Machine Learning and Digital Sketch Recognition Methods to Support Neuropsychological Diagnosis and Identification of Cognitive Decline
Abstract
With approximately 15 to 20 percent of adults aged 65 and older living with Mild Cognitive
Impairment (MCI), researchers in neuropsychology have placed increasing emphasis in early detection to best preserve quality of life for MCI patients before the turnover to full Alzheimer’s or similar types of dementia. Current methods of diagnosis in clinical neuropsychology involve a
lengthy process of developing a full behavioral profile, in which tests are administered to examine the existence and extent of cognitive impairment. Efforts to digitize and semi-automate the process have emerged, but solutions tend to be limited in scope and there exists ample room for improvement in developing new and faster diagnosis techniques.
This dissertation presents digital neuropsychological diagnosis tools by adapting existing neuropsychological tests or presenting new prototypes based on existing methodology. Digitizing these examinations allows for the collection of high-granularity data that allows for the creation of machine learning algorithms as well as automated graders designed to be supported with various intuitions about the test-taker’s behavior in a manner consistent with existing literature in neuropsychology. We sought to produce knowledge on the mapping of sketch recognition analysis to identifying MCI in patients. On a technical level, we present novel sketch recognition techniques that identify specific degrees of shape correctness for automated grading of complex-figure tests, as well as sketch segmentation techniques specific to the TMT.
To that end, we present three digitized systems that collect fine-grained behavioral usage data
through cognitive examinations: 1) a digital TMT that simulates the paper-and-pencil experience
on a tablet, 2) an automated grader for digitized ROCF tests, and 3) a new examination that
leverages modern touch tablet technology to expand on the inherent diagnosis weaknesses of
paper-based examinations as patients manipulate a virtual 3-dimensional object. We present our
technical approach, results from experiments, expected goals, and the intellectual contributions of this dissertation research.
Our studies produced over 350 digital sketches from neuropsychological examinations across
our three presented systems. Data segmentation techniques have resulted in over 6000 pieces of
segmented sketch data. We believe the findings presented in this dissertation can provide valuable insight into the process of developing semi-automated or fully automated neuropsychological diagnosis tools.
Subject
Machine LearningHuman-Computer Interaction
Digital Sketch Recognition
Applied computing
Health care information systems
Gestural Input
Intelligent User Interfaces
Mobile devices
Cognitive decline
Activity recognition
Citation
Lara Garduno, Raniero A (2021). Machine Learning and Digital Sketch Recognition Methods to Support Neuropsychological Diagnosis and Identification of Cognitive Decline. Doctoral dissertation, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /195764.