363 ‒ A new frontier in neurosurgery: brain-computer interfaces, new hope for brain diseases, & more

KEYPOINTS

• 01

  Neurosurgery has steadily moved toward less collateral damage, though the craniotomy still has its place.

  UCSF neurosurgeon Edward Chang traces modern neurosurgery to Harvey Cushing, who diagnosed early pituitary tumors and helped bring in electrocautery, and to Wilder Penfield, who popularized awake surgery and the homunculus map. Aneurysm repair used to mean a 7–9 inch incision and a microscope. Today about 90% are treated with coils through a catheter in the groin. Clot retrieval has turned many severe strokes into 'brain attacks' handled like heart attacks, and some patients go home the next day. Laser probes and focused ultrasound can now treat deep targets such as the nuclei behind tremor.

• 02

  Glioblastoma is still lethal, but molecular profiling and immune strategies are shifting the approach.

  GBM arises from glial support cells and grows so fast it outstrips its blood supply, leaving necrosis. It isn't considered heritable, and each tumor has a different combination of mutations. Genetic profiling is now routine at academic centers, opening the door to targeted chemotherapy. GBMs also hide from the immune system, so 'unmasking' them is a major goal. For now, the more of the tumor surgeons remove, the longer patients survive, but microscopic cells beyond what MRI shows grow back. Focused ultrasound that temporarily opens the blood–brain barrier for drug delivery is being studied at UCSF. Chang's 2040 hope is to turn an average survival of about 18 months into a chronic condition.

• 03

  Awake surgery works because the brain itself feels no pain.

  The scalp is numbed with lidocaine and epinephrine. Bone and brain tissue have no pain receptors, but the dura and some blood vessels do, so they're numbed too. Patients aren't intubated. The head is fixed in a holder and light propofol sedation is used at the start. The awake window is usually only 1–2 hours of a 6–8 hour operation, for mapping language or movement with electrical stimulation. The trade-off is resecting more tumor or seizure tissue against the risk of paralysis or aphasia.

• 04

  The '10% of the brain' myth has a grain of truth: some regions are redundant and plastic.

  Roughly 10–15% is critical for basics like moving, speaking and seeing. Chang has removed an entire frontal lobe on one side many times. With slow-growing lesions, the other hemisphere, which was already sharing the work, takes over through synaptic plasticity. Language sits in the left hemisphere for about 99% of right-handers, about 2 cm above the left ear, and that real estate can't be sacrificed.

• 05

  Severing the corpus callosum stops seizures from spreading rather than curing them.

  For drop attacks, where seizures spread so fast that patients collapse and may need to wear helmets, surgeons usually cut the front two-thirds of this heavily myelinated left–right highway. Chang compares a seizure to a whole stadium doing the wave instead of holding separate conversations. Cutting it fully risks a disconnection syndrome, with one hemisphere unaware of what the other is feeling, which is why partial cuts are preferred. The delicate step is protecting the pericallosal arteries, since injuring them can paralyze the legs.

• 06

  Recording from the brain's surface is the sweet spot for brain–computer interfaces.

  Taking scalp EEG as 1, electrodes on the cortical surface (ECoG) give roughly 1,000x the resolution, because skull and scalp blur the signal. Penetrating single-neuron electrodes add perhaps another 5x, but scar tissue forms around them and their recordings drift from neuron to neuron within hours or days, so decoders must be recalibrated. Because ECoG doesn't break the brain's surface membrane, it avoids much of that reaction and can blanket a whole region, which absorbs individual variation in exactly where functions sit.

• 07

  The decoder translates attempted speech, not inner thoughts.

  The array sits over the speech-motor cortex controlling the lips, jaw, tongue and larynx, so the patient must actually try to speak. Reading or imagining a sentence isn't enough. AI breaks the signal into 10–20 ms windows and maps them to statistically derived speech units, using an open-source speech-recognition model from Meta. A language model then infers the likeliest words, much as autocorrect does, and a synthesizer can recreate the person's pre-injury voice.

• 08

  Hearing loss deprives the brain and may accelerate cognitive decline.

  Many older adults with normal hearing tests struggle in noisy restaurants because the problem is signal processing in the brain. Unrecognized hearing loss leads to social isolation and less stimulation, and Chang says many studies link it to accelerated memory loss. Peter's practice treats the link as causal, while noting that a recent trial suggesting otherwise had methodological problems. Chang thinks vision loss has less of this effect.