Innovative AI Technique Uses Speech Analysis to Detect Early Signs of Neurological Disorders

Researchers led by Professor Li Hai from the Hefei Institutes of Physical Science at the Chinese Academy of Sciences have developed a groundbreaking deep learning framework that enhances both the accuracy and interpretability of identifying neurological disorders through speech. Published in Neurocomputing, this new approach leverages advanced artificial intelligence to analyze voice recordings, revealing early symptoms of conditions such as Parkinson's, Huntington's, and Wilson disease.

The core idea is that subtle changes in speech can serve as warning signals of underlying brain health issues. Since many neurodegenerative diseases initially manifest as speech abnormalities like dysarthria, voice signals provide a noninvasive method for early screening and ongoing monitoring.

Traditional speech analysis methods often depend heavily on manually crafted features and struggle with modeling complex temporal interactions, which limits their effectiveness and transparency. To overcome these challenges, the research team introduced the Cross-Time and Cross-Axis Interactive Transformer (CTCAIT), a sophisticated model designed for multivariate time series data.

This framework first employs a large-scale audio model to extract high-dimensional temporal features, creating multidimensional embeddings that represent speech signals across time and feature axes. Subsequently, it utilizes an Inception Time network to identify multi-scale and multi-level patterns within the speech data. By integrating multi-head attention mechanisms across time and channels, CTCAIT can detect pathological speech patterns that span different dimensions.

The model achieved remarkable detection accuracy—92.06% on a Mandarin Chinese dataset and 87.73% on an external English dataset—showing its strong potential for cross-linguistic applications. Additionally, the researchers conducted interpretability analyses to understand the model’s decision-making process better and compared various speech tasks to assess their efficacy for clinical implementation.

This innovative approach offers a promising tool for the early detection and continuous monitoring of neurological conditions, potentially enabling timely intervention and better patient outcomes.

For more details, see the original publication in Neurocomputing. This research was conducted by Zhang Zhenglin and colleagues and supported by the Chinese Academy of Sciences.

Source: https://medicalxpress.com/news/2025-07-ai-speech-early-neurological-disorders.html