Transfer Learning and Neural Network-Based Approach on Structural MRI Data for Prediction and Classification of...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alzheimer's disease (AD) lacks definitive treatment, making early diagnosis crucial for slowing progression.
• Structural magnetic resonance imaging (sMRI) and artificial intelligence (AI) show promise for AD detection.
• Distinguishing between early mild cognitive impairment (EMCI) and late mild cognitive impairment (LMCI) is vital for timely intervention.

Purpose of the Study:

• To differentiate Alzheimer's disease (AD) from normal controls (NC).
• To distinguish between early mild cognitive impairment (EMCI) and late mild cognitive impairment (LMCI).
• To evaluate the diagnostic performance (accuracy and AUC) of sMRI for early AD prediction.

Main Methods:

• Utilized a dataset of 398 participants from ADNI and OASIS databases.
• Included 98 individuals with AD, 102 with EMCI, 98 with LMCI, and 100 with NC.
• Employed a model incorporating DenseNet169, transfer learning, and class decomposition.

Main Results:

• Achieved an overall accuracy of 99.7% across all four classes.
• Demonstrated high Area Under the Curve (AUC) values for various comparisons (e.g., NC vs. AD: 0.985, EMCI vs. LMCI: 1.000).
• Successfully differentiated EMCI from LMCI with an AUC of 1.000.

Conclusions:

• The AI model effectively classifies AD stages, particularly distinguishing EMCI from LMCI.
• High accuracy and AUC confirm the model's strong performance in early AD diagnostics.
• Accurate diagnosis of cognitive impairment stages facilitates early AD prediction and management.