Automated Analysis of Clock Drawing Test for Differential Diagnosis of Cognitive Decline and AD

RUSSELL L. BINACO, SEAN MCGUIRE, JACOB R. EPIFANO, and NICHOLAS J. CALZARETTO

Early and accurate diagnosis of Alzheimer’s disease is an unsolved problem. Practitioners have been seeking a noninvasive diagnostic tool to assess a patient’s level of cognitive impairment. The clock drawing test is one such tool, where patients are asked to draw an analog clock showing the time 10 past 11. Traditionally the clock test, combined with a battery of other neuropsychological tests, is evaluated by a neuropsychologist in order to reach a diagnosis. However, this is a time consuming as well as subjective process, with a significant rate of misdiagnosis. Researchers are now looking at machine learning algorithms to help form a relationship between the various features obtained from the clock drawing test and a degree of cognitive impairment. To do so, we extracted hundreds of features from the clock drawing test using a smart pen and tablet. We then used information theoretic feature selection to determine the most relevant features, and used them to train a neural network type classifier. The classifier is used to analyze these features and identify each patient as SCI (Subtle Cognitive Impairment), MCI (Mild Cognitive Impairment), or AD (Alzheimer's Disease). Preliminary results with a non-optimized classifier indicate a differential diagnosis accuracy of mid 70% to low 80% are achievable. These results, if further improved upon, can allow general health practitioners to use clock drawing test as a preliminary screening tool and recommend patients for further neuropsychological testing if warranted.