Raleigh, North Carolina - When seven-month-old Natalie Beard's body arrived in the autopsy room, there were no outward signs of physical abuse. But behind her pretty brown eyes there was chaos.

Both retinas were puckered and clouded red. And there was acute bleeding outside and beneath the brain's outer membrane - the kind of bleeding most often associated with a burst aneurysm. To forensic experts, these were classic signs that Natalie was shaken to death.

The common wisdom in such "shaken-baby" cases was that the last person with the child before symptoms appeared was the guilty party, and a jury convicted babysitter Audrey Edmunds of first-degree reckless homicide.

Edmunds is now 10 years into her 18-year prison sentence, and she's seeking a new trial.

In the decade since her conviction, her attorneys say, many experts have studied the physics and biomechanics of shaken-baby syndrome and have concluded that shaking alone could not have produced Natalie's injuries without leaving other evidence of abuse.

Among those now questioning the diagnosis is Dr Robert Huntington, the forensic pathologist who examined Natalie's body and whose testimony helped put Edmunds away.

If the trial were held today, Huntington said recently: "I'd say she died of a head injury, and I don't know when it happened. There's room for reasonable doubt."

Three-and-a-half decades after the term was first used, there seems to be no middle ground in the debate.

"It doesn't exist," contends Dr John Plunkett. "You can't cause the injuries said to be caused by shaking."

Many paediatricians disagree.

"People confess to it. So it has to be possible," counters Dr Suzanne Starling.

She and her colleagues analysed 81 cases in which an adult confessed to shaking and/or battering a child. In cases where only shaking was admitted, the children were 2,39 times more likely to have retinal hemorrhages than victims of impact alone, they found, "suggesting that shaking is more likely to cause retinal haemorrhages than impact".

Plunkett scoffs: "What is the number-two cause of wrongful convictions? False confessions. You don't base scientific conclusions on what people confess to."

Much of the debate has centred on how quickly symptoms begin after a brain trauma. That question was central to the Edmunds' case.

On October 16, 1995, Edmunds was caring for her two daughters and another child when Cindy Beard dropped off her daughter, Natalie.

Natalie had had an ear infection and had vomited in recent days, but her parents say that appeared to have cleared up. But Edmunds says Natalie was unusually fussy that morning and refused to take a bottle.

Edmunds says she put Natalie down with a propped bottle and went to tend to the other children. When she went back to retrieve Natalie, the girl was crying and limp, her face slick with regurgitated formula.

At her 1996 trial, Huntington testified it was "highly probable" that Natalie was injured within two hours of being treated. That would mean the fatal injury occurred while Natalie was in Edmunds' care.

What changed his mind was a later case involving a child with injuries similar to Natalie's. That child had a "lucid interval" of more than 15 hours before the onset of symptoms, leading Huntington to acknowledge that Natalie could have been injured long before she was dropped off at Edmunds'.

The case continues. - Sapa-AP