An Integrated Approach to Non-Invasive Diagnosis of Dementia Using Natural Language Processing and Machine Learning
Aniket Dey, Sanam Mittal · 2022 IEEE 2nd International Conference on Data Science and Computer Application (ICDSCA) · 2022
Current diagnosis of dementia revolves around the usage of lengthy clinical tests such as brain scans and blood tests that are also often invasive for the elderly populations that cognitive impairment is common to. A rising trend in care for dementia is the usage of qualitative tests like neuropathological and cognitive assessments to serve as preliminary diagnosis, especially since these do not require invasive procedures and instead are accessible to delicate individuals. Machine learning techniques have slowly gained traction within healthcare, especially as algorithms have the unique ability to process a large amount of quantitative and qualitative data to simple and often binary outputs. Machine learning algorithms are thus particularly suited to the task of diagnosis of dementia. This paper aims to provide a qualitative and non-invasive method for diagnosis of dementia in patients and utilizes a combination of machine learning and natural language processing (NLP). All data for testing and training of the model was sourced from the Behavioral Risk Factor Surveillance System (BRFSS), an annual survey assessment conducted by the CDC. In our proposed model, a conversational chatbot collects data from individuals surrounding multiple qualitative predictors, and these values are then fed into a machine learning algorithm to provide an output for presence of dementia. The strongest performing algorithm was a random forest algorithm, which outperforms the current state-of-the-art, with an AUC-PR score of 0.832 and an MCC score of 0.53. Models like those proposed by our paper have the potential to revolutionize effective care for patients facing dementia, making diagnosis much quicker and easier to access.