Evaluation of DYSLEXPERT: A comparison of a knowledge-based system with experienced clinicians in the diagnosis of dyslexia

摘要

DYSLEXPERT is a knowledge-based system designed to support clinicians in the diagnosis of dyslexia. The system answers four questions that usually have to be addressed by human clinicians: Which areas of psychological functioning appear to be problematic? What tests would provide the relevant information? Are we actually dealing with dyslexia? How can the reading and spelling problems best be alleviated? In the present paper, we compare the competence of the program to the performance of four experienced diagnosticians on 20 cases of dyslexia and 20 cases of nondyslexia.The evaluation of the competence of DYSLEXPERT was based on three sets of data: the original case work, a reanalysis of the cases, and the results produced by DYSLEXPERT. The agreement between human and computer performance was measured using Cohen's kappa and Dice's measure of specific agreement. Statistical testing revealed the agreement between DYSLEXPERT and the reanalysis to be similar to the agreement between the original case work and the reanalysis.