Artificial intelligence for brain-to-speech decoding in paralysis: a systematic review.

Area of Science:

• Neuroscience and Artificial Intelligence
• Biomedical Engineering
• Rehabilitation Technology

Background:

• Communication loss is a significant challenge for individuals with paralysis.
• Brain-computer interfaces (BCIs) combined with artificial intelligence (AI) offer a potential solution for restoring speech.
• Decoding speech from brain signals is an active area of research with various approaches.

Purpose of the Study:

• To systematically review the use of AI in decoding speech from brain signals using both invasive and non-invasive neural interfaces.
• To evaluate the performance, methodologies, and quality of existing studies in this field.
• To identify key challenges and propose a framework for future research and clinical application.

Main Methods:

• Systematic literature review following PRISMA guidelines, analyzing 115 studies (2019-2025).
• Data extraction on acquisition protocols, signal preprocessing, and AI architectures.
• Quality assessment using QUADAS-2, with a focus on bias and validation in paralyzed populations.

Main Results:

• Invasive BCIs showed higher median classification accuracy (77.7%) than non-invasive methods (73.0%), though task complexity and evaluation paradigms differ.
• Hybrid CNN/RNN architectures and transformers outperformed traditional AI models.
• Significant limitations were found in study quality, with high risk of selection bias (62.6%) and limited validation in paralyzed individuals (5.2%).
• No non-invasive study demonstrated functional speech decoding in paralyzed populations, highlighting a critical translational gap.

Conclusions:

• While AI-powered BCIs show feasibility for speech decoding, particularly invasive methods, substantial validation and quality improvements are needed.
• The lack of non-invasive functional speech decoding in paralyzed individuals is a priority for future research.
• A decision framework is proposed to guide future development considering accuracy, cost, and clinical applicability.