Automated classification of Alzheimer's disease, mild cognitive impairment, and cognitively normal patients using 3D convolutional neural network and radiomic features from T1-weighted brain MRI: A comparative study on detection accuracy

Abstract

Objectives

Alzheimer's disease (AD) is a common neurodegenerative disorder that primarily affects older individuals. Due to its high incidence, an accurate and efficient stratification system could greatly aid in the clinical diagnosis and prognosis of AD patients. Convolutional neural networks (CNN) approaches have demonstrated exceptional performance in the automated stratification of AD, mild cognitive impairment (MCI) and cognitively normal (CN) participants using MRI, owing to their high predictive accuracy and reliability. Therefore, we aimed to develop an algorithm based on CNN and radiomic features derived from ROIs of bilateral hippocampus and amygdala in brain MRI for stratification between AD, MCI and CN.

Methods

In this study, we proposed a CNN and radiomic features-based algorithm using the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. T1-weighted images were used. We utilized three datasets, including AD (199 cases, 602 images), MCI (200 cases, 948 images), and CN (200 cases, 853 images), to perform binary classification (AD vs. CN, AD vs. MCI, and MCI vs. CN). Finally, we obtained the accuracy (ACC) and the area under the curve of the receiver operating characteristic curve (AUC) to evaluate the performance of the algorithm.

Results

Our proposed algorithm achieved acceptable overall discrimination accuracy. In the term of AD vs CN, radiomic-based algorithm alone obtained ACC of 82.6 % and AUC of 88.8, CNN-based algorithm obtained ACC of 80 % and AUC of 87.2 and their fusion showed ACC of 84.4 % and AUC of 90. In the term of MCI vs CN, radiomic-based algorithm alone obtained ACC of 71.6 % and AUC of 77.8, CNN-based algorithm obtained ACC of 69 % and AUC of 75 and their fusion showed ACC of 72.7 % and AUC of 80. In the term of AD vs MCI, radiomic-based algorithm alone obtained ACC of 57 % and AUC of 57.5, CNN-based algorithm obtained ACC of 56.6 % and AUC of 57.7 and their fusion showed ACC of 58 % and AUC of 59.5.

Conclusion

In conclusion, it has been determined that hippocampus and amygdala-based stratification using CNN features and radiomic features-based algorithm is a promising method for the classification of AD, MCI, and CN participants.

Advances in knowledge

This study proposed an automated procedures based on MRI-derived radiomic features and CNN for classification between AD, MCI and CN.