A Speech-based classification model for Mild Cognitive Impairment screening

Abstract

Mild Cognitive Impairment (MCI) is a condition that often presents symptoms similar to those of normal aging, which makes distinguishing between the two challenging. While some cognitive decline is expected as people age, MCI involves more noticeable memory and thinking difficulties that do not yet interfere significantly with daily life. Depending on its underlying cause, MCI may progress to dementia, a neurodegenerative condition that leads to a significant decline in cognitive function and quality of life. Dementia has become a global public health issue, with increasing numbers of cases due to aging populations worldwide. Early intervention at the MCI stage is critical, as it provides an opportunity to slow down or possibly prevent the progression of dementia, improving patient outcomes and reducing the burden on healthcare systems. Various cognitive screening tools are currently used in clinical settings to assess cognitive function and detect early signs of impairment, but they have exhibited several challenges. This study leveraged Rapid Application Development due to its flexibility and iterative structure to develop an effective model for Mild Cognitive Impairment Screening. The study used an open-source database containing audio samples from MCI patients and healthy controls to build the model. Consequently, the study aimed to improve classification performance by extracting universal features such as Mel-Frequency Cepstral Coefficients (MFCC), jitter, shimmer, and fundamental frequency. Various models were trained, including Random Forest, convolutional neural network (CNN), Convolutional Neural Network, Long Short-Term Memory (CNN-LSTM) with Deep Neural Network (DNN), achieving the highest accuracy of 84%. This research demonstrated that universal features can effectively support the early detection of Mild Cognitive Impairment using deep learning, offering a non-invasive and scalable screening alternative for clinical settings. It also provided a foundation for future research into speech biomarkers for cognitive disorders and encourages the integration of machine learning in health technology applications. Based on the results, further improvements are recommended, such as exploring additional audio features and applying transfer learning to enhance the model’s robustness. Keywords: Mild Cognitive Impairment, Speech Analysis, Speech-Based Detection, Early Diagnosis, Classification Model