Leveraging QEEG neuro-biomarkers of alcohol-related visual cognitive impairment for alcohol abuse and dependence diagnosis.

This paper presents a novel approach for leveraging Quantitative Electroencephalography (QEEG) neuro-biomarkers of alcohol-induced impairment of visual memory for alcohol abuse and dependence diagnosis. To achieve this, a spectral filter bank with a wide frequency range (0-100 Hz) is used in conjunction with a spatial filter bank constructed using the Common Spatial Pattern algorithm. We extract a broad set of QEEG features, including power, spectral distribution, and inter-hemisphere functional connectivity, from filtered EEG signals. A total of 1620 QEEG features are extracted from two independent cohorts to demonstrate the generalization ability of the proposed method. Further, Sequential Forward Selection (SFS) with stratified 10-fold cross-validation is used as a wrapper technique to select the subset of features with maximum predictive power, which is determined as 248 and 263 for the two cohorts. SFS was selected for its computational efficiency and effectiveness in optimizing feature subsets within a wrapper-based framework, while mitigating overfitting and preserving model interpretability. The proposed approach outperforms state-of-the-art models, achieving top diagnostic accuracies of 99.63% and 99.25% for the two cohorts using a Support Vector Machine classifier. Our findings reveal that features extracted from the lowest frequencies (delta, theta, and lower alpha bands) and the highest frequencies (higher gamma band) are most discriminative for identifying alcoholic individuals.