No abstract

In this work, a dense recurrent convolutional neural network (DRCNN) was constructed to detect sleep disorders including arousal, apnea and hypopnea using Polysomnography (PSG) measurement channels provided in the 2018 Physionet challenge database. Our model structure is composed of multiple dense convolutional units (DCU) followed by a bidirectional long-short term memory (LSTM) layer followed by a softmax output layer. The sleep events including sleep stages, arousal regions and multiple types of apnea and hypopnea are manually annotated by experts which enables us to train our proposed network using a multi-task learning mechanism. Three binary cross-entropy loss functions corresponding to sleep/wake, target arousal and apneahypopnea/normal detection tasks are summed up to generate our overall network loss function that is optimized using the Adam method. Our model performance was evaluated using two metrics: the area under the precision-recall curve (AUPRC) and the area under the receiver operating characteristic curve (AUROC). To measure our model generalization, 4-fold cross-validation was also performed. For training, our model was applied to full night recording data. Finally, the average AUPRC and AUROC values associated with the arousal detection task were 0.505 and 0.922, respectively on our testing dataset. An ensemble of four models trained on different data folds improved the AUPRC and AUROC to 0.543 and 0.931, respectively. Our proposed algorithm achieved first place in the official stage of the 2018 Physionet challenge for detecting sleep arousals with AUPRC of 0.54 on the blind testing dataset.

In this work, a dense recurrent convolutional neural network (DRCNN) was constructed to detect sleep arousals using available Polysomnography (PSG) measurement channels provided in the 2018 Physionet challenge database. Our model structure is composed of multiple dense convolutional units (DCU) followed by a bidirectional long-short term memory (LSTM) layer followed by a softmax output layer. The sleep events including sleep stages, arousal regions and multiple types of apnea-hypopnea/normal are manually annotated in 2018 Physionet challenge database which enable us to train our proposed network using a multi-task learning mechanism. Three binary cross-entropy loss functions corresponding to sleep/wake, arousal presence/absence and apnea/hypopnea presence/absence detections are summed up to generate our overall network loss function that is optimized using the Adam method. Our model performance was evaluated using two metrics: the area under the precision-recall curve (AUPRC) and the area under the receiver operating characteristic curve (AUROC). To measure our model generalization, 4-fold cross-validation was also performed. For training, full night recording data was applied to our model. Finally, our proposed algorithm achieves the first place in the official stage of the Physionet challenge with AUPRC of 0.54 on the blind testing dataset.

In this paper, we compare two low cost time-domain tracking algorithms based on passive acoustics. The problem consists in tracking an unknown number of sperm whales (Physeter catodon). Clicks are recorded on two datasets of 20 and 25 minutes on an open-ocean widely-spaced bottom-mounted hydrophone array. The output of the method is the track(s) of the Marine Mammal(s) (MM) in 3D space and time. Firstly, we briefly review studies of the Stochastic Matched Filter (SMF) detector and its performances with a reflected click cancellation, the Teager-Kaiser-Mallat (TKM) filtering, the source separation methods and the main characteristics of MM signals. Then, we propose a real-time algorithm for MM transient call localization. We also recall the Cramér-Rao Lower Bound (CRLB) Kay (1993) and the confidence ellipses theory to predict the reachable accuracy and compare it to the tracking results. In Section 3 we show and compare results of track estimates with results from specialized teams and compare SMF versus TKM localization. Then, the system is evaluated with the confidence ellipses on the trajectories. Finally, we discuss on the possible dynamic behavior of the whale that these localizations offer, like hunting and foraging strategies. This paper deals with the 3D tracking of MM using a widely-spaced bottom-mounted hydrophone array in deep water. It focuses on sperm whale clicks. There were previous algorithms developed in the state of the art Giraudet & Glotin (