Study Assesses Amyloid PET/MRI Combination in Detection of Alzheimer’s Disease

Can a machine learning combination of amyloid beta positron emission tomography (Aβ PET) and structural MRI (sMRI) enhance differentiation between older healthy control (OHC) participants, those with mild cognitive impairment (MCI) and patients with Alzheimer’s disease (AD)?

For the study, recently published in Academic Radiology, researchers compared the single modalities sMRI and Aβ PET as well as a machine learning combination of the modalities in their review of 261 study participants. The cohort was comprised of 94 patients with AD (mean age of 69.62), 82 patients with MCI (mean age of 74.37) and 85 OHC (mean age of 70.76) participants, according to the study.

The study authors found an 81 percent AUC for sMRI, an 86 percent AUC for Aβ PET and an 89 percent AUC for the combined use of sMRI and Aβ PET features in differentiating between OHC and AD. The researchers noted the combination of sMRI/Aβ PET also demonstrated a 95 percent sensitivity rate, an 89 percent accuracy rate and an 80 percent specificity rate for detecting AD.

“Our results demonstrated Aβ distribution in the precuneus/posterior cingulate, lateral temporal and parietal cortex was more effective than other regions in the comparison between (OHC) and AD. The regions above are early affected by Aβ deposition in the hierarchical regional progression pattern from cognitively normal to AD, commonly represented as stage I and II according to Aβ staging,” wrote lead study author Yi-Wen Bao, M.D., who is affiliated with the Department of Medical Imaging at the Affiliated Huaian No. 1 People’s Hospital of Nanjing Medical University in Jiangsu, China, and colleagues.

The researchers noted that the top 10 features for differentiating between OHC and AD with the combined use of modalities were Aβ PET features with the parietal lobe’s amyloid load being the top distinguishing feature.

When evaluating the modalities in differentiating OHC versus MCI, the study authors found that higher accuracy (89 percent) and sensitivity rates (95 percent) with sMRI/Aβ PET. They also noted that the combined model had equivalent AUC (74 percent) and accuracy rates (74 percent) to Aβ PET imaging alone. For this part of the analysis, the study authors pointed out that sMRI features comprised the top 10 features with cortical thickness of the left entorhinal cortex being the top distinguishing feature.

For the differentiation of MCI and AD, the combined model demonstrated a slightly higher AUC (71 percent) and accuracy rate (75 percent) in contrast to either sMRI or Aβ PET alone, equivalent sensitivity to Aβ PET (95 percent) and low equivalent specificity to sMRI (47 percent). In this analysis, the study authors said the Aβ load of the left occipital lobe was the top distinguishing characteristic, and the remaining top 10 features were related to regional Aβ PET.

“ … We speculate that brain atrophy can be more significant in the classification between earlier stages of AD (such as MCI) and OHC, but quantitative Aβ load may play a deciding role in differentiating advanced AD from others due to the role of Aβ in defining AD as a unique neurodegenerative disease,” maintained Bao and colleagues.

(Editor’s note: For related content, see “Can Deep Learning Automate Amyloid Positivity Assessment on Brain PET Imaging?,” “Could MRI-Guided Ultrasound Facilitate Improved Reduction of Amyloid-Beta Load in Patients with Alzheimer’s Disease?” and “Emerging MRI and PET Research Reveals Link Between Visceral Abdominal Fat and Early Signs of Alzheimer’s Disease.”)

In regard to study limitations, the authors acknowledged a limited cohort size and incomplete demographic information for some of the study participants. The researchers also conceded possible underestimation of AB uptake due to the use of non-specific amyloid tracer binding in white matter and minor segmentation artifacts within those regions.