MCI Detection Based on Deep Learning with Voice Spectrogram

We conduct a study on the classification of mild cognitive impairment (MCI) which is an early stage of dementia. We employ deep learning with a voice spectrogram for MCI detection. In accordance with the psychiatrist's diagnosis and the patient's score on the Mini-Mental State Examination (MMSE), the true status of the potential patient is labeled. Using the images of the voice spectrogram, we applied the convolutional neural networks (CNN) for MCI classification, where the optimal decision threshold was explored. With the optimal threshold, the precision and recall of patient-based classification are 75 and 82%, respectively.