Combinations of multimodal neuroimaging biomarkers and cognitive test scores to identify patients with cognitive impairment

Yuriko Nakaoku¹Soshiro Ogata¹Kiyotaka Nemoto²Chikage Kakuta¹Eri Kiyoshige¹Kanako Teramoto¹Kiyomasa Nakatsuka¹Gantsetseg Ganbaatar¹Masafumi Ihara¹Kunihiro Nishimura^1*

• ¹National Cerebral and Cardiovascular Center, Suita, Japan
• ²University of Tsukuba, Tsukuba, Japan

Background: Early detection of mild cognitive impairment (MCI), defined as the prodromal stage of dementia, is key to delaying the progression to dementia through lifestyle interventions and/or pharmacological treatments. This study aimed to develop and test new identification models for MCI in community settings based on multiple sources of clinical features, including neuroimaging biomarkers. Methods: This cross-sectional study analyzed cognitive testing and MRI examination data from 148 community-dwelling older adults in Nobeoka City. MCI was assessed using the Memory Performance Index from the MCI Screen. The variables used for model development were multisource features, including MRI-derived biomarkers and cognitive test scores. Finally, MCI identification models were developed using a penalized logistic regression model with an elastic net algorithm. Results: Among the 148 participants (mean age, 78.6±5.2 years), 44.6% were identified as having MCI. The area under the curve for the elastic net model using baseline variables (i.e., age, sex, and education) and the multisource model were 0.74 (95% confidence interval, 0.59 to 0.89) and 0.81 (0.67 to 0.94) in the test datasets, respectively. The addition of neuroimaging biomarkers and cognitive test scores significantly improved the performance of the model identifying MCI (p = 0.012 by DeLong's test). The structural, perfusion, and diffusion MRI-derived biomarkers remained in the identification model with variable selection with the elastic net algorithm, and were thus considered important variables. Conclusion: Our multisource elastic net model demonstrated high performance at detecting MCI, suggesting that the combination of multimodal neuroimaging biomarkers contributes to MCI discrimination.