Want to Read saving…. Want to Read Currently Reading Read. Other editions. Enlarge cover. Error rating book. Refresh and try again. Open Preview See a Problem? Details if other :.
Thanks for telling us about the problem. Return to Book Page. Parker Editor ,.
Philip M. Parker Editor. This book has been created for patients who have decided to make education and research an integral part of the treatment process. Although it also gives information useful to doctors, caregivers and other health professionals, it tells patients where and how to look for information covering virtually all topics related to narcolepsy also Gelineau's Syndrome; narcoleptic This book has been created for patients who have decided to make education and research an integral part of the treatment process.
Although it also gives information useful to doctors, caregivers and other health professionals, it tells patients where and how to look for information covering virtually all topics related to narcolepsy also Gelineau's Syndrome; narcoleptic syndrome; paroxysmal sleep; sleep epilepsy , from the essentials to the most advanced areas of research.
Online Submission Information
The title of this book includes the word official. This reflects the fact that the sourcebook draws from public, academic, government, and peer-reviewed research. Selected readings from various agencies are reproduced to give you some of the latest official information available to date on narcolepsy. Given patients' increasing sophistication in using the Internet, abundant references to reliable Internet-based resources are provided throughout this sourcebook. Where possible, guidance is provided on how to obtain free-of-charge, primary research results as well as more detailed information via the Internet.
E-book and electronic versions of this sourcebook are fully interactive with each of the Internet sites mentioned clicking on a hyperlink automatically opens your browser to the site indicated. Hard-copy users of this sourcebook can type cited Web addresses directly into their browsers to obtain access to the corresponding sites.
In addition to extensive references accessible via the Internet, chapters include glossaries of technical or uncommon terms. Get A Copy. Paperback , pages. More Details Other Editions 1. Friend Reviews. To see what your friends thought of this book, please sign up.
The Official Patient's Sourcebook on Narcolepsy by James N. Parker
To ask other readers questions about The Official Patient's Sourcebook on Narcolepsy , please sign up. Lists with This Book. The study is ongoing, and dates to The PSGs in subjects were used for training, while randomly selected PSGs were used for validation testing of the sleep stage scoring algorithm and narcolepsy biomarker training. A detailed description of the sample can be found in Young et al.
The sample does not contain any T1N patients, and the three subjects with possible T1N were removed The cohort contains thousands of PSG recordings, but for this study we used diagnostic no positive airway pressure recordings in independent patients that have been used in prior studies This subset contains patients with a range of different diagnoses including: sleep disordered breathing , insomnia , REM sleep behavior disorder 4 , restless legs syndrome 23 , T1N 25 , delayed sleep phase syndrome 14 and other conditions Description of the subsample can be found in Andlauer et al.
The randomly selected subjects were used for training the neural networks, while randomly selected PSGs were kept for validation testing of the sleep stage scoring algorithm. These subjects were also used for training the narcolepsy biomarker algorithm. A total of 26 subjects were removed from the study—4 due to poor data quality, and the rest because of medication use. The Korean Hypersomnia Cohort is a high pretest probability sample for narcolepsy. These PSGs were used for testing the sleep scoring algorithm and for training the narcolepsy biomarker algorithm.
No data were used for training the sleep scoring algorithm. Detailed description of the sample can be found in Hong et al. Two subjects were removed from the narcolepsy biomarker study because of poor data quality. Patients in this cohort were examined at the Innsbruck Medical University in Austria as described in Frauscher et al. The rest of the sample has idiopathic hypersomnia and type 2 narcolepsy. Almost all subjects had two sleep recordings performed, which were kept together such that no two recordings from the same subject were split between training and testing partitions.
As Rosenberg and Van Hout 16 have shown, variation between individual scorers can sometimes be large, leading to an imprecise gold standard. To quantify this, and to establish a more accurate gold standard, 10 scorers from 5 different institutions, University of Pennsylvania, St. For this study, scoring data from University of Pennsylvania, St.
The Official Patient's Sourcebook on Narcolepsy
This allowed for a much more precise gold standard, and the inter-scorer reliability could be quantified for a dataset, which could also be examined by automatic scoring algorithms. Detailed description of the sample can be found in Kuna et al. The sample does not contain any T1N patients. The few patients included are those with clear and frequent cataplexy a requirement of the trial who had no stimulant or antidepressant treatment at baseline All seven subjects in this sample were used exclusively for training the narcolepsy biomarker algorithm.
As non-T1N cases with unexplained daytime somnolence, the cohort includes 77 other patients: 19 with idiopathic hypersomnia, 7 with type 2 narcolepsy and normal CSF hypocretin-1, 48 with a subjective complaint of excessive daytime sleepiness not confirmed by MSLT and 3 with secondary hypersomnia. Patients in this cohort were examined at the Rigshospitalet, Glostrup, Denmark, as described in Christensen et al.
All 79 subjects in this cohort were used exclusively for training the narcolepsy biomarker algorithm. The FHC was used as data for the replication study of the narcolepsy biomarker algorithm. Design of this dataset is described in Rosenberg and Van Hout In doing so, technicians assign each epoch with a discrete value. With a probabilistic model, like the one proposed in this study, a relationship to one of the fuzzy sets is inferred based on thousands of training examples labeled by many different scoring technicians. The hypnodensity graph refers to the probability distribution over each possible stage for each epoch, as seen in Fig.
This allows more information to be conveyed, since every epoch of sleep within the same stage is not identical. For comparison with the gold standard, however, a discrete value must be assigned from the model output as:. This means that when multiple sleep stages are represented, more than half of the epoch may not match the assigned label.
This is evident in the fact that the label accuracy decreases near transition epochs One solution to this problem is to remove transitional regions to purify each class. However, this has the disadvantage of under-sampling transitional stages, such as N1, and removes the context of quickly changing stages, as is found in a sudden arousal. This also assumes that the noise is randomly distributed with an accurate mean—a bias cannot be canceled out, regardless of the amount of training data. For these reasons, all data including those containing sleep transitions were included.
- Formal Logical Methods for System Security and Correctness.
- 101 Ways to Love Your Job!
- International Monetary Economics, 1870-1960: Between the Classical and the New Classical (Historical Perspectives on Modern Economics).
Biases were evaluated by incorporating data from several different scoring experts cohorts and types of subjects. A full-night PSG involves recording many different channels, some of which are not necessary for sleep scoring Poor electrode connections are common when performing a PSG analysis. This can lead to a noisy recording, rendering it useless. These were then log-transformed, averaged and compared with a previously established multivariate distribution, based on the WSC 32 , 33 and SSC 10 , 32 training data.
The channel with lowest Mahalanobis distance 57 to this distribution was selected. Additionally, all channels were filtered with a fifth-order two-direction infinite impulse response IIR high-pass filter with cut-off frequency of 0. All steps of the pre-processing are illustrated in Fig.
Convolutional neural networks CNNs are a class of deep learning models first developed to solve computer vision problems A CNN is a supervised classification model in which a low level, such as an image, is transformed through a network of filters and sub-sampling layers. Each layer of filters produces a set of features from the previous layer, and as more layers are stacked, more complex features are generated.
This network is coupled with a general-purpose learning algorithm, resulting in features produced by the model reflecting latent properties of the data rather than the imagination of the designer. This property places fewer constrictions on the model by allowing more flexibility, and hence the predictive power of the model will increase as more data are observed. This is facilitated by the large number of parameters in such a model, but may also necessitate a large amount of training data.
Sleep stage scoring involves a classification of a discrete time series, in which adjacent segments are correlated. Models that incorporate memory may take advantage of this and may lead to better overall performance by evening out fluctuations. However, these fluctuations may be the defining trait or anomaly of some underlying pathology such as narcolepsy, a pathology well known to involve abnormal sleep stages transitions , present in only a fraction of subjects, and perhaps absent in the training data.
This can be thought of similarly to a person with a speech impediment: the contextual information will ease the understanding, but knowing only the output, this might also hide the fact that the person has such a speech impediment. To analyze the importance of this, models with and without memory were analyzed. Memory can be added to such a model by introducing recurrent connections in the final layers of the model.
- narcolepsy current treatment: Topics by cordkako.tk.
- Related Articles?
- Idiopathic Hypersomnia: A Study of 77 Cases.
- Data structures and Program Design in C++.
This turns the model into a recurrent neural network RNN. Classical RNNs had the problem of vanishing or exploding gradients, which meant that optimization was very difficult.
This problem was solved by changing the configuration of the simple hidden node into a LSTM cell Models without this memory are referred to as FF models. A more in-depth explanation of CNNs including application areas can be found in the review article on deep learning by LeCun et al. For a more general introduction to machine learning concepts, see the textbook by Bishop Biophysical signals, such as those found in a PSG, inherently have a low signal to noise ratio, the degree of which varies between subjects, and hence learning robust features from these signals may be difficult.
To circumvent this, two representations of the data that could minimize these effects were selected. An example of each decomposition is shown in Fig. Neural network strategy. These processed data are fed into the neural networks in one of the two formats. The data in the octave encoding are offset for visualization purposes. Color scale is unitless. Octave encoding maintains all information in the signal, and enriches it by repeatedly removing the top half of the bandwidth i.
At no point is a high-pass filter applied. Instead, the high frequency information may be obtained by subtracting lower frequency channels—an association the neural networks can make, given their universal approximator properties After filtration, each new channel is scaled to the 95th percentile and log modulus transformed:. Very large values, such as those found in particularly noisy areas, are attenuated greatly. Some recordings are noisy, making the 95th percentile significantly higher than what the physiology reflects.