EPFL’s low-power neural chip reads synchrony across brain regions

A brain implant can record a signal from one location and respond to it. But many of the symptoms researchers want to understand, including changes in mood and cognition, involve interactions among brain regions rather than one centre acting alone. A team at Switzerland's EPFL has built a chip designed to detect that coordination, analyse patterns and produce timing signals that could one day inform adaptive stimulation.[1]

The device is called LogSync. EPFL describes a chip approximately four by two millimetres in size, designed to operate below 200 microwatts. It measures features such as the strength of neural oscillations and the synchrony between signals from different regions. A machine-learning classifier uses those measurements to distinguish categories of cognitive and emotional activity in previously collected human recordings. Separate experiments in rats tested its ability to track oscillations and trigger stimulation at chosen phases.[1]

The human work used existing recordings; the live implant experiments were in rats. EPFL’s account does not report treatment of depression or anxiety in human patients. The result is a compact signal-processing and triggering system intended to support that future clinical research.[1]

Why communication may be a better signal

Deep brain stimulation is already used in the treatment of conditions such as Parkinson's disease, where precisely located stimulation can alter abnormal neural activity. Much psychiatric research is exploring more complicated patterns. If several regions communicate differently during a particular mental state, the timing relationship between their signals may be useful as a biomarker. That does not mean the biomarker is unique to a disorder, but it offers a measurement richer than activity at a single electrode.[1]

In an open-loop stimulator, the system delivers pulses according to a prescribed schedule or fixed settings. A closed-loop design reads neural activity and adjusts stimulation in response. LogSync is an engineering attempt to support the second approach without requiring a large external computer to perform all analysis. The chip computes compact features of network activity and feeds them into a classifier and trigger logic.[1]

There is a severe power constraint. Any implanted system must work within limits created by battery capacity, heat and the surrounding tissue. Moving raw recordings continuously to an external processor may be costly in energy and communication. EPFL's team uses logarithmic computation to simplify mathematical operations needed for its synchrony estimates. The reported sub-200-microwatt consumption is part of that engineering result, not a clinical efficacy score.[1]

Two datasets, two kinds of evidence

For cognitive decoding, the researchers used archived human brain recordings from experiments in which participants performed tasks or viewed emotionally arousing video. The system distinguished state categories represented in those datasets. How reliably it would generalise to different people, implanted electrode arrangements or unscripted daily life is a further question. Such transfer is often where promising biosignal systems encounter serious variability.[1]

The rat experiments address another aspect: the integrated device tracked neural rhythms and generated stimulation triggers at specified phases. That demonstrates a component of a responsive control loop in a living system. It is not equivalent to demonstrating that a particular pattern of stimulation safely changes a psychiatric outcome. The team reports that experiments and signal decoding need to be built together, and much of the clinical validation lies ahead.[1]

The engineering opportunity is to place network-level analysis close to the electrodes, within an implant’s tight energy budget. Clinical usefulness will depend on a further link: whether responding to the detected state improves a patient’s condition.[1]

Report an error or suggest a correction