Biomarkers.

Area of Science:

• Neuroimaging and computational neuroscience
• Artificial intelligence in medicine
• Neurodegenerative disease research

Background:

• Alzheimer's disease (AD) diagnosis requires early detection for intervention effectiveness.
• Mild Cognitive Impairment (MCI) is a prodromal stage of AD, but predicting its progression is challenging due to heterogeneity.
• Advanced diagnostic tools are needed to accurately identify MCI individuals who will progress to AD.

Purpose of the Study:

• To develop a highly generalizable deep learning approach for the early diagnosis of MCI progression to AD.
• To utilize Magnetic Resonance Imaging (MRI) for predicting conversion from MCI to AD.
• To improve diagnostic accuracy and generalizability in distinguishing between cognitive normal (CN), stable MCI (sMCI), and progressive MCI (pMCI) individuals.

Main Methods:

• A deep learning model was developed using multi-plane feature extraction and an attention-based mechanism.
• The model was trained on MRI data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.
• Generalizability was validated on an independent dataset from the Gwangju Alzheimer's and Related Dementia (GARD) database, comparing CN, MCI, pMCI, and sMCI groups.

Main Results:

• The model achieved high accuracy (91.92%) and AUC (96.22%) for CN vs. MCI, and (78.16%, 83.77%) for pMCI vs. sMCI on the ADNI dataset.
• On the GARD dataset, the model maintained robust performance with accuracy scores of 86.25% (CN vs. MCI) and 76.14% (pMCI vs. sMCI).
• t-SNE visualization confirmed effective feature separation across diagnostic groups, indicating strong discriminative capability.

Conclusions:

• The proposed deep learning framework demonstrates high accuracy and generalizability for early AD diagnosis via MCI progression prediction.
• This approach can reliably distinguish between CN and MCI individuals, and predict MCI progression to AD.
• The findings support the potential of this neuroimaging analysis tool for clinical decision-making in early Alzheimer's disease detection.