Feasibility of using an automated analysis of formulation effort in patients’ spoken seizure descriptions in the differential diagnosis of epileptic and nonepileptic seizures

AbstractObjective There are three common causes of Transient Loss Consciousness (TLOC), syncope, epileptic and psychogenic nonepileptic seizures (PNES). Many individuals who have experienced TLOC initially receive an incorrect diagnosis inappropriate treatment. Whereas syncope can be distinguished relatively easily with a small number "yes"/"no" questions, the differentiation other two is more challenging. Previous qualitative research based on methodology Conversation Analysis has demonstrated that descriptions contain formulation effort than accounts PNES. This investigates whether features likely to reflect level automatically elicited from audio recordings transcripts speech used differentiate between seizures. Method Verbatim conversations patients neurologists were manually produced video 45 interactions (21 epilepsy 24 PNES). The subsection each transcript containing person's account their first seizure was extracted for analysis. Seven detectable designed as markers effort. These train Random Forest machine learning classifier. Result significantly hesitations repetitions in Using nested leave-one-out cross validation approach, 71% correctly classified by Discussion pilot study provides proof principle linguistic been could distinguish PNES thereby contribute differential TLOC. Future should explore additional observations incorporated into diagnostic stratification tool compare performance these when they combined information provided witnesses about manifestations medical history.