Comparison of machine learning methods for classifying aphasic and non-aphasic speakers.

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Aphasia classification is crucial for diagnosis and treatment.
• Machine learning offers potential tools for analyzing complex language data in aphasia.

Purpose of the Study:

• To evaluate the performance of eight machine learning classifiers on three distinct aphasia-related classification tasks.
• To determine if a universally optimal classifier exists for aphasia data analysis.

Main Methods:

• Comparison of eight machine learning classifiers using three datasets: Philadelphia Naming Test data, Finnish Boston Naming Test data (Alzheimer's vs. vascular disease), and Aachen Aphasia Test data.
• Artificial data generation for smaller datasets to augment original naming data.
• Experimental testing of classifiers on confrontation naming and aphasia syndrome data.

Main Results:

• Classifiers performed successfully on the first (aphasic vs. non-aphasic) and third (aphasic syndromes) datasets.
• Less encouraging results were observed with the second dataset (Alzheimer's vs. vascular disease).
• No single machine learning classifier demonstrated superior performance across all tested aphasia datasets.

Conclusions:

• Machine learning classifiers can be effective for certain aphasia classification tasks.
• The choice of classifier should be tailored to the specific characteristics of the aphasia dataset being analyzed.
• Further research may be needed to optimize classifier performance for challenging datasets like distinguishing between Alzheimer's and vascular dementia based on naming data.