Title

Design of a Program for Automatic Detection of Stroke-Induced Aphasia

School Name

Spring Valley High School

Grade Level

10th Grade

Presentation Topic

Computer Science

Presentation Type

Non-Mentored

Abstract

It is critical for people experiencing stroke symptoms to be admitted to the hospital immediately to receive proper treatment. One way to ensure timely treatment is to automate the detection of aphasia, a symptom of stroke. The goal of this project was to develop a machine learning system based on Support Vector Machines (SVM) for detecting aphasia using features extracted from speech transcripts. The system was coded in Python using the scikit-learn library. The system was trained and tested with a dataset from the C-STAR Aphasia Lab, consisting of speech transcripts of responses for three different prompts (requiring the patients to describe a scenario) from 80 participants with differing levels of aphasia and without aphasia. The average aphasia detection accuracy of the system was 85-88% when only responses for each individual prompt were considered, while the accuracy of the system went up to 95% when the responses of all the prompts together were used. The inaccuracy is mainly due to false positives, which decreased significantly with all prompts combined. While these results are based on responses to prompts in a clinical setting as opposed to daily living, they show promise towards the creation of automatic at-home aphasia detection systems.

Location

HSS 206

Start Date

4-2-2022 10:45 AM

Presentation Format

Oral and Written

Group Project

No

COinS
 
Apr 2nd, 10:45 AM

Design of a Program for Automatic Detection of Stroke-Induced Aphasia

HSS 206

It is critical for people experiencing stroke symptoms to be admitted to the hospital immediately to receive proper treatment. One way to ensure timely treatment is to automate the detection of aphasia, a symptom of stroke. The goal of this project was to develop a machine learning system based on Support Vector Machines (SVM) for detecting aphasia using features extracted from speech transcripts. The system was coded in Python using the scikit-learn library. The system was trained and tested with a dataset from the C-STAR Aphasia Lab, consisting of speech transcripts of responses for three different prompts (requiring the patients to describe a scenario) from 80 participants with differing levels of aphasia and without aphasia. The average aphasia detection accuracy of the system was 85-88% when only responses for each individual prompt were considered, while the accuracy of the system went up to 95% when the responses of all the prompts together were used. The inaccuracy is mainly due to false positives, which decreased significantly with all prompts combined. While these results are based on responses to prompts in a clinical setting as opposed to daily living, they show promise towards the creation of automatic at-home aphasia detection systems.