Computer programs aid Alzheimer's diagnosis
Computer programs aid Alzheimer's diagnosis
As more physicians begin to use FDG-PET for diagnosing Alzheimer's disease, the demand for automated software systems that help interpret complex metabolic scans is increasing. Several new automated expert systems have been developed that can improve diagnostic accuracy and help assess risk for the disease.
The Centers for Medicare and Medicaid Services' decision last year to provide reimbursement for PET imaging of suspected Alzheimer's patients opened the field to physicians with varying levels of experience.
"We're now seeing substantial increases in demand for software programs that can help physicians with these complicated reads, especially in places that haven't been doing these types of diagnoses before," said Dr. Daniel Silverman, head of neuronuclear imaging at the University of California, Los Angeles Medical Center.
Silverman and colleagues developed the NeuroQ software program, which received FDA approval last year and is produced commercially by Atlanta-based Syntermed. The program quantifies the amount of activity in each of 240 regions of a PET scan and identifies abnormal areas. It returns information in a color-coded display in which blue represents normal metabolic activity, red represents abnormal, and violet indicates in-between activity.
Researchers at Gutenberg University Mainz in Germany have developed a computer-based expert system that can diagnose Alzheimer's disease with an accuracy comparable to that of experienced nuclear medicine physicians, according to a study presented at the Society of Nuclear Medicine meeting in June.
Nuclear medicine physicians often look for a typical pattern of impaired cerebral glucose metabolism in determining this diagnosis, said coauthor Dr. Peter Bartenstein, chair of nuclear medicine at Gutenberg.
Bartenstein and colleagues used 3D standard surface projections of stereotactically normalized PET brain scans and a data set of standardized regions of interest. These were projected on frontal, central, parietal, temporal, and occipital areas of the brain as the basis for an automated system. Two expert readers established a set of rules for diagnosis by comparing the 3D surface projections with 20 normal controls. They used the rules to develop an automated system that would generate a straightforward AD or non-AD diagnosis.
The researchers tested the system on 150 PET data sets, comparing the automated system results with reads done by three experts who had been blinded to all other imaging or clinical data. Concordance between the automated system and the nuclear medicine experts for all data sets had a kappa value of 0.76 to 0.83. A kappa value greater than 0.7 indicated satisfactory congruence.
The use of the system did not significantly increase the time needed for analysis, which took less than 15 minutes, Bartenstein said.
One major application for the system could be training physicians to diagnose AD. Inexperienced readers reported that the system was both a welcome aid and a learning tool, he said.
Ongoing enhancements could include artificial intelligence that would improve the quality of the program's decisions. This implementation might also extend the system's capability to include identification of specific patterns in other dementia disorders such as frontotemporal or Lewy body dementia.
"Ultimately, the final diagnosis of the patient's PET scan should not be based on the results provided by the program alone," Bartenstein said. "It should be used mainly for self-evaluation. In difficult cases, it could be used to support the decision the physician has already made."
Researchers led by Lisa Mosconi, Ph.D., a research scientist in the department of psychiatry at New York University School of Medicine, have also developed a software system that delineates metabolic activity in the regions of the brain-in the case cited, the hippocampus-affected by Alzheimer's on PET scans. Details of the program, called HipMask, were published in the June 2005 issue of Neurology.
The software program's technique of sampling PET scans from Alzheimer's patients was based on research conducted by Mony de Leon, Ed.D., director of the Center for Brain Health at the university. De Leon's research demonstrated that the hippocampus decreased in size in patients with Alzheimer's whose disease had progressed clinically from mild cognitive impairment to dementia.
Obtaining accurate measurements of the hippocampus has been difficult in the past because the region is so small and can change in both shape and size in patients with Alzheimer's. The image analysis program used in the study employs a sampling technique to anatomically probe PET scans, using MRI.
Mosconi and colleagues applied the program to 136 PET scans from 53 healthy, normal subjects between the ages of 54 and 80. They followed the patients for nine to 24 years.
In 25 patients who would later experience either mild cognitive impairment or Alzheimer's disease, the software program indicated that glucose metabolism in the hippocampus was significantly reduced (15% to 40%) on the first scan compared with controls.
"We were confident that by using an anatomically precise sampling procedure, we could find hippocampal metabolic abnormalities. However, the accuracy of hippocampal metabolism in predicting decline from normal to AD was 85%, and this is very good if you think that decline to AD occurred on average nine years after the PET scan was performed," Mosconi said.
The sampling technique could eventually enable researchers to screen for Alzheimer's in patients who do not show signs of cognitive impairment, according to the researchers.