FORM 2
THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENT RULES, 2003
COMPLETE SPECIFICATION (See Section 10 and Rule 13)
Title of invention:
METHOD AND SYSTEM FOR MONITORING SLEEP AROUSAL
Applicant
Tata Consultancy Services Limited A company Incorporated in India under the Companies Act, 1956
Having address:
Nirmal Building, 9th floor,
Nariman point, Mumbai 400021,
Maharashtra, India
Preamble to the description
The following specification particularly describes the invention and the manner in which it is to be performed.

CROSS-REFERENCE TO RELATED APPLICATIONS AND PRIORITY [001] The present application claims priority from Indian provisional application no. 201921029537, filed on July 22, 2019.
TECHNICAL FIELD [002] The disclosure herein generally relates to the field of sleep monitoring, and, more particular, to a method and system for monitoring sleep arousal.
BACKGROUND
[003] Sleep is a very critical part for health and well-being. Arousals are the brief wakefulness into the sleep, after which sleep resumes. Spontaneous arousals are normal. However, arousals that happen because of the various sleep disturbances can cause harm. One of the most common type of disturbances is caused by malfunction of the respiratory system resulting into ailments such as Apnea, Hypopnea and Respiratory Effort Related Arousal (RERA). Teeth grinding (bruxism), muscle jerks (including periodic limb movements during sleep), pain, insomnia, and snoring - all these can cause pathological arousals. Frequent sleep arousals can cause day time sleepiness resulting degraded cognitive performance.
[004] The gold standard for the sleep disorder monitoring is Polysomnography (PSG) study which includes manual scoring of the arousal events by one or more sleep experts by following a standard set of rules given by American Academy of Sleep Medicine (AASM). The duration of average sleep event is approximately 8 hours. Annotating 8 hours of data, in 10-second resolution is a long, tedious job, which requires careful observation by medical experts. This not only makes it a costly process, but also the annotation is often fairly time consuming. Hence, automated methods are preferred for sleep arousal monitoring. However there is a challenge in automated methods due to lack of accuracy.
SUMMARY

[005] Embodiments of the present disclosure present technological improvements as solutions to one or more of the above-mentioned technical problems recognized by the inventors in conventional systems. For example, in one embodiment, a method for monitoring sleep arousal is provided. The method includes receiving, a plurality of raw signals pertaining to a subject under test from a plurality of channels. Further, the method includes resampling each of the plurality of raw signals to a predetermined frequency. Further, the method includes extracting a plurality of features associated with the plurality of resampled signals based on a plurality of deep neural networks, wherein the plurality of features comprises a normalized energy trend of all the resampled signals excluding EEG and a non-stationarity present in EEG signals. Further, the method includes computing a plurality of relationship values associated with the plurality of resampled signals based on the plurality of features. Further, the method includes computing a probability time-series for sleep arousal based on the plurality of relationship values. Further, the method includes obtaining a smooth contour of the probability time series by smoothing the probability time series by utilizing a plurality of filters. Further, the method includes identifying a plurality of arousal locations by selecting a plurality of peaks of the smooth contour of the probability time-series, wherein the plurality of peaks are with an amplitude higher than a predetermined threshold and separated by a predetermined time duration. Furthermore, the method includes computing a Respiratory Disturbance Index (RDI) based on the plurality of arousal locations, by counting the average number of arousals per a predefined time. Finally, the method includes classifying the subject under test into at least one of a normal sleep conditioned subject or abnormal sleep conditioned subject based on the RDI, wherein the RDI is compared with a predefined threshold.
In another aspect, a system for monitoring sleep arousal is provided. The system includes a computing device wherein the computing device includes, at least one memory comprising programmed instructions, at least one hardware processor operatively coupled to the at least one memory, wherein the one or more hardware processors are configured by the programmed instructions to receive a plurality of

raw signals pertaining to a subject under test from a plurality of channels. Further, the one or more hardware processors are configured by the programmed instructions to resample each of the plurality of raw signals to a predetermined frequency. Further, the one or more hardware processors are configured by the programmed instructions to extract a plurality of features associated with the plurality of resampled signals based on a plurality of deep neural networks, wherein the plurality of features comprises a normalized energy trend of all the resampled signals excluding EEG and a non-stationarity present in EEG signals. Further, the one or more hardware processors are configured by the programmed instructions to compute a plurality of relationship values associated with the plurality of resampled signals based on the plurality of features. Further, the one or more hardware processors are configured by the programmed instructions to compute a probability time-series for sleep arousal based on the plurality of relationship values. Further, the one or more hardware processors are configured by the programmed instructions to obtain a smooth contour of the probability time series by smoothing the probability time series by utilizing a plurality of filters, wherein the plurality of filters comprises a median filter and a moving average filter. Further, the one or more hardware processors are configured by the programmed instructions to identify a plurality of arousal locations by selecting a plurality of peaks of the smooth contour of the probability time-series, wherein the plurality of peaks are with an amplitude higher than a predetermined threshold and separated by a predetermined time duration. Furthermore, the one or more hardware processors are configured by the programmed instructions to compute a Respiratory Disturbance Index (RDI) based on the plurality of arousal locations, by counting the average number of arousals per a predefined time. Finally, the one or more hardware processors are configured by the programmed instructions to classify the subject under test into at least one of a normal sleep conditioned subject or abnormal sleep conditioned subject based on the RDI, wherein the RDI is compared with a predefined threshold.
[006] In yet another aspect, a computer program product comprising a non-transitory computer-readable medium having the is configured to embodied

therein a computer program for method and system for monitoring sleep arousal is provided. The computer readable program, when executed on a computing device, causes the computing device to receive a plurality of raw signals pertaining to a subject under test from a plurality of channels. Further, the computer readable program, when executed on a computing device, causes the computing device to resample each of the plurality of raw signals to a predetermined frequency. Further, the computer readable program, when executed on a computing device, causes the computing device to extract a plurality of features associated with the plurality of resampled signals based on a plurality of deep neural networks, wherein the plurality of features comprises a normalized energy trend of all the resampled signals excluding EEG and a non-stationarity present in EEG signals. Further, the computer readable program, when executed on a computing device, causes the computing device to compute a plurality of relationship values associated with the plurality of resampled signals based on the plurality of features. Further, the computer readable program, when executed on a computing device, causes the computing device to compute a probability time-series for sleep arousal based on the plurality of relationship values. Further, the computer readable program, when executed on a computing device, causes the computing device to obtain a smooth contour of the probability time series by smoothing the probability time series by utilizing a plurality of filters, wherein the plurality of filters comprises a median filter and a moving average filter. Further, the computer readable program, when executed on a computing device, causes the computing device to identify a plurality of arousal locations by selecting a plurality of peaks of the smooth contour of the probability time-series, wherein the plurality of peaks are with an amplitude higher than a predetermined threshold and separated by a predetermined time duration. Furthermore, the computer readable program, when executed on a computing device, causes the computing device to compute a Respiratory Disturbance Index (RDI) based on the plurality of arousal locations, by counting the average number of arousals per a predefined time. Finally, the computer readable program, when executed on a computing device, causes the computing device to classify the subject under test into at least one of a normal sleep

conditioned subject or abnormal sleep conditioned subject based on the RDI, wherein the RDI is compared with a predefined threshold.
[007] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.
BRIEF DESCRIPTION OF THE DRAWINGS
[008] The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and, together with the description, serve to explain the disclosed principles:
[009] FIG. 1 is a functional block diagram of a system for monitoring sleep arousal, according to some embodiments of the present disclosure.
[010] FIG. 2 is a schematic block diagram illustrating a method for monitoring sleep arousal, in accordance with some embodiments of the present disclosure.
[011] FIG. 3 is a block diagram of a Channel Invariant Electroencephalography (EEG) network (CIEN) of the method for monitoring sleep arousal, in accordance with some embodiments of the present disclosure.
[012] FIG. 4 is a block diagram of a Trend Statistics Network (TSN) of the method for monitoring sleep arousal, in accordance with some embodiments of the present disclosure.
[013] FIG. 5A and 5B are exemplary flow diagram for a processor implemented method for monitoring sleep arousal, according to some embodiments of the present disclosure.
[014] FIG. 6A to 6K illustrates graphical representations of the experimental results for the method for monitoring sleep arousal, in accordance with some embodiments of the present disclosure.
DETAILED DESCRIPTION OF EMBODIMENTS

[015] Exemplary embodiments are described with reference to the accompanying drawings. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. Wherever convenient, the same reference numbers are used throughout the drawings to refer to the same or like parts. While examples and features of disclosed principles are described herein, modifications, adaptations, and other implementations are possible without departing from the spirit and scope of the disclosed embodiments. It is intended that the following detailed description be considered as exemplary only, with the true scope and spirit being indicated by the following claims.
[016] Embodiments herein provide a method and system for monitoring sleep arousal. The system for monitoring sleep arousal addresses the problem of automatic scoring of sleep arousal. The present disclosure provides an end-to-end trainable network by utilizing a plurality of raw signals associated with Polysomnography (PSG) and Sleep Arousal. The trainable network includes a Trend Statistics Network (TSN) to compute energy trends associated with the plurality of signals and a Chanel Invariant Electroencephalography (EEG) Network (CIEN) for computing a non-stationarity associated with the energy trends. An implementation of the method and system for monitoring sleep arousal is described further in detail with reference to FIGS. 1 through 6K.
[017] In an embodiment, according to American Academy of Sleep Medicine (AASM) manual, a sleep arousal is defined corresponding to the plurality of signals given below.
[018] Electroencephalography (EEG): Used to monitor electrical activity of brain. For example an arousal with respect to EEG means,“10s of sleep followed by sudden changes in the EEG frequency for atleast 3s" in any EEG channel,
[019] Chin-Electromyography (EMG): Reflects an inhibitory influence on motor activity and muscle tone. For example, the sleep arousal corresponding to EMG indicates an increase in amplitude of EMG signal depending on the sleep stage.

[020] Airflow (AF): Sequence of breaths lasting greater than 10 seconds characterized by increasing respiratory effort or flattening of the inspiration portion of the nasal pressure channel indicates the sleep arousal.
[021] Oxygen Saturation (SaO2): Reduction in Sao2 from the baseline prior to the event can cause the sleep arousal.
[022] Electrocardiography (ECG): The sleep arousal can be identified using ECG based on increase in heart rate depending on the scale of the arousal.
[023] Referring now to the drawings, and more particularly to FIG. 1 through 6K, where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments and these embodiments are described in the context of the following exemplary system and/or method.
[024] FIG. 1 is a functional block diagram of a system for monitoring sleep arousal, according to some embodiments of the present disclosure. The system 100 includes or is otherwise in communication with one or more hardware processors, such as a processors 102, at least one memory such as a memory 104, an I/O interface 122. The memory 104 may include the sleep arousal monitoring unit 120. The processor 102, memory 104, and the I/O interface 122 may be coupled by a system bus such as a system bus 108 or a similar mechanism.
[025] The I/O interface 122 may include a variety of software and hardware interfaces, for example, a web interface, a graphical user interface, and the like. The interface 122 may include a variety of software and hardware interfaces, for example, interfaces for peripheral device(s), such as a keyboard, a mouse, an external memory, a plurality of sensor devices, a printer and the like. Further, the interface 122 may enable the system 100 to communicate with other devices, such as web servers and external databases.
[026] The interface 122 can facilitate multiple communications within a wide variety of networks and protocol types, including wired networks, for example, local area network (LAN), cable, etc., and wireless networks, such as Wireless LAN (WLAN), cellular, or satellite. For the purpose, the interface 122 may include one or more ports for connecting a number of computing systems with

one another or to another server computer. The I/O interface 122 may include one or more ports for connecting a number of devices to one another or to another server.
[027] The hardware processor 102 may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuitries, and/or any devices that manipulate signals based on operational instructions. Among other capabilities, the hardware processor 102 is configured to fetch and execute computer-readable instructions stored in the memory 104.
[028] The memory 104 may include any computer-readable medium known in the art including, for example, volatile memory, such as static random access memory (SRAM) and dynamic random access memory (DRAM), and/or non-volatile memory, such as read only memory (ROM), erasable programmable ROM, flash memories, hard disks, optical disks, and magnetic tapes. In an embodiment, the memory 104 includes a plurality of modules 106 and a repository 110 for storing data processed, received, and generated by one or more of the modules 106 and the image analysis unit 120. The modules 106 may include routines, programs, objects, components, data structures, and so on, which perform particular tasks or implement particular abstract data types.
[029] The memory 104 also includes module(s) 106 and a data repository 110. The module(s) 106 include programs or coded instructions that supplement applications or functions performed by the system 100 for monitoring sleep arousal. The modules 106, amongst other things, can include routines, programs, objects, components, and data structures, which perform particular tasks or implement particular abstract data types. The modules 106 may also be used as, signal processor(s), state machine(s), logic circuitries, and/or any other device or component that manipulates signals based on operational instructions. Further, the modules 106 can be used by hardware, by computer-readable instructions executed by a processing unit, or by a combination thereof. The modules 106 can include various sub-modules (not shown). The modules 106 may include computer-

readable instructions that supplement applications or functions performed by the system 100 for monitoring sleep arousal.
[030] The data repository 110 may include a “Physionet challenge 2018” data set and other data. Further, the other data 118 amongst other things, may serve as a repository for storing data that is processed, received, or generated as a result of the execution of one or more modules in the module(s) 106 and the modules associated with the sleep arousal monitoring unit 120.
[031] Although the data repository 110 is shown internal to the system 100, it will be noted that, in alternate embodiments, the data repository 110 can also be implemented external to the computing device 100, where the data repository 110 may be stored within a database (not shown in FIG. 1) communicatively coupled to the computing device 100. The data contained within such external database may be periodically updated. For example, new data may be added into the database (not shown in FIG. 1) and/or existing data may be modified and/or non-useful data may be deleted from the database (not shown in FIG. 1). In one example, the data may be stored in an external system, such as a Lightweight Directory Access Protocol (LDAP) directory and a Relational Database Management System (RDBMS). In another embodiment, the data stored in the data repository 110 may be distributed between the computing device 104 and the external database (not shown).
[032] FIG. 2 is a schematic block diagram illustrating a method for monitoring sleep arousal, in accordance with some embodiments of the present disclosure. Now referring to FIG. 2, the method includes a CIEN (Channel Invariant EEG Network), a plurality of Trend Statistics Network (TSN) and a channel fusing network. The CIEN receives EEG signal as input and provides batch normalized features as output. Each of the plurality of TSN channels receives Airflow (AF), Abdomen-Electromyography (ABD), Chest-Electromyography (CHST), Electro-Oculo-Graphy (EOG), Chin-Electromyography (EMG), Electrocardiography (ECG) and Oxygen Saturation (SaO2) as input
[033] In an embodiment, the TSN is indicated by TSNx {1 < x < 8}. Two values in the bracket below TSNx indicate (N,L), where N is the number of filters

and L is order of the filter. When TSNx is used in multiple blocks corresponding to different input channels in FIG. 2, it indicates that the parameters are shared between the blocks. For EMG, only the energy change matters, and hence, no filters are used. Change in Sao2 is needed for arousal detection and hence no filters or square operation is used for Sao2 channel. For this reason, (N,L) pair associated with the EMG and Sao2 are indicated as (0,0) for TSN6 and TSN8. The number of filters used for EEG channels is 16 and for rest of the channels (AF, CHST, ABD,EOG, ECG) four filters are used. The order of all filters is experimentally set to 64. The feature outputs of the CIEN and the TSN networks are provided as input to the channel fusion network. The channel fusion network includes a plurality of separable convolution 1 Dimensional network with order (80,3) 202, 206, batch normalization block 204, 3 layered Binary LSTM (Long Short Term Memory) 208 and a sigmoid activation function 210. The feature output from the CIEN and TSN networks are provided as input to the first separable convolution 1D block 202 where a plurality of features are extracted and the plurality of features are batch normalized. Further, the batch normalized features are provided as input to the second separable convolution 1D block 206 and the 3 layered BLSTM 208 simultaneously to obtain a plurality of relationship values associated with the plurality of features. The plurality of relationship values obtained from the second separable convolution 1D block 206 and the 3 layered BLSTM 208 are combined and provided as input to the sigmoid activation function. The sigmoid activation function calculates the probability of arousal.
[034] FIG. 3 is a block diagram of a Channel Invariant Electroencephalography (EEG) network (CIEN) of the method for monitoring sleep arousal, in accordance with some embodiments of the present disclosure. Now referring to FIG. 3, the CIEN network includes a plurality of TSN channels. In an embodiment, the plurality of channel are six numbers. Each of the plurality of TSN network receives the corresponding EEG signal and obtains a plurality of energy trend values corresponding to each of the TSN channel. The plurality of energy trend values provided as input to an activation sorting block, wherein the sorting of the normalized energy trend values corresponding to each of the 6 EEG

channels is performed in descending order. Further the non-stationarity present in the plurality of EEG signals are computed based on the sorted energy trend values, wherein the non-stationarity is a weighted combination of sorted energy trends. Further, the weighted combination of sorted energy trends are batch normalized.
[035] In an embodiment, the CIEN network is trained as follows: An arousal is marked if a sudden change is detected in any of the plurality of EEG channels. In an embodiment, the training is performed without augmentation of training data. Here, the first stage of the CIEN architecture is the TSN to measure the change in the features of EEG as depicted in FIG. 3. Identical TSN is used for all six EEG channels to get a total of 16x3 (=48) TSN outputs denoted by
Further, the activation of n-th TSN output from all six EEG channels are sorted in descending order to obtain a TSN sorted and the linear combination of the sorted outputs are
taken to get the n-th EEG feature as given in equation 1.

Where are the weights learnt as a part of the network training. A
higher value of the TSN output indicates a sudden change in the EEG frequency and the respective EEG channel comes to the top after sorting; w decides the importance of the sorted channel.
[036] In an embodiment, the database includes labels for no arousal or sleep (0), targeted arousal (1) and non-targeted arousal (-1). The network uses the entire sleep recording of a subject as one batch and the goal of the present disclosure is to detect the targeted arousal and sleep. Hence the following objective function given in equation 2 is utilized.

Where F is the indicator function, ypred(i) denoted the prediction, is the
ground truth label. BC is Binary Entropy function and T is the number of time steps. [037] FIG. 4 is a block diagram of a Trend Statistics Network (TSN) of the method for monitoring sleep arousal, in accordance with some embodiments of the present disclosure. Now referring to FIG. 4, the TSN includes a filter, squaring

function, a plurality of moving average functions, a plurality of ratio functions and an average pooling.
[038] In an embodiment, the sleep arousal is defined using a change in the features for at least 3 seconds(s) in any of the channels with respect to pre-arousal baseline (10s). To establish a pre-event baseline and to measure the change in the signal feature, the present disclosure utilizes trend statistics network shown in the FIG. 3. In order to extract the feature from the signal, the signal is first passed through a filter of order L and the energy of the filter output is computed. To measure the energy trends at different time scales, a moving average of the filter output for different lengths is computed using causal (1s, 3s, 10s) and anti-causal (3s) filters. A moving average filter of length x is denoted by MAVG(x). The absolute value of energy of the filter output may not be relevant for arousal detection as it significantly changes for different sleep stages and across subjects. Hence, to measure the change, a normalization is performed by taking the ratio of the output of the moving average filter output with the output of a MAVG(10) filter. The 50Hz signal is average pooled to get a signal of 1Hz. In an embodiment, three TSN outputs (denoted by r1,r3,r3ac are obtained corresponding to two causal (1s, 3s) and one anti-causal (3s) moving average filters.
[039] In an embodiment, the PSG recordings having 13 time-series from 7 different physiological sensors includes 6-channel EEG (F3-M2, F4- M1, C3-M2, C4-M1, O1-M2, O2-M1), single channel Electrooculography (EOG) (from left eye with right ear EEG and M2 as reference), EMG, Chest (CHST) and Abdomen (ABD) effort, single-lead ECG, Respiratory Airflow (AF) and SaO2. Excluding SaO2, all other signals are sampled at 200 Hz and are measured in microvolts. F3-M2, F4- M1, C3-M2, C4-M1, O1-M2, O2-M1 are the standard electrode location points for EEG. For the convenience of data analysis, SaO2 is resampled to 200 Hz. A total of seven scorers have annotated the database for arousal, but with one scorer per PSG. EEG signals were scored in non-overlapping 30s epochs according to the AASM (American Academy of Sleep Medicine) manual.
[040] The sleep arousal unit 120 , executed by one or more processors of the system 100, receives a plurality of raw signals pertaining to a subject under test

from a plurality of channels. The plurality of raw signals comprises a plurality of Electroencephalography (EEG) signals, Airflow (AF), Abdomen-Electromyography (ABD), Chest-Electromyography (CHST), Electro-Oculo-Graphy (EOG), Chin-Electromyography (EMG), Electrocardiography (ECG) and Oxygen Saturation (SaO2).
[041] Further, the sleep arousal unit 120, executed by one or more processors of the system 100, resamples each of the plurality of raw signals to a predetermined frequency. For example the predetermined frequency range can be 50Hz,
[042] Further, the sleep arousal unit 120, executed by one or more processors of the system 100, extracts a plurality of features associated with the plurality of resampled signals based on a plurality of deep neural networks, wherein the plurality of features includes a normalized energy trend of all the resampled signals excluding EEG and a non-stationarity present in EEG signals. The plurality of deep neural networks includes a pre-trained Trend Statistics Network (TSN) and a pre-trained Channel Invariant EEG Network (CIEN) and a channel fusion network as illustrated in FIG. 2. The TSN is a tailor-made deep neural network to capture energy trends of the signal at different time scales, and the CIEN is a tailor-made deep neural network to compute non-stationarity of the multi-channel EEG signal. [043] In an embodiment, the method of extracting the plurality of features associated with the plurality of resampled signals based on TSN includes the following steps: (1) filter each of the plurality of resampled signals using a filter of certain pre-defined order (2) compute an energy associated with each of the plurality of filtered signals by squaring each filtered signal. (3) compute an energy trend associated with the corresponding computed energy at varying time scales based on a moving average of the signal energies, wherein the moving average is computed by causal and anti-causal filtering (4) normalize the energy trend by computing the ratio of moving average of each time scale to the reference moving average obtained using 10sec causal filter and (5) compute average of the normalized energy trend to obtain the energy trend in a particular frequency.

[044] In an embodiment the method of extracting the plurality of features associated with the plurality of resampled EEG signals based on CIEN includes the following steps: (1) compute a normalized energy trend values for each of the plurality of EEG channels by utilizing the TSN network (2) sort the normalized energy trend values corresponding to each of the 6 EEG channels in descending order and (3) computing the non-stationarity present in EEG signals based on the sorted energy trend values, wherein the non-stationarity is a weighted combination of sorted energy trends.
[045] Further, the sleep arousal unit 120, executed by one or more processors of the system 100, computes a plurality of relationship values associated with the plurality of resampled signals based on the plurality of features. The relationship values comprising a plurality of spatial relationship values and a plurality of temporal relationship values. The plurality of relationship values are computed by utilizing a convolutional neural network and a bi-directional Long Short Term Memory (LSTM) network as illustrated in FIG. 2.
[046] Further, the sleep arousal unit 120, executed by one or more processors of the system 100, computes a probability time-series for sleep arousal based on the plurality of relationship values.
[047] Further, the sleep arousal unit 120, executed by one or more processors of the system 100, obtains a smooth contour of the probability time series by smoothing the probability time series by utilizing a plurality of filters. The plurality of filters includes a median filter and a moving average filter.
[048] Further, the sleep arousal unit 120, executed by one or more processors of the system 100, identifies a plurality of arousal locations by selecting a plurality of peaks of the smooth contour of the probability time-series, wherein the plurality of peaks are with an amplitude higher than a predetermined threshold and separated by a predetermined time duration. The plurality of arousal locations comprises arousals pertaining to Apnea, Hypopnea, Respiratory Effort Related Arousal (RERA), teeth grinding, muscle jerks, pain, insomnia and snoring.
[049] Further, the sleep arousal unit 120, executed by one or more processors of the system 100, computes a Respiratory Disturbance Index (RDI)

based on the plurality of arousal locations, by counting the average number of arousals per unit time. In an embodiment, RDI is a number measured by the average number of arousal events occurred per hour in a single sleep event. It is defined as given in equation 3.

where TST is total sleep time in minutes. Hence, the lower the RDI, the better is the subject’s sleep arousal status. The higher the RDI, the more a person feels tired and deprived of sleep. The normal range for RDI is below 15. 15<=RDI<= 30 is a range that indicates moderate problem, while RDI>30 generally indicates severe problem.
[050] Further, the sleep arousal unit 120, executed by one or more processors of the system 100, classifies the subject under test into at least one of a normal sleep conditioned subject or abnormal sleep conditioned subject based on the RDI, wherein the RDI is compared with a predefined threshold.
[051] FIG. 5 is an exemplary flow diagram for a processor implemented method for monitoring sleep arousal, according to some embodiments of the present disclosure. The method 500 may be described in the general context of computer executable instructions. Generally, computer executable instructions can include routines, programs, objects, components, data structures, procedures, modules, functions, etc., that perform particular functions or implement particular abstract data types. The method 400 may also be practiced in a distributed computing environment where functions are performed by remote processing devices that are linked through a communication network. The order in which the method 400 is described is not intended to be construed as a limitation, and any number of the described method blocks can be combined in any order to implement the method 400, or an alternative method. Furthermore, the method 400 can be implemented in any suitable hardware, software, firmware, or combination thereof.
[052] At 502, the system 100, receives, by a one or more hardware processors, the plurality of raw signals pertaining to a subject under test from a plurality of channels. The plurality of raw signals comprises a plurality of EEG signals, AF, ABD, CHST, EOG, EMG, ECG and SaO2. At 504, the system 100, resamples, by a one or more hardware processors, each of the plurality of raw

signals to a predetermined frequency. At 506, the system 100, extracts, by a one or more hardware processors, the plurality of features associated with the plurality of resampled signals based on a plurality of deep neural networks, wherein the plurality of features comprises a normalized energy trend of all the resampled signals excluding EEG and a non-stationarity present in EEG signals. The plurality of deep neural networks comprising a pre-trained Trend Statistics Network (TSN) and a pre-trained Channel Invariant EEG Network (CIEN), wherein the TSN is a tailor-made deep neural network to capture energy trends of the signal at different time scales, wherein the CIEN is a tailor-made deep neural network to compute non-stationarity of the multi-channel EEG signal. The method of extracting the plurality of features associated with the plurality of resampled signals based on TSN includes the following steps: (1) filtering each of the plurality of resampled signals using a filter of certain pre-defined order (2) computing an energy associated with each of the plurality of filtered signals by squaring each filtered signal (3) computing an energy trend associated with the corresponding computed energy at varying time scales based on a moving average of the signal energies, wherein the moving average is computed by causal and anti-causal filtering (4) normalizing the energy trend by computing the ratio of moving average of each time scale to the reference moving average obtained using 10sec causal filter and (5) averaging the normalized energy trend to obtain the energy trend in a particular frequency. The method of extracting the plurality of features associated with the plurality of resampled EEG signals based on CIEN includes the following steps: (1) computing normalized energy trend values for each of the plurality of EEG channels by utilizing the TSN network (2) sorting the normalized energy trend values corresponding to each of the plurality of EEG channels in descending order and (3) computing the non-stationarity present in EEG signals based on the sorted energy trend values, wherein the non-stationarity is a weighted combination of sorted energy trends. At 508, the system 100, computes, by a one or more hardware processors, the plurality of relationship values associated with the plurality of resampled signals based on the plurality of features, wherein the relationship values comprising a plurality of spatial relationship values and a plurality of temporal

relationship values. The plurality of relationship values are computed by utilizing a convolutional neural network and a bi-directional Long Short Term Memory (LSTM) network. At 510, the system 100, computes, by a one or more hardware processors, a probability time-series or a probability for sleep arousal based on the plurality of relationship values. At 512, the system 100, obtains, by a one or more hardware processors, a smooth contour of the probability time series by smoothing the probability time series by utilizing a plurality of filters, wherein the plurality of filters comprises a median filter and a moving average filter. At 514, the system 100, identifies, by a one or more hardware processors, the plurality of arousal locations by selecting a plurality of peaks of the smooth contour of the probability time-series, wherein the plurality of peaks are with an amplitude higher than a predetermined threshold and separated by a predetermined time duration. The plurality of arousal locations comprises arousals pertaining to Apnea, Hypopnea, Respiratory Effort Related Arousal (RERA), teeth grinding, muscle jerks, pain, insomnia and snoring. At 516, the system 100, computes, by a one or more hardware processors, the RDI based on the plurality of arousal locations, by counting the average number of arousals per unit time. At 518, the system 100, classifies, by a one or more hardware processors, the subject under test into at least one of a normal sleep conditioned subject or abnormal sleep conditioned subject based on the RDI, wherein the RDI is compared with a predefined threshold. The higher the RDI, the more a person feels tired and deprived of sleep. The normal range for RDI is below 15. 15<=RDI<= 30 is a range that indicates moderate problem, while RDI>30 generally indicates severe problem. In an embodiment, the RDI is taken as the measure of the low resolution sleep arousal study.
[053] FIG. 6A to 6K illustrates graphical representation of experimental results for the method for monitoring sleep arousal, in accordance with some embodiments of the present disclosure.
[054] In an embodiment, the system 100 is experimented as follows: To conduct the experiments, the available annotated data (994 recordings) are divided into 10 folds, where each includes approximately 690 training, approximately 200 validation and approximately 100 test recordings. Each of 994 files is used exactly

once as a test file. The network is implemented in tensorflow and keras. The network is initialized randomly and the objective function given in equation. 2 is optimized using adam optimizer for 20 iterations and the model with lowest validation error has been chosen. To measure the binary classification accuracy, the Area Under the Precision Recall Curve (AUPRC) is utilized as the evaluation metric. To measure the accuracy of RDI prediction, three kinds of metrics are considered. In the first metric, the mean absolute error between the predicted RDI and the RDI computed from the annotations is computed. In the second one, the classification performance (sensitivity and specificity) of normal (class 1: RDI<=15) and abnormal (class 2: RDI >=15) range classification is computed. In the third one, the severity classification performance, among normal (class 1: RDI<=15) and mild (class 2: 15 30) classes is considered.
[055] In an embodiment, high resolution study (AUPRC cross validation) is performed and the results are discussed below.
[056] FIG. 6A depicts the AUPRC for each subject’s recording vs number of RERA arousals in the recording. It is clear from the figure that as the number of the arousals increases the AUPRC increases. In fact, the AUPRC value is more than 0.8 for recordings where majority of them have large number of arousals. AUPRC is low for recording with less number of arousals. The correlation coefficient between the two variables is found to be 0.642. This could happen because of the fact that the AUPRC penalizes more for the error in predicting the arousal when there is a large imbalance between the arousal and non-arousal regions in a recording. The cumulative AUPRC for all test files is shown in FIG. 6B. It is clear from the figure that there is an initial rise in cumulative AUPRC, and is because the initial recordings might have more number of arousal and hence more AUPRC which cause the cumulative AUPRC to increase and after 200 files the AUPRC becomes stable.
[057] In an embodiment, low resolution study (AUPRC cross validation) is performed and the results are discussed below.
[058] In an embodiment, The RDI is predicted from the probability predicted by the network as discussed in the previous section. The histogram of the

error in predicting RDI is shown FIG. 6C. It is clear from the figure that approximately 80% of the errors are within +10, with a mean absolute error of 6.11 with the standard deviation of 5.64. FIG. 6D shows the Bland-Altman plot for ground truth and predicted RDI. It is clear from the figure that the mean of the RDI error is close to zero and that linear fit to the data has slope close to zero (0.02), which indicates that the error in RDI does not correlate with the magnitude of ground truth RDI.
[059] In an embodiment, the main parameter in any sleep study report is the quantification of the severity based on the RDI. Confusion matrices for the two class and three class classification are shown in Table IA and 1B. Tables IA and IB depicts that, for two class classification problem, a specificity of 75% and sensitivity of 83% has been achieved. However, for three class classification problem, the confusion among class 1&2 and class 2&3 is more. Even though the objective function in equation 2 uses just the RERA arousal for the training, the performance of both low and high resolution study drops when all the arousals are included in the training. It could be because of the training of the network is biased in detecting apnea arousals and fail to detect the RERA arousals. However, since both class 2&3 relate to medical condition of the sleep arousal, the two class classification is sufficient for separating normal and abnormal people, which is the primary requirement of any screening process.



[060] In an embodiment, the plurality of the networks of the system 100 has been trained and an interpretation of the plurality of trained network is given below:
[061] First, a plurality of filters are learned in the TSNs for all channels. To learn the filters, a cumulative frequency response of all the filters for the corresponding channels are measured. The cumulative frequency response is obtained by summing up all the discovered filters’ power spectral densities and is useful to highlight which frequency bands are emphasized by the learned filters. FIG. 6E to 6J depicts the cumulative frequency response of the filter for each channel with mean and standard deviation across the 10 folds. The solid line indicates the mean of the cumulative responses for the 10 channels and the dotted lines indicates the corresponding standard deviation. The figures 6E to 6J indicates the cumulative filter response is similar for all 10 folds.
[062] In an embodiment, the plurality of filters learned for the EEG has peaks around 10Hz, around 18Hz and most of the filters have high gain in the low frequency. The peaks of EEG channels correlates with AASM manual which mentions that the arousal is defined as a sudden change in the frequency of theta waves (3-5Hz) and the frequency >16Hz, but not alpha (8-12Hz) waves. For chest and abdomen channels, the filters are mostly low-pass filters that primarily measures the envelope of the signal. Similarly for EOG, the learned filters are also low-pass in nature and many filters have peaks around 1Hz which could measure the eye-movement (eye-blink during wake and rem sleep stage). The airflow filter is also low pass in nature but has significant gain is greater than 1Hz frequency. This could be because the flattening of the airflow needs to detected for the RERA arousal. The flattening can be detected by measuring the energy at the fundamental frequency and its harmonics.
[063] In an embodiment, a high gain for frequency range >1Hz could help in measuring the strength of the harmonics. ECG filter has a different structure which shows high gain around 1-5Hz, which is the frequency range of the P/T-waves. It also has high gain around 5-15Hz, which is the frequency range of QRS

complex. The moving average energy in these frequency regions will be correlated with the heart rate, which is an indicator of the arousal and the REM sleep stage.
[064] In an embodiment, FIG. 6K illustrates the mean and standard deviation of the trained wlt 1 < i < 6 across 10 folds after the sorting the EEG activation. Now, referring to FIG. 6K, the channel 1 and 2 have the highest weight compared to others. It indicates that change in features in primarily two channels is enough for the classification. Here, the solid line indicates the mean and the dotted lines indicates the standard deviation.
[065] The written description describes the subject matter herein to enable any person skilled in the art to make and use the embodiments. The scope of the subject matter embodiments is defined by the claims and may include other modifications that occur to those skilled in the art. Such other modifications are intended to be within the scope of the claims if they have similar elements that do not differ from the literal language of the claims or if they include equivalent elements with insubstantial differences from the literal language of the claims.
[066] The embodiments of present disclosure herein addresses unresolved problem of accurate and automated monitoring of sleep arousal study. Here the system 100 is able to perform a high resolution sleep arousal monitoring and utilize the high resolution sleep arousal monitoring to monitor the low resolution sleep arousal monitoring. Here, tailor made CIEN and TSN networks are implemented and utilized to classify the sleep arousal of the subject.
[067] It is to be understood that the scope of the protection is extended to such a program and in addition to a computer-readable means having a message therein; such computer-readable storage means contain program-code means for implementation of one or more steps of the method, when the program runs on a server or mobile device or any suitable programmable device. The hardware device can be any kind of device which can be programmed including e.g. any kind of computer like a server or a personal computer, or the like, or any combination thereof. The device may also include means which could be e.g. hardware means like e.g. an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination of hardware and software means, e.g. an

ASIC and an FPGA, or at least one microprocessor and at least one memory with software modules located therein. Thus, the means can include both hardware means and software means. The method embodiments described herein could be implemented in hardware and software. The device may also include software means. Alternatively, the embodiments may be implemented on different hardware devices, e.g. using a plurality of CPUs.
[068] The embodiments herein can comprise hardware and software elements. The embodiments that are implemented in software include but are not limited to, firmware, resident software, microcode, etc. The functions performed by various modules described herein may be implemented in other modules or combinations of other modules. For the purposes of this description, a computer-usable or computer readable medium can be any apparatus that can comprise, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
[069] The illustrated steps are set out to explain the exemplary embodiments shown, and it should be anticipated that ongoing technological development will change the manner in which particular functions are performed. These examples are presented herein for purposes of illustration, and not limitation. Further, the boundaries of the functional building blocks have been arbitrarily defined herein for the convenience of the description. Alternative boundaries can be defined so long as the specified functions and relationships thereof are appropriately performed. Alternatives (including equivalents, extensions, variations, deviations, etc., of those described herein) will be apparent to persons skilled in the relevant art(s) based on the teachings contained herein. Such alternatives fall within the scope and spirit of the disclosed embodiments. Also, the words “comprising,” “having,” “containing,” and “including,” and other similar forms are intended to be equivalent in meaning and be open ended in that an item or items following any one of these words is not meant to be an exhaustive listing of such item or items, or meant to be limited to only the listed item or items. It must also be noted that as used herein and in the appended claims, the singular

forms “a,” “an,” and “the” include plural references unless the context clearly dictates otherwise.
[070] Furthermore, one or more computer-readable storage media may be utilized in implementing embodiments consistent with the present disclosure. A computer-readable storage medium refers to any type of physical memory on which information or data readable by a processor may be stored. Thus, a computer-readable storage medium may store instructions for execution by one or more processors, including instructions for causing the processor(s) to perform steps or stages consistent with the embodiments described herein. The term “computer-readable medium” should be understood to include tangible items and exclude carrier waves and transient signals, i.e. non-transitory. Examples include random access memory (RAM), read-only memory (ROM), volatile memory, nonvolatile memory, hard drives, CD ROMs, DVDs, flash drives, disks, and any other known physical storage media.
[071] It is intended that the disclosure and examples be considered as exemplary only, with a true scope and spirit of disclosed embodiments being indicated by the following claims.

WE CLAIM: 1. A processor implemented method, the method comprising:
receiving a plurality of raw signals pertaining to a subject under test from a plurality of channels;
resampling each of the plurality of raw signals to a predetermined frequency;
extracting a plurality of features associated with the plurality of resampled signals based on a plurality of deep neural networks, wherein the plurality of features comprises a normalized energy trend of each of the resampled signals excluding EEG signals and a non-stationarity present in the EEG signals;
computing a plurality of relationship values associated with the plurality of resampled signals based on the plurality of feature, wherein the plurality of relationship values comprising a plurality of spatial relationship values and a plurality of temporal relationship values;
computing a probability time-series for sleep arousal based on the plurality of relationship values;
obtaining a smooth contour of the probability time series by smoothing the probability time series by utilizing a plurality of filters;
identifying a plurality of arousal locations by selecting a plurality of peaks of the smooth contour of the probability time-series, wherein the plurality of peaks are with an amplitude higher than a predetermined threshold and separated by a predetermined time duration;
computing a Respiratory Disturbance Index (RDI) based on the plurality of arousal locations, by counting the average number of arousals per a predefined time; and
classifying the subject under test into at least one of a normal sleep conditioned subject or abnormal sleep conditioned subject based on the RDI, wherein the RDI is compared with a predefined threshold.

2. The method as claimed in claim 1, wherein the method of extracting the
plurality of features associated with the plurality of resampled signals based
on TSN comprises:
filtering each of the plurality of resampled signals using a filter of certain pre-defined order;
computing an energy associated with each of the plurality of filtered signals by squaring each filtered signal;
computing an energy trend associated with the corresponding computed energy at varying time scales based on a moving average of the signal energies, wherein the moving average is computed by causal and anti-causal filtering;
normalizing the energy trend by computing the ratio of moving average of each time scale to the reference moving average obtained using 10 sec causal filter; and
averaging the normalized energy trend to obtain the energy trend in a particular frequency.
3. The method as claimed in claim 1, wherein the method of extracting the
plurality of features associated with the plurality of resampled EEG signals
based on CIEN comprises:
computing normalized energy trend values for each of the plurality of EEG channels by utilizing the TSN network;
sorting the normalized energy trend values corresponding to each of the plurality of EEG channels in descending order; and
computing the non-stationarity present in EEG signals based on the sorted energy trend values, wherein the non-stationarity is a weighted combination of sorted energy trends.
4. The method as claimed in claim 1, wherein the plurality of deep neural
networks comprising a pre-trained Trend Statistics Network (TSN) and a
pre-trained Channel Invariant EEG Network (CIEN), wherein the TSN is a

tailor-made deep neural network to capture energy trends of the signal at different time scales, wherein the CIEN is a tailor-made deep neural network to compute non-stationarity of the multi-channel EEG signal.
5. The method as claimed in claim 1, the plurality of relationship values are computed by utilizing a convolutional neural network and a bi-directional Long Short Term Memory (LSTM) network.
6. The method as claimed in claim 1, wherein the plurality of arousal locations comprises arousals pertaining to Apnea, Hypopnea, Respiratory Effort Related Arousal (RERA), teeth grinding, muscle jerks, pain, insomnia and snoring.
7. The method as claimed in claim 1, wherein the plurality of raw signals comprises a plurality of Electroencephalography (EEG) signals, Airflow (AF), Abdomen-Electromyography (ABD), Chest-Electromyography (CHST), Electro-Oculo-Graphy (EOG), Chin-Electromyography (EMG), Electrocardiography (ECG) and Oxygen Saturation (SaO2).
8. The method as claimed in claim 1, wherein the plurality of filters comprises a median filter and a moving average filter.
9. A system (100), the system (100) comprising:
at least one memory (104) storing programmed instructions; one or more Input /Output (I/O) interfaces (112); and one or more hardware processors (102) operatively coupled to the at least one memory (104), wherein the one or more hardware processors (102) are configured by the programmed instructions to:
receive a plurality of raw signals pertaining to a subject under test from a plurality of channels;
resample each of the plurality of raw signals to a predetermined frequency;

extract a plurality of features associated with the plurality of resampled signals based on a plurality of deep neural networks, wherein the plurality of features comprises a normalized energy trend of all the resampled signals excluding EEG and a non-stationarity present in EEG signals;
compute a plurality of relationship values associated with the plurality of resampled signals based on the plurality of features, wherein the relationship values comprising a plurality of spatial relationship values and a plurality of temporal relationship values;
compute a probability time-series for sleep arousal based on the plurality of relationship values;
obtain a smooth contour of the probability time series by smoothing the probability time series by utilizing a plurality of filters;
identify a plurality of arousal locations by selecting a plurality of peaks of the smooth contour of the probability time-series, wherein the plurality of peaks are with an amplitude higher than a predetermined threshold and separated by a predetermined time duration;
compute a Respiratory Disturbance Index (RDI) based on the plurality of arousal locations, by counting the average number of arousals per a predefined time; and
classify the subject under test into at least one of a normal sleep conditioned subject or abnormal sleep conditioned subject based on the RDI, wherein the RDI is compared with a predefined threshold.
10. The system of claim 9, wherein the sleep arousal monitoring unit is configured to extract the plurality of features associated with the plurality of resampled signals based on TSN by:
filtering each of the plurality of resampled signals using a filter of certain pre-defined order;
computing an energy associated with each of the plurality of filtered signals by squaring each filtered signal;

computing an energy trend associated with the corresponding computed energy at varying time scales based on a moving average of the signal energies, wherein the moving average is computed by causal and anti-causal filtering ;
normalizing the energy trend by computing the ratio of moving average of each time scale to the reference moving average obtained using 10sec causal filter; and
averaging the normalized energy trend to obtain the energy trend in a particular frequency.
11. The system of claim 9, wherein the sleep arousal monitoring unit is
configured to extract the plurality of features associated with the plurality
of resampled EEG signals based on CIEN by:
computing normalized energy trend values for each of the plurality of EEG channels by utilizing the TSN network;
sorting the normalized energy trend values corresponding to each of the plurality of EEG channels in descending order; and
computing the non-stationarity present in EEG signals based on the sorted energy trend values, wherein the non-stationarity is a weighted combination of sorted energy trends.
12. The system of claim 9, wherein the plurality of deep neural networks comprising a pre-trained Trend Statistics Network (TSN) and a pre-trained Channel Invariant EEG Network (CIEN), wherein the TSN is a tailor-made deep neural network to capture energy trends of the signal at different time scales, wherein the CIEN is a tailor-made deep neural network to compute non-stationarity of the multi-channel EEG signal.
13. The system of claim 9, the plurality of relationship values are computed by utilizing a convolutional neural network and a bi-directional Long Short Term Memory (LSTM) network.

14. The system of claim 9, wherein the plurality of arousal locations comprises arousals pertaining to Apnea, Hypopnea, Respiratory Effort Related Arousal (RERA), teeth grinding, muscle jerks, pain, insomnia and snoring.
15. The system as claimed in claim 9, wherein the plurality of raw signals comprises a plurality of Electroencephalography (EEG) signals, Airflow (AF), Abdomen-Electromyography (ABD), Chest-Electromyography (CHST), Electro-Oculo-Graphy (EOG), Chin-Electromyography (EMG), Electrocardiography (ECG) and Oxygen Saturation (SaO2).
16. The system of claim 9, wherein the plurality of filters comprises a median filter and a moving average filter.