[Research of Partial Least Squares Decoding Method for Motion Intent].

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Neural ensemble signals from microelectrode arrays are sparse, noisy, and contain redundant information.
• This noise and redundancy reduce the stability and precision of decoding motion intent.
• Existing feature extraction methods may not effectively handle noise and information redundancy.

Purpose of the Study:

• To propose and evaluate a novel decoding method for motion intent using partial least squares (PLS) feature extraction.
• To enhance the stability and precision of neural decoding by addressing noise and redundant information.
• To compare the performance of the proposed PLS-based method against common feature extraction techniques.

Main Methods:

• Extracted features from neural spike signals using Partial Least Squares (PLS).
• Classified the extracted features using a Support Vector Machine (SVM) classifier.
• Decoded motion intent from neural ensemble signals recorded during a plus-maze task.

Main Results:

• The proposed PLS combined with SVM classification model demonstrated improved stability and higher decoding accuracy compared to traditional methods.
• The PLS method effectively reduced noise by extracting fewer, more informative features.
• Achieved decoding accuracies of 93.59%, 84.00%, and 83.59% on real datasets.

Conclusions:

• The PLS feature extraction method, when combined with SVM classification, offers a robust solution for decoding motion intent from noisy neural data.
• This approach overcomes the limitations of PLS regression in handling accumulated noise effects.
• The method provides a more stable and precise decoding of neural ensemble signals for brain-computer interfaces.