Deep Learning Based Diagnosis of Mild Cognitive Impairment Using Resting-State Functional MRI

Mild cognitive impairment (MCI) is the prodromal stage of Alzheimer’s Disease (AD) and patients with MCI are of high risk developing AD and dementia. Therefore, accurate detection of MCI is crucial to provide appropriate treatment as soon as possible. The purpose of this study was to develop a deep learning method named MCI-ARnet for the automatic diagnosis of MCI using resting-state functional MRI (rs-fMRI).

This method utilizes amplitude of low-frequency fluctuation (ALFF) and regional homogeneity (ReHo) extracted from rs-fMRI as inputs. These inputs are then passed into a developed dual-branch cross-collaboration feature extractor for advanced feature extraction. This feature extraction method enables the model to capture complementary information from both features more finely, thereby extracting more comprehensive and diversified feature representations. Finally, the two extracted features are effectively fused by the developed Feature-Enhanced Fusion module, and the fused features are passed to a fully connected layer classifier to automatically identify MCI.

We employed five classification evaluation metrics to validate the diagnostic performance of MCI-ARnet. The results indicate that MCI-ARnet achieved an accuracy of 92.50%, a precision of 93.75%, a recall of 92.11%, an F1 score of 92.92%, and an AUC of 0.9424.

The proposed method has the potential to be extended to clinical applications, assisting clinicians in diagnosing MCI.