Awake brain mapping has been used for decades during tumor surgery. The surgeon exposes the brain, wakes the patient, stimulates a region with an electrode, and asks the patient to perform a task — name an object, move a finger, speak a sentence. If stimulation causes an error, the region is marked as functional and the surgeon avoids it. If stimulation causes no error, the region is considered expendable. The test is binary: error or no error.
A study published in Science Advances, led by Raouf Belkhir, showed that the binary test discards the most informative part of the signal. Stimulation doesn't produce binary responses. It produces continuous ones. A patient who takes 400 milliseconds to name a picture instead of 300 milliseconds has not made an error. But the slowdown indicates that the stimulated region participates in the naming circuit. A binary test records this as “no effect.” A continuous analysis records it as partial disruption — evidence that the region contributes to the function without being essential for it.
The structural insight is about information loss from quantization. Converting a continuous signal to a binary one destroys the gradient — the information about how much a region contributes, not just whether it contributes at all. The gradient is precisely the information the surgeon needs most. A region that produces errors when stimulated is unambiguously functional. A region that produces no change at all is unambiguously expendable. The difficult decisions are in between: regions that slow responses, increase hesitation, or degrade performance without producing outright errors. These are the regions where the tumor margin is negotiated, and the binary test says nothing about them.
The University of Rochester Medical Center has deployed MindTrace, a surgical decision-support tool that captures continuous performance data during awake mapping. The tool predicts how different surgical approaches might affect speech, movement, and other functions based on the full response profile rather than the binary classification. The gradient that was always present in the data is now visible.
The lesson extends beyond surgery. Any binary classification of a continuous phenomenon discards the gradient, and the gradient is often where the interesting information lives. Pass/fail testing, threshold-based alarms, presence/absence diagnostics — all convert a rich signal into a single bit. The bit is robust and easy to act on. But the signal it replaced contained the answer to the harder question.