An aphasia database on the internet: a model for computer-assisted analysis in aphasiology

Area of Science:

• Neuroscience
• Computer Science
• Medical Informatics

Background:

• Aphasiology research often involves complex data analysis.
• Educating medical and engineering students on data mining in aphasiology requires accessible tools.

Purpose of the Study:

• To develop a web-based software model for data mining in aphasiology.
• To educate students on data mining techniques using aphasia patient data.
• To classify aphasia types using artificial neural networks and patient data.

Main Methods:

• A database of 254 aphasic patients was compiled, including diagnosis, Aachen Aphasia Test (AAT) profiles, and clinical information.
• Cerebral lesion data from 147 patients were standardized onto a 3D reference brain using the ACPC coordinate system.
• Two artificial neural networks were employed for aphasia type classification, with feature selection enhancing diagnostic accuracy.

Main Results:

• A coarse classification of aphasia type based on spontaneous speech achieved 87% accuracy.
• Selecting four key features from 30 available AAT features improved diagnostic accuracy to 92%.
• The nearest-neighbor method facilitated interindividual comparison of neurolinguistic and lesion data.

Conclusions:

• The developed Aphasia Diagnoser software model effectively utilizes data mining for aphasia classification.
• The model supports both group study (neural network) and case study (nearest-neighbor) approaches in aphasiology.
• This web-based tool promotes discussion on the reliability and potential of data mining in understanding aphasia.