This work presents a new method for detection of target non-apneic arousals applying deep neural network architecture on selected physiological signals. The proposed architecture merges the convolutional layers for feature extraction with long-short term memory layers for temporal combination of features. It is noteworthy that, for each recording, one such optimal convolutional recurrent neural networks classifier is developed while training and all of these classifiers are used for prediction based on majority vote decision for detecting arousals. When evaluated with 2018 PhysioNet/CinC Challenge dataset, the experimental outcomes demonstrate overall AUROC and AUPRC scores of 0.65±0.15 and 0.60±0.05 respectively for the training data.