Beyond neuropsychological tests: AI speech analysis in PKU

Area of Science:

• Neuroscience
• Metabolic Disorders
• Artificial Intelligence

Background:

• Phenylketonuria (PKU) is an inherited metabolic disorder causing toxic phenylalanine (Phe) buildup.
• Traditional neuropsychological tests may not fully capture the cognitive and linguistic impacts of PKU.
• Speech analysis offers a novel avenue for understanding PKU-related challenges.

Purpose of the Study:

• To investigate if AI-driven speech analysis can detect differences in verbal discourse associated with PKU.
• To correlate linguistic features identified by AI with biochemical markers in individuals with PKU.
• To introduce a new measure, Proficiency in Verbal Discourse, for assessing PKU-related cognitive function.

Main Methods:

• Adults with PKU (n=42) and without PKU (n=41) completed the Cookie Theft Picture Task for spontaneous speech.
• AI analyzed speech responses, identifying 23 distinct linguistic features.
• Multidimensional scaling (MDS) aggregated features into a quantifiable metric (Dimension 1) correlated with biomarkers.

Main Results:

• AI analysis revealed significant differences in 23 linguistic features between PKU and non-PKU groups.
• Dimension 1, derived from these features, significantly correlated with biochemical markers in PKU patients.
• ChatGPT analysis of extreme Dimension 1 examples highlighted differences in attention, clarity, cohesion, awareness, and emotion recognition.

Conclusions:

• AI-powered speech analysis can detect subtle verbal discourse characteristics associated with PKU, undetectable by traditional methods.
• Proficiency in Verbal Discourse shows promise as a novel measure for assessing cognitive function and learning disabilities in PKU.
• This AI paradigm may offer new insights for PKU management and be applicable to other rare diseases and clinical trials.