process involved a large number of image cells or voxels. With the correct processing, background noise and spurious signals were cancelled out, showing no activity. However, by selectively using uncorrected data, Bennett ‘demonstrated’ brain activation in certain areas.
“The more chances you have to find a result, the more likely you are to find one, even by chance,” says Bennett. “We have accepted statistical methods to correct for this, but not all scientists use these methods in their neuroimaging analysis.”
Weeding out false positives remains a chronic issue in fMRI studies. In 2016 the New York Times suggested that 15 years of fMRI work was invalidated by a flaw in analysis. Anders Eklund and Thomas Nichols of Linköping University in Sweden had published a study showing the correction process for false positives did not work in all circumstances. Some of the 40,000 studies published using fMRI data undoubtedly reported phantom activations as bogus as the signal from the dead salmon. However, according to Nichols, the number of studies affected was less than 4,000, and only a fraction of these were actually wrong. More importantly, the methods used to analyse fMRI have been corrected.
Seeing which brain areas are activated only takes us so far. We also need to understand how connections between brain areas work. Ultimately, the brain is nothing but connections. Neural networking software, which mimics how a brain processes information by changing the strength of connections, is increasingly used for tasks from spotting cancer on X-rays to controlling robot fruit-pickers. To build a neural network model that matches the brain we need to have its ‘connectome’, the connection equivalent of the human genome. This requires a process known as network analysis.
We already know that brain areas with many interconnections are important in largescale brain functions. A group of 12 brain areas known as the “rich club” is a hub of
brain activity. These well-connected areas handle information processed in other brain areas. Some researchers believe the “rich club” is where human consciousness resides.
Again, the methodology is challenged, this time by a recent paper, called “Could a Neuroscientist Understand a Microprocessor?” in the journal PLOS One. Two researchers applied the network analysis techniques used by neurobiologists to a simple computer chip, the 6502. (Older readers will recognise this as the processor in the venerable BBC Micro.) The 6502 is infinitely simpler than the human brain, but the traffic analysis still failed to establish accurately the hierarchy of information processing.
In particular, while the analysis located some of the key structures present in the chip, but gave little idea of how they were working or how they related to each other. It also turned up spurious connections that were coincidental or trivial. For example, some transistors appeared to be activated only while playing “Donkey Kong”; they were not related specifically to the game as neuroscientists might assume, but only to some lower-level functions. The “Rich club” may be equally misleading.
In a sense, this is not a surprise. Researchers have had the complete connectome of a worm called Cænorhabditis elegans since 1986, which has just 302 neurons. Even with this, they are still unravelling how the worm responds to temperature, pressure and light and how it coordinates movement.
Understanding the human brain is perhaps the ultimate long haul in science. Sequencing the three billion base pairs of the human genome took 20 years; understanding and applying that knowledge to treat inherited conditions will take much longer. Recording and deciphering the connectome will make the genome look like child’s play.
The path of scientific progress is likely to be erratic as usual. In a couple of decades fMRI has provided more insights than all the previous centuries. More and better tools are needed. However, there will always be critics keen to seize on failures of neuroscience. After all, who want to believe their mind can be reduced to a set of numbers?