Abstract Details

Title Towards Machine Learning-Based Cognitive Examination

Topic Behavioral and Cognitive Neurology

Presentation(s) S2 - Behavioral Neurology (2:48 PM-3:00 PM)

Poster/Presentation
Number 010

Objective The objectives of this study is to create the foundation for an automated computerized cognitive examination. Further, the second objective was to train subcomponents of the cognitive examination to be graded by deep-learning methods.

Background With the burgeoning population of patients with cognitive disorders such as dementia, healthcare is already having significant difficulty in caring for this new population of patients. However, with the last decade’s advances in neural networks, it is possible to begin creating software which may aid in healthcare for these patients.

Design/Methods A completely custom cognitive examination was coded in Python 3.7. A graphic user interface was created, and questions were adapted from the Addenbrooke Cognitive Examination-3. Back-end software was coded to utilize the answers provided for each question. The answers were graded according to Addenbrooke Cognitive Examination-3 guidelines with custom-coded Python-based software depending on the nature of each question. Methods used to grade inputs included: natural language processing, custom computer vision neural network, object-collision detection, and simple comparison of inputs to a pre-defined list of acceptable inputs. 10 healthy subjects under the age of 50 were recruited to take the examination to assess the quality.

Results All questions performed highly with no bugs in software resulting in false elevation or depression of score. However, one question failed to appropriately score all patients, resulting in false lowering of all scores by 1/100. Overall, none of the healthy individuals scored within known range for cognitive impairment, and the outcomes of the healthy individuals are comparable to outcomes on paper-based tests for similar cohorts.

Conclusions The work presented here demonstrates automated cognitive examinations may assist with healthcare for patients with cognitive disorders, having impact upon the fields of neurology, psychiatry, and family medicine. Future goals include the development of machine-learning driven approach to analyzing cognitive examination scores.

Authors/Disclosures
Calvin W. Howard, MD (Calvin Howard) PRESENTER	Dr. Howard has or had stock in CogNet.Dr. Howard has received intellectual property interests from a discovery or technology relating to health care.

��ɫ��

��ɫ��