They then woke up the dreamer and asked him or her what images they had seen, a process that was repeated 200 times. These answers were compared with the brain maps that the MRI scanner produced.

Researchers were then able to predict what images the volunteers had seen with a 60 per cent accuracy rate, rising to more than 70 per cent with around 15 specific items including men, words and books, they said.

"We have concluded that we successfully decoded some kinds of dreams with a distinctively high success rate," said Yukiyasu Kamitani, a senior researcher at the laboratories and head of the study team. "I believe it was a key step towards reading dreams more precisely."

The subject of centuries of speculation that has captivated humanity since ancient times, dreams are hard to study. Experiments with mice have shown they experienced dreamless sleep and dreaming, the content of which can be supplied by memories. But animals cannot confirm what they have dreamed about.

Also rapid-eye movement (REM) sleep - the stage richest in dreams - usually only starts about 90 minutes into sleep. This makes it difficult to collect data, such as having to wake up the subject. Conventional MRI, alas, is noisy.

The Japanese researchers sidestepped these issues by recording the brain activities of three adult male volunteers during the early stages of sleep. They were repeatedly awakened and asked for details about what they had experienced when asleep. A recall by a volunteer cited in the paper said: "There were persons, about three persons, inside some sort of hall. There was a male, a female and maybe like a child. Ah, it was like a boy, a girl and a mother. I don't think that there was any colour."

After gathering 200 such reports from the volunteers, the researchers used a lexical database to group the dreamed objects in general categories, such as street, furniture and girl. The participants were then asked to look at images of things in those categories while their brains were scanned. Computer algorithms then matched these patterns of brain activity with those general object categories.

Once the computer had been programmed to make these matches, it scanned the subjects to assign object categories to the detected brain activities. On average, the computer program made accurate object matching in a dream 70 per cent of the time, a rate high enough to convince the researchers that it was not achieved by chance or luck.

"It's striking work," Vanderbilt University cognitive psychologist Frank Tong told the journal. "It's a demonstration that brain activity during dreaming is very similar to activity during wakefulness."

Tong, who was not involved in the research, said it could lead to a better understanding of what the brain did during different states of sleep, such as those experienced by coma patients.

Kamitani first revealed his team's work in October at the annual meeting of the Society for Neuroscience in New Orleans.

They find that the same parts of the brain are used to process visual patterns, whether the person is awake or asleep. He believes this may explain why dreams are often so vivid to the dreamers. "Our study shows that during dreaming, some brain areas show activity patterns similar to those elicited by pictures of related content," Kamitani told the conference in October.

"Thus using a database of picture-elicited brain activity and a pattern recognition algorithm, we can read out or decode what a person might be seeing from brain scans during dreaming."

He argues that further research should reveal not only simple patterns, but more dynamic and emotional aspects of dreaming.