Biomarkers.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Mild Cognitive Impairment (MCI) is an early indicator of dementia, necessitating early detection for timely intervention.
• Traditional 2D and 3D Convolutional Neural Networks (CNNs) face limitations in spatial correlation analysis and computational efficiency for MCI classification.
• Accurate classification of MCI converters (MCI-C) and non-converters (MCI-NC) is crucial for predicting dementia progression.

Purpose of the Study:

• To develop and validate a novel sequence-based framework for classifying MCI-C and MCI-NC using structural MRI (s-MRI).
• To overcome the computational and spatial correlation limitations of existing deep learning models for MCI diagnosis.
• To analyze gray matter decay patterns for improved early dementia detection.

Main Methods:

• Utilized s-MRI data from 187 subjects (75 MCI-C, 112 MCI-NC) from the Alzheimer's Disease Neuroimaging Initiative (ADNI).
• Extracted features from 106 MRI slices per subject using Local Binary Patterns (LBP) and its variants, creating feature vectors.
• Employed a Layer-wise Adaptive Sine Activation (LASA) based Bidirectional Recurrent Neural Network (BiRNN) to model temporal and spatial relationships between MRI slices.

Main Results:

• The proposed model demonstrated strong generalization performance with validation accuracy exceeding training accuracy.
• Achieved an average accuracy of 97.4% (±0.2 standard deviation) across 30 experiments.
• The model's effectiveness and reliability in classifying MCI subtypes were confirmed.

Conclusions:

• The novel framework offers high accuracy and edge-device compatibility for early MCI diagnosis.
• This approach facilitates cost-effective and accessible dementia diagnosis in diverse healthcare settings, including resource-constrained environments.
• The method provides an efficient and practical alternative to traditional deep learning models for early dementia detection.