Room: Exhibit Hall | Forum 8
Purpose: Alzheimer's disease (AD) is one of the most prominent diseases caused by aging and cognitive disorder, which will further lead to dementia. This study is to classify AD patients from healthy subjects based on resting-state functional MRI (rs-fMRI) with novel deep learning method to aid the early diagnosis of AD.
Methods: Resting-state fMRI data of 37 AD patients and 36 healthy subjects were selected from the Alzheimerâ€™s Disease Neuroimaging Initiative (ADNI) database following same protocols (3.0T Philips scanners, TR/TE =3000/30, FA= 80Â°). Then the rs-fMRI were preprocessed, and divided into 200 cerebral regions according to the Craddock 200 atlas. The regional mean time series of each region was extracted and normalized to the average signal of the corresponding ROI. Each time series contained 130 time points, and was then randomly cropped into 10 sections with a scan length of 3 min (60 time points). In total, the number of time series of the AD subset is increased to 370, and the CN subset is increased to 360. These time series were utilized as the input of the proposed multi-Long Short-Term Memory based network, and were divided into 90% for training and 10% for testing. Accuracy (ACC), area under the curve (AUC) and sensitivity (SEN) was calculated to evaluate the performance of the proposed network.
Results: The classification results showed that our proposed method can distinguish AD patients from healthy subjects only based on the rs-fMRI time series with an averaged ACC of 78.1%, AUC of 88.6% and SEN of 79.4%.
Conclusion: Our proposed method can well classify AD based on only rs-fMRI time series signal. This result demonstrates that there is discriminative power of AD carried by the latent time-varying correlation among brain ROIs, and our method could helped to reveal and utilized it to aid the classification process.
Funding Support, Disclosures, and Conflict of Interest: This work is partially supported by NIH R01 CA235723 and the seed grant of Radiation Oncology Department at University of Texas Southwestern Medical Center.