At RSNA this year, Philips showed DOTmed News a work-in-progress system that could combine functional information from PET imaging with structural insights from MR scans to predict dementia before patients even show clinical symptoms.

According to the National Institutes of Health, dementia strikes an estimated 6.8 million Americans. In some areas, almost half of all people age 85 and over suffer from a neurodegenerative disease, such as Alzheimer's or frontotemporal dementia.

Ordinarily, people only get a diagnosis of dementia once they go to their doctor after symptoms begin to show, but by then it might be too late for new treatments in the pipeline designed to halt the progress of the disease.

Unfortunately, with current imaging techniques catching dementia at an early stage is hard, Philips says.

In general, if a doctor suspects a patient has dementia, the patient will undergo an MR scan to rule out lesions that can cause dementia-like symptoms. The scan can also reveal increased ventricle size, a tell-tale sign of some kinds of dementia. But if the MR is inconclusive, the neurologist might request a follow-up PET scan. PET imaging can help doctors see which regions of the brain suffer from hypometabolism, low metabolic activity often associated with Alzheimer's, for instance.

At an advanced stage, say 20 years into the disease, differences between healthy and abnormal brains are obvious in most imaging studies; one brain looks healthy and full with lots of metabolic activity, the other shrunken, decayed, blighted by swollen ventricles and metabolically sluggish.

"The challenge is to see the difference already after one year, when it is much more subtle," Steve Klink, a spokesman for Philips' research division, tells DOTmed News.

"In reality," he continues, "only highly experienced neurologists can interpret a PET image [for early dementia], because you have to look for really small intensity differences."

Some digital assistance

To help practitioners, Philips is developing software that will act like one of the greatest PET or MR readers, and automatically code patients' scans to say how closely they match those of diseased brains.

The program works by comparing brain scans of patients against a database of scans from healthy ones. The software then checks to see how much the patient differs from the healthy subjects, and if those differences match up with known disease-specific patterns that indicate dementia. Once the analysis is done, the program will provide a score, shown as a percentage, indicating how closely the patient's brain matches a certain disease pattern, and thus it can aid the doctor in determining how likely it is the patient has the disease.

"The outcome is the system says, 'This has a 60 percent resemblance to a patient with Alzheimer's, 20 percent with frontotemporal dementia, and 10 percent to others,'" Klink says.

So far, studies are quite preliminary, but results look promising, according to Philips.

One retrospective study looked at PET scans of patients who had Alzheimer's and frontotemporal dementia, as well as healthy subjects, with diagnoses previously given by a medical expert. Philips says the software's own diagnosis matched the expert's opinion in around 98 percent of the cases. A second study involving 48 subjects, which also included scans from patients with Lewy-Body dementia, was a tad weaker, with the software agreeing with the experts around 80 percent of the time.

"The second study was a bit more challenging," Klink admits. But he notes that the study worked with PET alone, not PET combined with MR, which is what Philips ultimately hopes for the program.

"In the end, we want highest sensitivity and specificity," says Klink. "We don't know if you would reach that alone with PET, that's why we also want to incorporate the MR information."