Comprehensive Summary
Afonso et al. studied the role of EEG signals in better Parkinson’s disease (PD) detection by employing the innovative hybrid quorum sensing optimization (HQSO) approach for feature selection. The authors transformed EEG time series into time-frequency images using the Stockwell transform, which provided better results than STFT or CWT methods. Features were learned by convolutional neural networks, they were further filtered through the two-step HQSO process: population-level optimization followed by statistical refinement and correlation pruning. After testing on two publicly available EEG datasets (San Diego and University of New Mexico), the model achieved detection accuracies of 98.09% and 94.96% respectively, while reducing nearly 60% of features, thereby reducing computational complexity. This performance was ahead of or relatively close to existing state-of-the-art models.
Outcomes and Implications
This study highlights the promise of EEG-based diagnostics as a safe, low-cost, and portable alternative to more invasive or costly tools such as imaging and deep brain stimulation. The model shows high diagnostic accuracy with significantly lower feature sets and thus serves to promote real-time wearable systems for PD detection and monitoring. This model would allow such applications to develop for earlier diagnosis, broad clinical and home-care accessibility, as well as a decreased dependence on motor tasks for the patient, which in later stages of PD may pose challenges. The findings lay the groundwork for incorporating optimized EEG-based deep learning tools into routine neurological care in the near future.