Related Experiment Videos
A parameter-efficient deep learning approach to predict conversion from mild cognitive impairment to Alzheimer's
Simeon Spasov1, Luca Passamonti2, Andrea Duggento3
1University of Cambridge, Cambridge, Department of Computer Science and Technology, William Gates Building, 15 J J Thomson Ave, Cambridge, CB3 0FD, UK.
Neuroimage
|January 18, 2019
Summary
This study introduces a novel deep learning model to predict Alzheimer's disease (AD) progression in mild cognitive impairment (MCI) patients. The AI accurately identifies individuals at high risk of developing AD within three years using MRI and clinical data.
Area of Science:
- Neuroscience
- Artificial Intelligence
- Medical Imaging
Background:
- Mild cognitive impairment (MCI) can precede Alzheimer's disease (AD), but some forms remain stable.
- Differentiating progressive MCI from stable MCI is crucial for timely intervention and personalized treatment strategies.
- Objective measures are needed to identify MCI patients at high risk of progressing to AD.
Purpose of the Study:
- To develop and validate a novel deep learning architecture for predicting the conversion of MCI to AD within three years.
- To identify MCI patients at high risk of developing AD, enabling early intervention.
- To improve the accuracy of AD diagnosis and prognosis by integrating multi-modal data.
Main Methods:
- A novel deep learning architecture using dual learning and 3D separable convolutions was developed.
- The model integrated structural magnetic resonance imaging (MRI), demographic, neuropsychological, and APOe4 genetic data.
- Multi-task learning was employed to simultaneously predict MCI to AD conversion and classify AD vs. healthy controls.
Main Results:
- The deep learning model achieved an Area Under the Curve (AUC) of 0.925 for distinguishing MCI patients who develop AD within 3 years from those with stable MCI.
- The model demonstrated a 10-fold cross-validated accuracy of 86%, sensitivity of 87.5%, and specificity of 85% for predicting MCI to AD conversion.
- The model achieved perfect AUC (1) and 100% accuracy, sensitivity, and specificity when classifying AD patients from healthy controls.
Conclusions:
- The developed deep learning model shows high performance in predicting AD progression in MCI patients.
- The multi-modal approach, particularly using structural MRI and clinical data, is effective for AD prognostication.
- This framework offers a flexible and robust tool for computer-aided diagnosis of neurodegenerative diseases.