BIOFEEDBACK SYSTEM

Abstract

A biofeedback system for headache patients comprises: a sensor system (2) for obtaining and transmitting data indicative of a plurality of physiological parameters of the patient, a personal computing device (4) arranged to receive data from the sensor system (2) and to interact with the patient via a user interface of the personal computing device (4), and a computer-implemented biofeedback agent (6). The sensor system (2) is configured for measurement of the physiological parameters of the patient, wherein the physiological parameters include at least two of muscle tension, body temperature, heart rate and heart rate variability. The personal computing device (4) together with the biofeedback agent (6) are configured to carry out the following steps: obtain data indicative of the physiological parameters from the sensor system (2); instruct the patient to control the physiological parameters; determine baseline levels for each of the physiological parameters; derive a score for each of the physiological parameters relative to the baseline levels, wherein the score increases in reaction to control of the respective physiological parameter; and use a weighting system to determine a weighting for the score associated with each of the physiological parameters and present a total score to the patient via the user interface, the total score consisting of a combination of the weighted scores, wherein the weighting system gives a higher weighting to the score for the physiological parameter that the patient most successfully controls and a lower weighting to the score for the physiological parameter that the patient least successfully controls.

Claims

1. A biofeedback system for headache patients, the system comprising: a sensor system for obtaining and transmitting data indicative of a plurality of physiological parameters of the patient, the sensor system being configured for measurement of the physiological parameters of the patient, wherein the physiological parameters include at least two of muscle tension, body temperature, heart rate, and heart rate variability; a personal computing device arranged to receive data from the sensor system and to interact with the patient via a user interface of the personal computing device; and a computer-implemented biofeedback agent; wherein the personal computing device together with the biofeedback agent are configured to carry out the following steps: obtain data indicative of the physiological parameters from the sensor system; instruct the patient to control the physiological parameters; determine baseline levels for each of the physiological parameters; derive a score for each of the physiological parameters relative to the baseline levels, wherein the score increases in reaction to control of the respective physiological parameter; and use a weighting system to determine a weighting for the score associated with each of the physiological parameters and present a total score to the patient via the user interface, the total score consisting of a combination of the weighted scores, wherein the weighting system gives a higher weighting to the score for the physiological parameter that the patient most successfully controls and a lower weighting to the score for the physiological parameter that the patient least successfully controls.

2. A biofeedback system as claimed in claim 1, wherein the sensor system includes the one or more sensors for measurement of the patient’s physiological parameters.

3. A biofeedback system as claimed in claim 1 or 2, wherein the plurality of parameters comprises at least three parameters, with scores being derived for each of at least three parameters, with the weighting system determining a higher weighting for the most successful of the at least three parameters and a lower weighting for the least successful of the at least three parameters.

4. A biofeedback system as claimed in claim 1, 2 or 3, wherein the sensor system takes measurements of all three of muscle tension, body temperature and heart rate.

5. A biofeedback system as claimed in any preceding claim, wherein the biofeedback agent is arranged to determine baseline levels for the physiological parameters, wherein these baseline levels include a low baseline level and a high baseline level, with the range between the low baseline level and the high baseline level being a range that encloses normal values of the physiological parameters.

6. A biofeedback system as claimed in claim 5, wherein one of the baseline levels defines a target value, which is a value of the physiological parameter toward which the patient is instructed to control their body; and the other of the baseline levels that is not the target value defines an outer bound for normal variation of the physiological parameter.

7. A biofeedback system as claimed in claim 6, wherein the biofeedback agent is arranged to determine the score for each parameter using the size of the range between the baseline levels for that parameter along with the degree of success of the patient in controlling the parameter toward the target value.

8. A biofeedback system as claimed in claim 6 or 7, wherein the score is calculated based on proximity of the current value for the physiological parameter to the target value, with the other of the low or high baseline levels being used to normalise the score to allow the difference to the target value to be presented as a decimal or percentage value.

9. A biofeedback system as claimed in claim 8, wherein the score is calculated using the difference between the current value and the target value, divided by the difference between the high and low baseline levels.

10. A biofeedback system as claimed in any preceding claim, wherein the biofeedback agent is configured to set initial baseline levels for a new patient based on ranges for the physiological parameter that are expected to capture normal values for all patients.

11. A biofeedback system as claimed in claim 10, wherein the determination of the baseline levels is done adaptively with changes to the baseline levels taking account of at least one of the patient’s past performance and the patient’s on-going performance during a biofeedback session.

12. A biofeedback system as claimed in claim 11, wherein the baseline levels are adjusted compared to the initial baseline levels in order: to reduce the size of the range between the high and low baselines; to move a target value toward an expected achievable value for the patient; and/or to make a target value more difficult to achieve as the patient becomes more successful at controlling the physiological parameter.

13. A biofeedback system as claimed in claim 11 or 12, wherein the target value is adjusted based on the best value achieved by the patient for the relevant parameter, with the best being defined as that which is closest to the target value.

14. A biofeedback system as claimed in any preceding claim, wherein the biofeedback agent is configured to: update the scores continually during use of the biofeedback system, by periodically sampling the data to obtain a current value for the physiological parameter and calculating a score for the sampled value; and update the total score continually and presented it to the patient via the user interface to provide real-time feedback during their attempts to control the physiological parameters.

15. A biofeedback system as claimed in any preceding claim, wherein a degree of success of the patient is assessed based on the scores for each physiological parameter, with the success of the patient in controlling the parameters being ranked from highest to lowest based on the highest to lowest scores.

16. A biofeedback system as claimed in any preceding claim, wherein the weighting for the score is a multiplier for the score and the higher and lower weightings that are used to reflect the more and less successful control of parameters are implemented by addition of a constant to the weighting multiplier for the physiological parameter with the highest score, and subtraction of a constant from the weighting multiplier for the physiological parameter with the lowest score.

17. A biofeedback system as claimed in any preceding claim, wherein the weightings are adjusted periodically based on a prescribed time period and/or when changes in the scores indicates a change in the ranking of the physiological parameters by the patient’s success.

18. A biofeedback system as claimed in any preceding claim, wherein the biofeedback agent is configured to request and record patient data in connection with the patient’s usage of other medication or therapy and/or the patient’s symptoms, with the latter including one or more of: the incidence of migraine/headache; headache parameters; headache frequency; and premonitory symptoms.

19. A biofeedback system as claimed in claim 18, wherein the biofeedback agent is configured to identify trends in physiological parameters and/or symptoms, either alone or in combination, that link with increases and/or decreases in the risk of a migraine or of another headache.

20. A biofeedback system as claimed in claim 19, wherein the biofeedback agent is configured to present the patient with a prediction of likelihood of migraine or of another headache.

21. A biofeedback system as claimed in claim 18, 19 or 20, wherein the biofeedback agent is configured to provide guidance to the patient to prompt behavioural changes to reduce the risk of a future migraine or to reduce the risk of another form or headache.

22. A biofeedback system as claimed in claim 21, wherein the guidance to the patient takes the form of a selection of a particular physiological parameter as the focus for biofeedback.

23. A biofeedback system as claimed in any of claims 18 to 22, wherein the biofeedback agent comprises a cost function that is optimised taking account of performance criteria including symptoms and/or the incidence of migraine or other headache, with the cost function taking account of the success of the patient in particular sessions with reference to headache/migraine reduction and/or success of the patient in relation to particular individual parameters.

24. A biofeedback system as claimed in claim 23, wherein the biofeedback agent is configured to take a measure of the engagement of the patient and to react to this by adjusting the weighting and/or by incorporating the measure of engagement into the cost function.

25. A biofeedback system as claimed in claim 24, wherein the biofeedback system comprises a user engagement sensor for determining if the patient is attentive or not.

26. A biofeedback system as claimed in claim 25, wherein a camera of the personal computing device acts as the user engagement sensor, with the camera being used to provide images for facial recognition and/or eye-tracking.

27. A biofeedback system as claimed in claim 24, 25 or 26 wherein the biofeedback agent is arranged to interact with the patient in order to increase or decrease the degree of engagement.

28. A biofeedback system as claimed in any preceding claim, wherein the biofeedback agent is arranged to make changes in the user interface in relation to one or more of adjusting the information presented to the patient, giving feedback or prompts to the patient and/or presenting choices to the patient.

29. A biofeedback system as claimed in claim 28, wherein changes to the user interface are done to prompt improved biofeedback scores and/or to aid the patient in controlling a migraine or other headache.

30. A biofeedback system as claimed in any preceding claim, wherein the personal computing device is a mobile device such as a smartphone.

31. A method for operating a biofeedback system for headache patients, the biofeedback system comprising: a sensor system; a personal computing device; and a computer-implemented biofeedback agent; wherein the method comprises: using the sensor system for obtaining and transmitting data indicative of a plurality of physiological parameters of the patient, wherein the physiological parameters include at least two of muscle tension, body temperature, heart rate and heart rate variability; using the personal computing device, receiving data from the sensor system; and using the personal computing device together with the biofeedback agent to carry out the following steps: obtain data indicative of the physiological parameters from the sensor system; instruct the patient to control the physiological parameters; determine baseline levels for each of the physiological parameters; derive a score for each of the physiological parameters relative to the baseline levels, wherein the score increases in reaction to control of the respective physiological parameter; and use a weighting system to determine a weighting for the score associated with each of the physiological parameters and present a total score to the patient via the user interface, the total score consisting of a combination of the weighted scores, wherein the weighting system gives a higher weighting to the score for the physiological parameter that the patient most successfully controls and a lower weighting to the score for the physiological parameter that the patient least successfully controls.

32. A method as claimed in claim 31, comprising use of a biofeedback system as claimed in any of claims 1 to 0.

33. A computer programme product for a biofeedback system comprising a sensor system and a personal computing device, the computer programme product comprising instructions that, when executed, will provide the biofeedback system with a biofeedback agent and configure the biofeedback system such that it will: using the personal computing device, receive data from the sensor system; and using the personal computing device together with the biofeedback agent, carry out the following steps: obtain data indicative of the physiological parameters from the sensor system; instruct the patient to control the physiological parameters; determine baseline levels for each of the physiological parameters; derive a score for each of the physiological parameters relative to the baseline levels, wherein the score increases in reaction to control of the respective physiological parameter; and use a weighting system to determine a weighting for the score associated with each of the physiological parameters and present a total score to the patient via the user interface, the total score consisting of a combination of the weighted scores, wherein the weighting system gives a higher weighting to the score for the physiological parameter that the patient most successfully controls and a lower weighting to the score for the physiological parameter that the patient least successfully controls.

Description

[0076] Certain example embodiments of the present invention will now be described by way of example only and with reference to the accompanying drawings in which:

[0077] FIG. 1 shows a schematic for a biofeedback system;

[0078] FIG. 2 shows a schematic of a reward system for a biofeedback system;

[0079] FIG. 3 shows a communication flow overview for a biofeedback agent; and

[0080] FIG. 4 shows a core feedback loop of a biofeedback session.

[0081] A basic biofeedback setup, as shown schematically in FIG. 1, includes a sensor system 2 as well as a personal computing device in the form of a mobile device 4, in this case a smartphone 4. A biofeedback agent 6 is provided via software on the smartphone 4 and at a remote server 8, which also provides data storage for the biofeedback agent 6. The sensor system 2 comprises a plurality of sensors 10, 12, 14 for measuring physiological parameters in the form of muscle tension, body temperature, and heart rate. The selected sensors 10, 12, 14 all transmit signals via Bluetooth® Smart/4.0 to the smartphone. Muscle tension can be measured via a small compact bipolar surface electromyography sensor 10, such as the NeckSensor™ marketed by EXPAIN AS of Oslo, Norway. In this example the electromyography sensor 10 is used for measuring muscle tension from the upper trapezius muscle fibres. For body temperature measurement the sensor system 2 includes a finger temperature sensor 12, which are widely available from numerous sources. Typical sensors are either mounted to a fingertip or may be held by the patient. The heart rate can be measured via a heart rate sensor 14 such as a wristband sensor, for example a MIO Fuse™ heart rate wristband as supplied by Mio Global, Physical Enterprises. Such a sensor can be used to measure heart rate over the dorsal aspect of the wrist. A heart rate sensor can also supply measurements of heart rate variability, which may provide a further physiological parameter.

[0082] The smartphone 4 includes a user interface, primarily provided by the display screen 16, but also optionally using a speaker for audible feedback or prompts to the patient. When the biofeedback agent (described in more detail below) is in use then the display screen 16 can present the patient with a total score along with a realtime indication of the physiological parameters as measured by the sensors 10, 12, 14. The total score may be shown in graphical form, such as through a chart representing a percentage value or the like. The physiological parameters may be presented on the same screen or on a different screen to the total score, either shown separately or in combination, such as via bar charts or a “slider” display.

[0083] In use, the biofeedback agent 6 together with the smartphone 4 and sensor system 2 are used to obtain data for the physiological parameters, which are then used to generate scoring, such as via the example set out below. This scoring combines all of the physiological parameters (i.e. all three of heart rate, body temperature and muscle tension in this case), which contrasts to traditional biofeedback where a single parameter is used. This overcomes challenges that arise in that not all biofeedback patients experience an influence over the physiological parameter measured, and that different parameters may be useful to varying degrees for different patients. For instance, if a patient excels at raising their finger temperature, but has trouble lowering their heart rate, the currently proposed biofeedback agent algorithm will “fade out” the latter to some extent throughout the session and thereby generate a more appropriate and “personalized” total score for each individual patient. Comparably, the parameter that is most efficient for each patient, i.e. where they are most successful in controlling it will be given the heaviest weighting in the combined feedback score. It has been shown that biofeedback systems have greater benefits to the patient when the patient is able to improve their “score”. The proposed biofeedback system can automatically adapt to give each different patient the best opportunity to achieve this, by focussing on where they achieve the most success. Additionally, the ability for personalised and adaptive scoring of physiological parameters selected from a suitable set of multiple parameters adds significant benefits for patients that display greater variance in their physiological properties, which is a known issue for the pediatric population. The proposed biofeedback agent can hence provide robust therapist independent treatment, with particular benefits for pediatric migraine patients.

EXAMPLE

[0084] An example is set out below to demonstrate one implementation for the proposed biofeedback system, including calculations that may be used for the scoring and weighting systems. It will however be appreciated that the biofeedback system may be implemented in various ways, with alternatives compared to those in the specific example below.

[0085] A biofeedback agent 6 is software that enables personalised biofeedback for the individual user. An exemplary biofeedback agent 6 is described as follows with reference to FIGS. 2, 3 and 4, and is built to maximize the efficacy of biofeedback training, by using principles from motivational psychology, neurology and machine learning.

[0086] The underlying motivational psychology is described below, with reference to FIG. 2. The architectural components of the biofeedback agent are described with reference to FIG. 3. The following description provides the hierarchical architecture of the different components of the biofeedback agent 6 and the composition of these components.

Principles

[0087] The successful use of biofeedback, either in collaboration with a real or an artificial therapist 38 is predicated on acquiring voluntary control of bodily functions. A natural goal for a biofeedback agent 6 is then to ensure that such acquisition is optimized with regards to time of acquisition and churn, and that such voluntary control converge towards some threshold value. In addition, once voluntary control is acquired for an individual, the goal is to maintain and further develop such voluntary control over an extended period of time to reap its benefits.

[0088] Biofeedback is a process whereby electronic monitoring of a normally automatic bodily function is used to train someone to acquire voluntary control of that function. The above-mentioned principle is not restricted to any particular use case. However, in the present invention it is applied towards the reduction of headache frequency, duration and intensity. The concepts described in this section do not depend on this particular application, but is a general framework for applying motivational theory to minimize acquisition time and churn, and maximize engagement and bodily control.

Terminology and Notation

[0089] Two reward systems can be distinguished: in-session rewards 20 and between-session rewards 22. The purpose of the former is to increase engagement and promote correct behaviour during biofeedback training and the latter to maintain interest in the training for long enough to see a lasting effect from gaining voluntary control of bodily functions. This distinction creates a clear boundary between the inner workings of the biofeedback agent 6 during a session and after/between sessions.

[0090] FIG. 2 illustrates the two reward systems used by the biofeedback agent 6, which is described in more detail below.

In-Session Rewards

[0091] In-session rewards 20 provide motivation on a short time-scale, i.e. minutes, seconds and milliseconds, during biofeedback training. The in-session reward 20 can further be divided into two categories, motivation through self-control 24 and operant conditioning 26, as shown by FIG. 2.

Operant Conditioning

[0092] Operant conditioning 26 is a type of associative learning process through which the strength of a behaviour is modified by reinforcement or punishment. It is also a procedure that is used to bring about such learning. The behaviour modifying techniques described herein are positive reinforcement and positive punishment.

[0093] Positive reinforcement increase correct behaviour by introducing a rewarding stimulus, for example, the graphical visualization of a reduced Heart Rate when the user 50 manages to reduce it.

[0094] Positive punishment reduce wrong behaviour by introducing an aversive stimulus, for example, the graphical visualization of an increased Heart Rate.

[0095] The focus is to use feedback as positive reinforcement and punishment, and avoid guidance to the degree possible. The degree of guidance necessary to achieve bodily control is determined by a therapist 38 and will vary as the user 50 trains over time. The effect of guidance and the reason why it varies is covered below. Additional layers of positive reinforcement are introduced through: a weighting scheme, see below, that prioritize the bodily functions that the user 50 has the most success in controlling when forming a total score; adaptive baselines, see below, that modify the scoring of an individual bodily function based on performance; and In-session rewards for achieving short-term goals, determined through gamification.

Motivation Through Self-Control

[0096] Motivation through self-control is the result of experiencing increased voluntary control of bodily functions through self-learning. That is, users become aware of their ability to control bodily functions and finds out for themselves how they can affect how these bodily functions vary. The result is an immediate reward resulting from, for example, seeing the changes occur through graphical elements on a screen. This stands in contrast to the more basic principles set out above, which describe how data presented to the user 50 is modified to unconsciously motivate him.

Between-Session Rewards

[0097] Between-session rewards 22 provide motivation on a long time-scale, read days, weeks and months, between biofeedback training sessions. The between-session reward 22 system can further be divided into two main categories, personal training 28 and operant conditioning 30, as shown in FIG. 2.

Operant Conditioning

[0098] The same basic principle as described above holds when applied to a longer time-scale. The mechanism, however, by which it is applied is different. In addition to the behaviour modifying techniques described above, negative reinforcement is also consistently applied. Take the application of promoting headache reduction, for example. The presentation of a reduction in headache attacks reinforce the behaviour of performing biofeedback training. This provides the foundation for the long-term strategy of the biofeedback agent 6.

Personal Training

[0099] Personal training is the effect of feedback and guidance adapting to the individual user 50 over time. The intention is that such adaption makes the training more captivating by providing helpful and meaningful guidance. The purpose is to avoid extinction of the desired behaviour, to put it in terms of operant conditioning 30, which is the user 50 completing biofeedback training sessions regularly and gradually improving while doing so. By changing the degree of guidance stagnation is reduced, and it avoids the user 50 not knowing what to do if they do not manage to improve. The focus is therefore to vary the degree of guidance and the scaling of feedback to adapt the in-session experience based on long-term performance. This can be thought of as tweaking the difficulty of the biofeedback training based on performance. Additional elements that are included, but not described in detail for brevity, is: performance rewarded with increased responsibility and less guidance; motivating remarks based on performance; statistics and data visualization.

Biofeedback Agent

[0100] The biofeedback agent 6 is software built to instantiate the principles described above. FIG. 3 illustrates a set of five conceptual layers of the biofeedback agent, each with a separate set of functions. The layers are respectively, the core 30 (biofeedback training session software), middleware 32, 34, 36, (application specific), therapist 38 (Al), gamification 40 (UI) and logging/analytics 42 (diary). The first layer, the core 30, consists of the working software that provides user feedback during a biofeedback training session. The second layer, the middleware 32, 34, 36, is composed of multiple independent algorithms designed to derive characteristics of the user state (performance, engagement, etc.) based on different data sources. The third layer, the therapist 38 (or AI if you will), finds the optimal strategy for the user 50, and provides guidance to maximize user performance. The outermost layer is the gamification layer 40, which modifies the user interface 16 according to the strategy determined by the therapist 38. The final layer 42 is for logging and analysing data related to long-term performance, for example a headache diary as depicted in the instance of this architecture in FIG. 3.

Core

[0101] The algorithm of the core 30 is an adaptive weighting scheme. During a biofeedback session, the user 50 wears sensors 10, 12, 14 that measure a set of physiological variables. These physiological variables in this example include: Heart rate (HR), Heart rate variability (HRV), Peripheral finger temperature (Temp), and Neck tension (sEMG), but may include just two of these variables or more.

[0102] These physiological measurements are streamed to a native mobile application which derives an individual score, S.sub.i ∈ [0,100], for each of these measurements relative to a baseline. The baseline is computed individually for each user 50 and for each modality. The scores are then rated relative to each other with the purpose of deriving user performance. This rating lays the foundation for modifying the relative weighting between measurement scores when computing a total biofeedback score:

[00001]$W_{HR}+W_{HRV}+W_{Temp}+W_{sEMG}=1$

[00002]$W_{HR}{S}_{HR}+W_{HRV}{S}_{HRV}+W_{Temp}{S}_{Temp}+W_{sEMG}{S}_{sEMG}=S_{tot}$

where S.sub.tot is the total biofeedback score. The scores are computed using the equations:

[00003]$\frac{y_i-B_{i,low}}{B_{i,high}-B_{i,low}}\cdot 100=S_i$

[00004]$1-\frac{y_i-B_{i,low}}{B_{i,high}-B_{i,low}}\cdot 100=S_i$

where i is the physiological modality being scored. Equation (3) provide scores for values that are to be controlled to larger values such as y.sub.temp, whereas equation (4) provide scores for values that are to be controlled to smaller values such as y.sub.HR. The values of the weights initially define the physiological parameters the user 50 manages to control, or more precisely, the parameters the user 50 manages to control the most. That is, the weights are adjusted on the fly using a constant adaption rate k>0. The physiological parameter the user 50 manages to control the most is thereby given the most significance in the computation of S.sub.tot, see equation (2).

[0103] Let I be a nonempty set containing the indices for the physiological parameters measured during a biofeedback training session.

[0104] Let be the minimum and maximum scores of the physiological parameters measured at each time instant during the biofeedback training session. The weight of the physiological parameter corresponding to the maximum score is adjusted according to the formulae

[00005]$W_{i+1}=W_i+k$

with the minimum score adjusted according to

[00006]$W_{i+1}=W_i-k$

[0105] The weights have to be non-negative W.sub.i > 0 and saturated such that W.sub.i ≤ 1 . This is mainly to avoid divergence and numerical issues in equations (1)-(2).

[0106] The general set of equations are;

[00007]${\displaystyle {.Math.}_{i\in I}{W}_i=1}$

[00008]${\displaystyle {.Math.}_{i\in I}{W}_i{S}_i=S_{tot}}$

with I from above.

[0107] The individual scores are displayed and visualized during a biofeedback session to provide immediate feedback to the user 50. The goal of such a session is for the user 50 to try and maximize the score and control his physiological state. The scoring scheme itself is a means to help the user 50 achieve that target by creating a closed feedback loop and being a source of immediate motivation, see above. This closed feedback loop is illustrated in FIG. 4 in which the user’s physiological variables are measured by the sensor system 2, scored with reference to the baseline values and displayed to the user 50 via the user interface 16 of the personal computing device 4.

Middleware

[0108] The middleware 32, 34, 36 is a set of algorithms that independently estimate qualities related to the state of the user 50. Examples of such algorithms are performance estimators 34 and engagement estimators 36. As illustrated in FIG. 3, the middleware 32, 34, 36 can be composed of multiple independent algorithms. These algorithms are plugged in between the core 30 and the therapist 38, and receives inputs from the core and the long-term data logger 42. In the current example inputs to the middleware may include:

TABLE-US-00001 Inputs Notation Origin layer Weights W.sub.i Core Scores Si Core Headaches H.sub.N Diary Headache days D.sub.Nj Diary Symptoms n/a Diary Medication intake n/a Diary Session history n/a Diary

where headaches, headache days, symptoms, medication intakes and the session history are arrays of data going N days back in time. An exemplary stack of middleware algorithms include a headache predictor 32, a performance estimator 34, and an engagement estimator 36.

[0109] The remaining parts of this section will describe these algorithms in detail including how the middleware 32, 34, 36 can be composed to provide data to the therapist 38.

Headache Predictor

[0110] The headache predictor 32 uses Machine Learning to predict whether a headache will occur tomorrow given session results and longitudinal data from a headache diary 42. The state space of the output is Boolean, meaning that in practice someone either has a headache or they do not. The output from the headache predictor on the other hand is continuous, P.sub.1 ∈ [0,1], and describes the probability of a headache occurring tomorrow.

[0111] A verified implementation of this algorithm uses the Random Forest algorithm with boosting. The validation of other algorithms and the iterative adaption of such algorithms towards the user 50 is only limited by data. As the current results are based on an ensemble of different people, it is expected that training models on a fewer amount of people over a longer period will increase accuracy.

Performance Estimator

[0112] Performance can have multiple definitions in the context of biofeedback applied to people with headaches and migraines. The performance estimator 34 can relate to the user’s 50 ability to control physiological signals or the actual reduction of headaches. They describe how good the user 50 is at biofeedback itself and how good an effect the biofeedback, as it is currently being performed, has on headache/migraine reduction. The following can henceforth be derived to describe such effects: Headache Occurrence (HO) rate — The reduction in the number of headaches per month; Headache Duration (HD) rate —The reduction in the duration of headaches, on average, per month; Headache Intensity (HI) rate — The reduction in the intensity of headaches, on average, per month; Score Improvement rate — Improvement rate of scores for each modality and in total; Weight disparities - Disparity between all weights.

Engagement Estimator

[0113] Engagement defines how dedicated the user 50 is with respect to performing biofeedback. The engagement estimator 34 is important for knowing if the user needs an extra push, or whether the user interface 16 needs modification. Examples of an engagement measures are: Session adherence — Adherence rate to session schedule; User engagement rating — Rating R∈[1,5] provided by the user as answer to an engagement query. The engagement estimator 34 may include a sensor to determine a degree of engagement of the user, such as via use of the camera of a smartphone.

Therapist - Baseline Adaption

[0114] From equations (3)-(4) it is clear that the scores S.sub.i are highly dependent on the baselines. The baselines therefore determine how easy or difficult it is to reach a certain score during a biofeedback training session. The main issues when determining baselines are; how should they adapt to the user 50 and how to initialize them.

[0115] At first, the baselines are initialized by the therapist 38 to plausible physiological values that ensures that every user 50 using the device for the first time get a score in the middle range S.sub.i ∈ (30,70). These thresholds are determined heuristically based on the physiological parameter being measured. For example,B.sub.HR,.sub.low = 25bpm, B.sub.HR,.sub.high = 110bpm. During the first few sessions, these upper and lower thresholdsgradually close in around a baseline value. After X sessions, the user 50 will have exhibited a set of physiological measurements that can be classified as .sub.Yi,best (the definition of best can be varied, and is application dependent). For example, the median value of the 30 lowest heart rate measurements during the X first sessions.

[0116] With the above mentioned definition of .sub.Yi,.sub.best a naive iterative baseline update scheme which depends on whether equation (3) or (4) is used for score computation can be defined.

[0117] Case 1 - When using equation (3) B.sub.i,low is kept fixed and B.sub.i,.sub.high is adapted according to:

[00009]$B_{i,high}=\left(1+k\right)\cdot y_{i,best}$

where .sub.K>0 determines how aggressively B.sub.i,.sub.high moves.

[0118] Case 2 - When using equation (4) B.sub.i,.sub.high is kept fixed and B.sub.i,.sub.low is adapted according to:

[00010]$B_{i,low}=\left(1-k\right)\cdot y_{i,best}$

where .sub.K>0 determines how aggressively B.sub.i,low moves.

[0119] It will be appreciated that the above calculations can be implemented for multiple sessions with the biofeedback system being provided with the ability to learn from prior sessions, as well as to record and assess data relating to the patient’s performance, symptoms, medication and so on as discussed above. In this way the biofeedback system can be provided with means to operate as an enhanced form of “headache diary”, along with basic prediction/forecasting abilities based on identifying patterns in the patient’s behaviour that may link to increasing or decreasing likelihood of a headache or migraine.