The presence of electroencephalogram (EEG) patterns can provide valuable information about the degree of neurological recovery of comatose patients hospitalised after cardiac arrest. Signal processing and machine learning-based automated systems can predict the chances of a patient's consciousness recovery. This work presents a novel method to classify EEG records for predicting such prognosis using TQWT-based signal refinement and Tensor decomposition-based feature extraction to generate good or poor outcome labels with probability. TQWT-based signal reconstruction with selected sub-bands obeying relevant statistical constraints can be a promising approach to enhance critical care patterns in EEG signals. The optimal TQWT settings and statistical properties are obtained to emphasise critical care EEG patterns like spikes and a variety of transient-like patterns in patients suffering cardiac arrest-induced coma. Spectrograms, capturing time-frequency based information, are used to create a 3-way tensor per record. The core tensor resulting from Tucker decomposition of the formed tensor is then used to get 1-D feature vectors for learning of Random Forest Classifier. Evaluation of the proposed methodology on The George B. Moody PhysioNet Challenge 2023 dataset obtained an average F-measure of 0.76 and a challenge score of 0.33 for 10-fold cross-validation. We participated in the challenge as team ‘Medics' and obtained a challenge score of 0.54 on challenge hidden test data.