Machine learning winnows memory-care cohort to only the most appropriate nuc-med patients

An AI-aided way has emerged to confidently select dementia patients who are likely to benefit from amyloid-PET imaging while appropriately de-selecting patients for whom the costly exam would probably be unhelpful.  

The selection method uses a computerized decision support (CDS) system based on personalized patient data as combed by supervised machine learning.

Researchers in the Netherlands designed the tool to help answer one question:

“If a clinician already has detailed information on key disease indicators—neuropsychological tests, apolipoprotein E (APOE) genotype status and brain imaging—would adding amyloid-PET guide the clinician to a more certain diagnosis?”

“Amyloid-PET” is shorthand for positron emission tomography augmented by injection with the radiotracer florbetaben (brand name Neuraceq), which helps neuroimaging specialists visualize beta-amyloid plaques in the brain.

The researchers found their homegrown AI tool narrowed a field of 286 amyloid-PET candidates—all of whom were clients of a memory-care clinic—to the 60 individuals (21%) who stood to benefit the most by undergoing the additional imaging exam.

The field included 135 controls, 108 persons with Alzheimer’s disease dementia, 33 with frontotemporal lobe dementia and 10 with vascular dementia.

Of the 60 amyloid-PET patients, 188 (66%) ended up receiving a diagnosis of “sufficient certainty.”

Publishing the results May 20 in PLOS One, lead investigator Hanneke Rhodius-Meester, MD, PhD, and colleagues at Amsterdam University Medical Center report that their computerized CDS approach bested three others with which they compared it:

• Without amyloid-PET, diagnostic classification was obtained in 155 (54% of) patients.  
• With amyloid-PET as recommended by the relevant appropriate use criteria, amyloid-PET was given to 113 (40% of) patients and diagnostic classification was obtained in 156 (55% of) patients.
• Performing amyloid-PET in all participants resulted in diagnostic classification in 154 (54% of) patients.

In their discussion, Rhodius-Meester and co-authors underscore that their computerized CDS approach advised performing an amyloid-PET scan in 21% of patients “without compromising proportion of correctly classified cases.”

More: 

“Our approach was thus more efficient than the other scenarios, where we would have performed PET in all patients, in none, or according to the appropriate use criteria (AUC). When implemented in a computer tool, this approach can support clinicians in making a balanced decision in ordering additional (expensive) amyloid-PET testing using personalized patient data.”

The study is available in full for free.