Dream Decoding with fMRI

In 2013, neuroscientists at ATR Labs in Kyoto, Japan, woke volunteers repeatedly inside an fMRI scanner, asked what they had been dreaming about, and then used a computer algorithm trained on their brain activity to predict the general content of future dreams with about 60 percent accuracy — turning 'reading minds' during sleep from science fiction into an early, real experiment.

How fMRI measures brain activity during sleep. What the 2013 Kamitani Lab dream-decoding study actually found. Why dream content is linked to the same visual brain regions used while awake. The current limits of dream-reading technology.

fMRI, or functional Magnetic Resonance Imaging, tracks blood flow to active brain regions moment to moment by detecting oxygen levels in blood — a measurement called the BOLD signal (Blood-Oxygen-Level-Dependent). More active neurons demand more oxygenated blood, so researchers can map which brain regions are working hardest at any given instant, including during sleep.

Researchers led by Yukiyasu Kamitani woke 3 participants dozens of times each right after they entered sleep, asked them to describe what they had seen in the dream, and matched the reports to broad visual categories such as person, car, or book. They trained a machine-learning classifier on fMRI patterns recorded from those same participants while they viewed similar images WHILE AWAKE. When the classifier was applied to new dream scans, it correctly predicted the dream's visual category about 60 percent of the time — well above the roughly 9 to 15 percent accuracy expected by random chance.

Current dream decoding can only guess broad categories, not full detailed scenes or storylines. It also requires the subject to lie inside a large, loud scanner, and requires waking someone up repeatedly through the night, which disrupts natural sleep cycles. That makes it a lab research tool for studying the brain, not a device you could use at home to record a bedside dream diary.