[Paper Review] A global feature extraction model for the effective computer aided diagnosis of mild cognitive impairment using structural MRI images

Chen Fang, Panuwat Janwattanapong+2 more|arXiv (Cornell University)|Dec 2, 2017

Dementia and Cognitive Impairment Research9 references3 citations

TL;DR

This paper proposes a global feature extraction model using Gaussian discriminant analysis (GDA)-based dual high-dimensional decision spaces to improve early diagnosis of mild cognitive impairment (MCI) from structural MRI. By integrating features from both brain hemispheres into a unified classifier, it achieves an F1 score of 91.06% and accuracy of 88.78% on held-out test data, outperforming prior local feature methods for CN vs. MCI classification.

ABSTRACT

Multiple modalities of biomarkers have been proved to be very sensitive in assessing the progression of Alzheimer's disease (AD), and using these modalities and machine learning algorithms, several approaches have been proposed to assist in the early diagnosis of AD. Among the recent investigated state-of-the-art approaches, Gaussian discriminant analysis (GDA)-based approaches have been demonstrated to be more effective and accurate in the classification of AD, especially for delineating its prodromal stage of mild cognitive impairment (MCI). Moreover, among those binary classification investigations, the local feature extraction methods were mostly used, which made them hardly be applied to a practical computer aided diagnosis system. Therefore, this study presents a novel global feature extraction model taking advantage of the recent proposed GDA-based dual high-dimensional decision spaces, which can significantly improve the early diagnosis performance comparing to those local feature extraction methods. In the true test using 20% held-out data, for discriminating the most challenging MCI group from the cognitively normal control (CN) group, an F1 score of 91.06%, an accuracy of 88.78%, a sensitivity of 91.80%, and a specificity of 83.78% were achieved that can be considered as the best performance obtained so far.

Motivation & Objective

• To address the limitation of local feature extraction methods that require separate feature sets for each binary comparison (CN vs. MCI, CN vs. AD, MCI vs. AD), which hinders clinical applicability.
• To develop a unified global feature extraction model that leverages both hemispheres’ data through dual high-dimensional decision spaces for improved classification performance.
• To identify the most significant cortical regions associated with Alzheimer’s disease progression using ANOVA-based feature ranking.
• To achieve state-of-the-art performance in early diagnosis of MCI using structural MRI and machine learning.
• To validate the model on a held-out test set to ensure robustness and generalization beyond cross-validation.

Proposed method

• The method employs a global GDA-based classifier trained on dual high-dimensional decision spaces derived from both left and right brain hemispheres.
• Features are extracted from 25 FreeSurfer-labeled cortical regions using seven morphometric measures: surface area, travel depth, geodesic depth, mean curvature, convexity, thickness, and volume.
• A 175-variable vector is generated per hemisphere (7 measures × 25 regions), and these are combined into a single global feature space for classification.
• ANOVA is used to rank features based on p-values, identifying the most discriminative features across the three diagnostic groups (CN, MCI, AD).
• Tenfold cross-validation is used to evaluate model performance, followed by a 20% held-out test set for final validation.
• The optimal feature set is selected based on global classifier performance, with the entorhinal cortex identified as the most significant region.

Experimental results

Research questions

• RQ1 Can a global feature extraction model that combines both hemispheres’ data outperform traditional local feature extraction methods in classifying MCI from cognitively normal controls?
• RQ2 Which cortical regions are most significantly associated with Alzheimer’s disease progression based on morphometric features from structural MRI?
• RQ3 Does the use of dual high-dimensional decision spaces in a GDA-based classifier improve early diagnosis performance compared to single-hemisphere or local feature approaches?
• RQ4 Can the proposed model achieve state-of-the-art performance on a held-out test set for CN vs. MCI classification?
• RQ5 Is the entorhinal cortex the most discriminative region for early detection of MCI, as confirmed by feature ranking and classification results?

Key findings

• The proposed global GDA-based model achieved an F1 score of 91.06%, accuracy of 88.78%, sensitivity of 91.80%, and specificity of 83.78% on the held-out test set for CN vs. MCI classification.
• The model significantly outperformed recent cross-validation results reported in the literature, particularly for the challenging CN vs. MCI comparison.
• The entorhinal cortex was identified as the most significant cortical region associated with AD progression, consistent with established neuropathological findings.
• The top five regions—entorhinal, middle temporal, inferior temporal, fusiform, and parahippocampal—were all confirmed as AD signature regions, validating the feature ranking method.
• The global classifier using dual decision spaces improved performance across all comparisons (CN vs. MCI, CN vs. AD, MCI vs. AD) compared to local models.
• The optimal feature set included 48 features from the left hemisphere and 61 from the right, indicating strong asymmetry in discriminative power.