An auxiliary diagnostic systematic for dysarthria in stroke patients based on spectrogram acoustic characteristics

Abstract

With the increasing prevalence of post-stroke dysarthria (PSD), the conventional diagnostic procedures, which mostly rely on subjective manual assessments and outdated computational methodologies, have become insufficient. Traditional methods are inadequate in meeting the current need for fast, thorough, and unbiased assessments. This study presents a new computer-aided diagnostic system that combines spectrogram analysis with the RcpViT-X model. The RcpViT-X model is a modified version of the lightweight convolutional neural network, based on the Repvit architecture and improved using the Xception framework. This integration is a component of a methodical processing workflow that greatly improves productivity and ensures effective production of results. This novel approach overcomes the constraints of conventional PSD diagnostic methods, providing a more effective and resilient option for diagnosing post-stroke dysarthria.