In the United States, 5.8 million individuals are currently living with Alzheimer’s and this number is expected to triple to 14 million by 2060 (CDC, 2020). Alzheimer’s disease is an incurable neurodegenerative disease that progressively kills brain cells causing a decline in cognitive function. Most current methods use CT and MRI scans for early detection of Alzheimer’s; however, these methods can be costly and inaccessible to patients. On the other hand, cognitive performance data is quantitative and highly convenient as it involves simple tests, such as reaction time and memory tests, that can be conducted without medical machinery. The goal of this project was to use cognitive performance data to model Alzheimer’s progression by creating an LSTM machine learning model that allows a patient to get regular updates about their Alzheimer’s status. First, data from the ADNI database on patient demographics and cognitive function for the training dataset was compiled. This data was preprocessed by matching patient data and converting dates into time intervals. From here, the data was used in an LSTM model that could predict the future decline in cognitive function of a patient using past cognitive data and time intervals. From here, the training data was used in a kNN that graphs data to diagnose Alzheimer’s and if so, what stage of the disease. The result of my project was a model that was able to predict Alzheimer’s progression with 80% accuracy and correctly identified which stage of Alzheimer’s a patient is in. Keywords: Alzheimer’s, cognitive performance data, machine learning, time progression series

Alzheimer’s disease is a deadly disease with most treatment options depending on when the disease is diagnosed. Currently, early diagnosis methods are expensive and inaccessible to many patients, especially those in poverty.

The goal of this project was to use cognitive performance data to diagnose Alzheimer’s disease by creating an LSTM machine learning model that can be used frequently by a patient to get regular updates about their Alzheimer’s status.

In the United States, 5.8 million individuals are currently living with Alzheimer’s, and this number is expected to triple to 14 million by 2060 (CDC, 2020). Alzheimer’s disease is an incurable neurodegenerative disease that progressively kills brain cells and causes memory loss, along with degradation in cognitive functions. Early detection of the disease allows for prolongation of life and the creation of better treatment options (Nawaz et al., 2020). Most current methods use CT and MRI scans to detect Alzheimer’s as early and as accurately as possible; however, these methods can be unaffordable to the patient and inaccessible. On the other hand, reaction time and other cognitive measures are some of the first indicators of Alzheimer’s progression and can be easily measured and acquired in comparison to other methods of tracking the progression of Alzheimer’s, such as MRI scans, CT scans, or tracking cerebrospinal fluid (Christ et al., 2018).

The thought behind the overall layout of the model was to first look at creating the LSTM portion inputting initial parameters, such as demographic data and previous function values. The LSTM model provided a prediction of the new values for function, using time as a basis for its parameters that allowed for more accurate predictions. These values were then inputted into a kNN, which used these values to predict the MMSE score of the person, based on their functional values. The MMSE score was then used to classify which stage of Alzheimer’s a patient was in, based on what portion of the range of 1-30 that the MMSE score falls in. Below is a figure that outlines the complete model architecture.