Method for predicting a kinetic biomarker associated with autism spectrum disorder

Abstract

The present invention generally relates to methods and systems for predicting one or more a kinetic biomarkers associated with one or more neurodevelopmental disorder, such as autism spectrum disorder, by monitoring and analysing various movement behaviours. Specifically, the present invention can be applied to monitor and record data on atypical movement behaviours, such as tiptoe behaviour and axis-spinning, when observed in subjects including children and adults. An aspect of the invention relates to a computer-implemented method for predicting a kinetic biomarker associated with a neurodevelopmental disorder; the method comprising: - receiving a first data flow indicative of a movement and orientation of the lower leg of the subject; - receiving a second data flow indicative of a movement and orientation of the foot of the subject; - determining using at least one machine learning model that the first data flow and second data flow correspond to an atypical movement event; - determining, based on a frequency of occurrence of the atypical movement event during a predetermined time window, that the first data flow and the second data flow correspond to at least one kinetic biomarker associated with the neurodevelopmental disorder; and, - generating a report comprising the kinetic biomarker.

Description

[0001] METHOD FOR PREDICTING A KINETIC BIOMARKER ASSOCIATED WITH AUTISM SPECTRUM DISORDER

[0002] FIELD

[0003] The present invention relates to the field of computer-implemented methods, and more specifically to methods for predicting one or more kinetic biomarkers associated with neurodevelopmental disorders by monitoring and analysing movement behaviours. The invention further encompasses systems configured to perform these methods. It is applicable in both clinical and healthcare settings, as well as in naturalistic environments— such as in-home monitoring— to enhance behavioural assessment strategies for neurodevelopmental disorders.

[0004] BACKGROUND

[0005] Many behavioural studies involve monitoring the movements of patients. Of particular interest is the observation of combinations of atypical movement behaviours in children, especially those exhibiting neurodevelopmental disorders such as autism spectrum disorder (ASD).

[0006] Traditionally, this process relies on the manual interpretation of in-person observations or video recordings. Such sessions are typically conducted in specialised facilities designed for this purpose, which poses challenges for accurately assessing behaviours in naturalistic settings. This approach increases operational costs and requires the involvement of trained healthcare personnel. Moreover, the manual interpretation of movement data is labour-intensive and demands substantial time, costs and human resources.

[0007] Certain systems in the art aim to address some of these challenges. For example, US2021 / 259579 Al describes a wearable system for treating a child patient with idiopathic toe walking, which includes a wearable shoe insole, a first pressure sensor located at a heel region of the shoe insole, a second pressure sensor located at a front region of the shoe insole, a vibration actuator included within the shoe insole, and may also include an inertial measurement unit. The wearable system further includes a processor that receives data from the pressure sensors and / or the inertial measurement unit, and determines a gait pattern of the child patient based on the received data.

[0008] In another example, W02020 / 018886 Al describes a method for detecting a motor developmental delay and / or neurodevelopmental disorder of an infant that includes receiving motion data associated with the infant's gross motor activity; analyzing, using a machine learning algorithm, the motion data to detect a kinematic feature; comparing the kinematic feature to an expected relationship between the kinematic feature and infant age; and detecting the neurodevelopmental disorder based on the comparison.

[0009] There remains a need to develop systems and methods capable of monitoring and recording high-quality evidential data on atypical movement behaviours in children, to support the assessment of neurodevelopmental disorders such as ASD. In particular, there is a need for solutions that can function not only within controlled clinical settings but also in natural environments, such as during playtime at home, in school, or in daycare.

[0010] SUMMARY

[0011] It is an objective of the present invention to provide systems and methods capable of monitoring and recording high-quality evidential data of atypical movement behaviours in a subject, in order to facilitate the assessment of neurodevelopmental disorders, such as ASD. This objective can be achieved by predicting one or more a kinetic biomarkers associated with the one or more neurodevelopmental disorder, such as ASD, and by monitoring and analysing various movement behaviours. Specifically, the present invention enables the detection and recording of atypical behaviours such as tiptoe walking or spinning behaviour in subjects, including both children and adults.

[0012] One advantage of the invention lies in its ability to detect atypical movement behaviours more quickly and reliably in naturalistic environments where children engage in routine activities, free from clinical constraints. By capturing behavioural data in these settings, the system provides a more accurate and holistic view of the subject's motor patterns. This can lead to improved assessment strategies, enabling healthcare professionals to make more accurate diagnoses, initiate timely interventions, and deepen their understanding of the individual's behavioural profile.

[0013] Another advantage of the present invention is its cost-effectiveness. It may reduce clinic and staffing expenses by minimising the need for labour-intensive, manual interpretation of movement data, which typically demands significant time and human resources. In contrast to conventional monitoring systems that often rely on complex motion capture or imaging technologies, the invention offers a lower- complexity solution without compromising detection accuracy.

[0014] An aspect of the invention relates to a computer-implemented method for predicting a kinetic biomarker associated with a neurodevelopmental disorder; the method comprises: receiving a first data flow, generated by a first inertial measurement unit externally mounted onto a lower leg portion of the subject, the first data flow being indicative of a movement and orientation of the lower leg of the subject; receiving a second data flow, generated by a second inertial measurement unit externally mounted onto an upper foot portion of the subject, the second data flow being indicative of a movement and orientation of the foot of the subject; determining using at least one machine learning model that the first data flow and second data flow correspond to an atypical movement event, wherein the atypical movement event includes at least tiptoe behaviour and spinning behaviour; determining from a frequency of the atypical movement event occurring during a predetermined time window that the first data flow and second data flow correspond to at least one kinetic biomarker associated with neurodevelopmental disorder; generating a report comprising the kinetic biomarker.

[0015] In some embodiments, the first inertial measurement unit is mounted onto a lower portion of the shin of the subject's leg, preferably adjacent to an ankle of the subject, and / or the second inertial measurement unit is mounted onto the instep of the subject's foot, preferably around the middle of the instep of the subject's foot.

[0016] In some embodiments, the first and / or second inertial measurement units comprise an accelerometer configured to generate linear acceleration data, a gyroscope configured to generate angular velocity data, and a magnetometer configured to measure magnetic field strength data used to determine orientation in three directions.

[0017] In some embodiments, the first and / or second data flows are indicative of one or more rate, rate change, frequency, or amplitude of rotations of the foot of the subject.

[0018] In some embodiments, the machine learning model is a classifier model configured to classify the first and second data flow into categories representing 'tiptoe behaviour' and 'non-tiptoe behaviour,' and 'spinning behaviour' and 'non-spinning behaviour'; preferably, the classifier model is selected from decision trees, random forests, gradient boosting, and / or neural networks; more preferably, the machine learning model includes random forests.

[0019] In some embodiments, the method further comprises the step of receiving third data generated by a pressure measurement unit externally mounted onto a lower foot portion of the subject, preferably on the sole of the subject's foot, the second data flow being indicative of a pressure exerted by the foot of the subject during movement.

[0020] In some embodiments, the neurodevelopmental disorder includes autism spectrum disorder, attention- deficit / hyperactivity disorder, and / or developmental coordination disorder.

[0021] In some embodiments, the report comprises an overview of the frequency of the atypical movement event occurring during one or more predetermined time windows.

[0022] In some embodiments, the method further comprises timestamping the occurrence of an atypical movement event; and wherein the report comprises a timeline of the timestamped atypical movement event, with the same frequency of the measurement rate of the original flows.

[0023] In some embodiments, the method further comprises wirelessly transmitting the first data flow and the second data flow via a data link communicatively coupled to the first and second inertial measurement units to a processing unit configured for performing the steps of the method; preferably, the processing unit is a mobile device. Another aspect of the invention relates to a computer program comprising instructions which, when the program is executed by a computer, performs the method as described herein.

[0024] Another aspect of the invention relates to a kinetic biomarker prediction system, comprising a first inertial measurement unit adapted to be externally mounted onto a lower leg portion of the subject, and configured to generate data indicative of a movement and orientation of the lower leg of the subject; a second inertial measurement unit adapted to be externally mounted onto an upper foot portion of the subject, and configured to generate data indicative of a movement and orientation of the foot of the subject; at least one data processing unit comprising a memory configured to store computer-executable instructions, and at least one processor configured to access the memory and execute the computerexecutable instructions to: receive a first data flow generated by the first inertial measurement unit, the first data flow indicative of a movement and orientation of the lower leg of the subject; receive a second data flow generated by the second inertial measurement unit, the second data flow indicative of a movement and orientation of the foot of the subject; determine using at least one machine learning model that the first data flow and second data flow correspond to an atypical movement event, wherein the atypical movement event includes at least tiptoe behaviour and spinning behaviour; determine from a frequency of the atypical movement events over a period of time that the first data flow and second data flow correspond to a kinetic biomarker associated with a neurodevelopmental disorder; generate a report comprising the kinetic biomarker.

[0025] Another aspect of the invention relates to a method of training a machine learning model to receive a first and second data flows, and from the first and second data flows determine an atypical movement event, the method comprising the steps of: externally mounting a first inertial measurement unit onto a lower leg portion of a reference subject, the first inertial measurement unit being configured for generating the first data flow indicative of a movement and orientation of the lower leg of the reference subject; externally mounting a second inertial measurement unit onto an upper foot portion of the reference subject, the second inertial measurement unit being configured for generating the second data flow indicative of a movement and orientation of the foot of the reference subject; using the first and the second inertial measurement units to simultaneously acquire the first and the second data flow of the reference subject during movement, thereby obtaining reference data; processing the reference data to label one or more atypical movement events, thereby obtaining a ground truth dataset; wherein the atypical movement event includes at least tiptoe behaviour and / or spinning behaviour; inputting the first and the second data flow of the reference subject to a machine learning model; comparing the output of the machine learning model to the ground truth dataset; iteratively adjusting operation of the machine learning model based on the results of the previous comparing step, thereby obtaining a trained machine learning model.

[0026] In some embodiments, the trained machine learning model is applied to determine that the first data flow and second data flow correspond to the atypical movement event, preferably in another subject.

[0027] DESCRIPTION OF THE FIGURES

[0028] The following description of the figures relate to specific embodiments of the disclosure which are merely exemplary in nature and not intended to limit the present teachings, their application or uses.

[0029] Figure 1 is a block diagram showing a system according to certain aspects of the present disclosure.

[0030] Figure 2 represents Original, Spinning versus Non-Spinning, Typical Toe-to-Heel (TTB) versus Non-TTB, Typical versus TTB versus Spinning in typical development (TD) subjects.

[0031] Figure 3 represents Original, Spinning versus Non-Spinning, TTB versus Non-TTB, Typical versus TTB versus Spinning in ASD subjects.

[0032] Figure 4 represents a test Receiver Operating Characteristic (ROC) curve for spinning behaviour in TD subjects.

[0033] Figure 5 represents a validation ROC curve for spinning behaviour in TD subjects

[0034] Figure 6 represents a test ROC curve for TTB in TD subjects.

[0035] Figure 7 represents a validation ROC curve for TTB in TD subjects.

[0036] Figure 8 represents a test ROC curve for spinning behaviour in ASD subjects.

[0037] Figure 9 represents a validation ROC curve for spinning behaviour in ASD subjects.

[0038] Figure 10 represents a test ROC curve for TTB in ASD subjects.

[0039] Figure 11 represents a validation ROC curve for TTB in ASD subjects.

[0040] Figure 12 represents a sequences of behaviours spinning (black), TTB (light gray), and typical (dark grey) using the ASD-model.

[0041] Figure 13 represents a sequences of behaviours spinning (black), TTB (light gray), and typical (dark gray) using the TD-model.

[0042] It is noted that the use of reference numerals in the description and drawings is solely for the purpose of improving the clarity and understanding of the embodiments of the invention, and does not imply any limitation on the scope of protection. Similarly, the specific arrangements and configurations illustrated in the drawings are provided by way of example only and should not be construed as limiting the invention to the particular forms disclosed.

[0043] DESCRIPTION

[0044] The present description sets forth various embodiments of a (computer-implemented) method for predicting one or more kinetic biomarkers associated with neurodevelopmental disorders and related aspects, including a kinetic biomarker prediction system and a method of training a machine learning model for us in the method. These embodiments are provided by way of example only and are not intended to limit the scope of the invention, which is defined solely by the appended claims. For ease of understanding, reference is made throughout the description to the accompanying drawings and reference numerals, which illustrate exemplary configurations and components. This description is meant to aid the reader in understanding the technological concepts more easily, but it is not meant to limit the scope of the present disclosure, which is limited only by the claims.

[0045] An aspect of the invention provides a (computer-implemented) method for predicting a kinetic biomarker associated with a neurodevelopmental disorder, the method comprising the steps of:

[0046] - receiving a first data flow, generated by a first inertial measurement unit externally mounted onto a lower leg portion of the subject, the first data flow being indicative of a movement and orientation of the lower leg of the subject;

[0047] - receiving a second data flow, generated by a second inertial measurement unit externally mounted onto an upper foot portion of the subject, the second data flow being indicative of a movement and orientation of the foot of the subject;

[0048] - determining using at least one machine learning model that the first data flow and second data flow correspond to an atypical movement event, wherein the atypical movement event includes at least tiptoe behaviour and spinning behaviour ;

[0049] - determining from a frequency of the atypical movement event occurring during a predetermined time window that the first data flow and second data flow correspond to at least one kinetic biomarker associated with the neurodevelopmental disorder;

[0050] - generating a report comprising the kinetic biomarker.

[0051] As used herein, "neurodevelopmental disorder" refers to a condition that can impact the development and function of the nervous system, resulting in abnormalities in brain function which can affect emotions, learning abilities, self-control, and memory. Such disorders typically manifest early in development and may cause developmental deficits that lead to impairments in personal, social, academic, or occupational functioning. Examples of neurodevelopmental disorder include, but are not limited to, autism spectrum disorder (ASD), attention-deficit / hyperactivity disorder (ADHD), and / or developmental coordination disorder (DCD).

[0052] As used herein, the term "autism spectrum disorder" (ASD) refers to a specific neurodevelopmental disorder characterised by persistent deficits in social communication and social interaction, alongside restricted, repetitive patterns of behaviour, interests, or activities. ASD represents a spectrum of symptoms and severity levels, ranging from mild to severe, and may be associated with atypical sensory processing, motor coordination difficulties, and variations in intellectual functioning.

[0053] As used herein, a "kinetic biomarker" refers to a measurable characteristic or parameter associated with the dynamic aspects of a subject's movement or motion. In certain embodiments, the kinetic biomarker may include quantifiable data related to tiptoe behaviour, spinning behaviour, walking or running speed, movement variability, gait, posture, motor function, or other dynamic physiological processes.

[0054] As used herein, the term "inertial measurement unit (IMU)" refers to an electronic device comprising sensors such as accelerometers, gyroscopes, and magnetometers, which are respectively configured to measure linear acceleration, angular velocity, and magnetic field strength and orientation surrounding a body. By integrating data from these sensors, the IMU can determine the body's orientation, velocity, and positional changes, thereby enabling accurate navigation, orientation tracking, stabilisation, and control. In embodiments, the term "accelerometer" refers to a sensor configured to measure acceleration, typically along one or more axes, preferably the X, Y, and Z axes. It detects changes in speed and provides data regarding the acceleration forces acting on the subject.

[0055] In embodiments, the term "gyroscope" refers to a sensor configured to measure angular velocity, typically along one or more axes, preferably the X, Y, and Z axes, based on the principle of angular momentum. In exemplary embodiments, the gyroscope may include a micro-electromechanical system (MEMS) structure.

[0056] In embodiments, the term "magnetometer" refers to a sensor configured to measure the strength and / or direction of a magnetic field. It detects the presence of magnetic fields and provides data on their intensity and orientation. In exemplary embodiments, the magnetometer may include fluxgate magnetometers, proton precession magnetometers, and magneto-resistive magnetometers.

[0057] As used herein, the term "upper foot portion of the subject" refers to the anatomical region of a subject encompassing the upper surface of the foot, extending from the toes to the ankle joint. This region includes the dorsal surface of the foot, comprising the tarsal bones, metatarsal bones, and associated soft tissues. It further includes the dorsal aspects of the toes, the metatarsophalangeal joints, and the proximal phalanges. s used herein, the term "tiptoe behaviour" refers to a recurrent pattern of walking, standing, or running in which the individual predominantly places weight on the balls of the feet and toes, rather than maintaining the feet flat on the ground. This behaviour may manifest as walking with elevated heels and minimal to no contact between the heels and the ground. It may be potentially indicative of an underlying neurodevelopmental disorder, such as ASD.

[0058] As used herein, the term "spinning behaviour" refers to a movement pattern characterised by circular or rotational motion about an axis of the ankle or an axis corresponding to a pivoting motion of the subject. This behaviour may manifest as the subject spinning in circles or repeatedly rotating oneself. It is considered a stereotypical or self-stimulatory behaviour, referred to as stimming. It may be potentially indicative of an underlying neurodevelopmental disorder, such as ASD.

[0059] In embodiments, the first inertial measurement unit is mounted onto a lower portion of the shin of the subject's leg, preferably adjacent to an ankle of the subject, and / or wherein the second inertial measurement unit is mounted onto the instep of the subject's foot, preferably around the middle of the instep of the subject's foot.

[0060] As used herein, the term "shin" refers to the anterior part of the lower leg of the subject, located between the knee and the ankle. In specific embodiments, the lower portion of the shin refers to the lower half of this region, preferably the lower third, and more preferably the area immediately above the ankle.

[0061] As used herein, the term "instep" refers to the upper surface of the foot extending from the base of the toes to the ankle of the subject. The instep typically encompasses the arched middle portion of the foot, where the foot curves upward.

[0062] In embodiments, the first and / or second inertial measurement units comprise an accelerometer configured to generate linear acceleration data, a gyroscope configured to angular velocity data, and a magnetometer configured to measure magnetic field strength data used to determine orientation, in three directions.

[0063] In embodiments, the first and / or second data flows are indicative of one or more rate, rate change, frequency, or amplitude of rotations of the foot of the subject and optionally movement of the foot of the subject.

[0064] In embodiments, the machine learning model is a classifier model selected from decision trees, random forests, gradient boosting and / or neural networks; preferably wherein the machine learning model includes random forests. In a preferred embodiment, the classifier model is random forests, primarily for its ability to achieve an elevated level of accuracy in classification tasks while also demonstrating computational efficiency, particularly in comparison to other classifier models.

[0065] As used herein, the term "classifier model" refers to a type of machine learning model known in the art that is used to categorize input data into different classes or categories based on its features. A classifier model can be used to predict the class label of new, unseen data points after being trained on a prelabelled , manually labelled or classified dataset. Classifier models are trained using algorithms that learn patterns and relationships between input features and class labels, enabling them to make accurate predictions on new data.

[0066] As used herein, the term "random forests" refers to an ensemble learning technique constructing multiple decision trees during training. Each tree is typically built using a random subset of the training data and a random subset of features. The final prediction is made by aggregating the predictions of all individual trees.

[0067] As used herein, the term "gradient boosting" refers to an ensemble learning method building a series of weak learners sequentially. Each subsequent learner corrects the errors made by the previous ones, gradually improving the model's predictive performance. It often uses decision trees as weak learners.

[0068] As used herein, the term "neural networks" refers to a class of machine learning models comprising interconnected layers of artificial neurons that process input data through weighted connections and nonlinear activation functions to produce output predictions.

[0069] In embodiments, the classifier model is configured to classify the first and second data flow.

[0070] In preferred embodiments, the classifier model is configured to classify the first and second data flow into categories representing 'tiptoe behaviour and 'non tiptoe behaviour'.

[0071] In preferred embodiments, the classifier model is configured to classify the first and second data flow into categories representing 'spinning behaviour' and 'non spinning behaviour'.

[0072] In embodiments, the first data flow and second data flow are sampled during a time window of at least 1 second to at most 30 seconds, preferably 10 to 20 seconds, more preferably about 15 seconds.

[0073] In embodiments, the first data flow and second data flow are collected for a time period of at least 1 hour, at least 2 hours; at least 4 hours, at least 5 hours, during which time period atypical movement of the subject is determined. In embodiments, the first data flow and second data flow are collected for a time period of at most 12 hours, at most 11 hours; at most 10 hours, at most 9 hours, at most 8 hours, at most 7 hours, at most 6 hours, at most 5 hours during which time period atypical movement of the subject is determined. In embodiments, the first data flow and second data flow are collected for a time period of between about 1 hour and 12 hours, about 2 hour and 11 hours, about 3 hour and 10 hours, about 4 hour and 9 hours, about 5 hour and 8 hours during which time period atypical movement of the subject is determined.

[0074] In embodiments, the accelerometer is sampled at a frequency of at least 10 Hz to at most 100 Hz; of at least 15 Hz to at most 195 Hz, of at least 20 Hz to at most 85 Hz, of at least 25 Hz to at most 75 Hz, of at least 30 Hz to at most 70 Hz, of at least 35 Hz to at most 65Hz, of at least 40 Hz to at most 60 Hz, of at least 45 Hz to at most 55 Hz, of about 50 Hz. In embodiments, the accelerometer is sampled at a frequency preferably of at least 25 Hz to at most 75 Hz, more preferably of about 50 Hz. In embodiments, the gyroscope is sampled at a frequency of at least 10 Hz to at most 100 Hz; of at least 15 Hz to at most 195 Hz, of at least 20 Hz to at most 85 Hz, of at least 25 Hz to at most 75 Hz, of at least 30 Hz to at most 70 Hz, of at least 35 Hz to at most 65Hz, of at least 40 Hz to at most 60 Hz, of at least 45 Hz to at most 55 Hz, of about 50 Hz. In embodiments, the gyroscope is sampled at a frequency preferably of at least 25 Hz to at most 75 Hz, more preferably of about 50 Hz.

[0075] In embodiments, the magnetometer is sampled at a frequency of at least 10 Hz to at most 100 Hz; of at least 15 Hz to at most 195 Hz, of at least 20 Hz to at most 85 Hz, of at least 25 Hz to at most 75 Hz, of at least 30 Hz to at most 70 Hz, of at least 35 Hz to at most 65Hz, of at least 40 Hz to at most 60 Hz, of at least 45 Hz to at most 55 Hz, of about 50 Hz. In embodiments, the magnetometer is sampled at a frequency preferably of at least 25 Hz to at most 75 Hz, more preferably of about 50 Hz.

[0076] In embodiments, the accelerometer, the gyroscope and / or the magnetometer are sampled at a frequency of at least 10 Hz to at most 100 Hz; preferably 25 Hz to 75 Hz, more preferably about 50 Hz. In any of the above embodiments, the sampling frequency is advantageously chosen to effectively capture all distinct movements while simultaneously minimizing the required data flow transfer. This selection ensures that the system operates efficiently, thereby enhancing its effectiveness in capturing and analysing movement data flow associated with a neurodevelopmental disorder or other relevant applications.

[0077] In embodiments, the method may further comprise the step of receiving a third data flow generated by a pressure measurement unit externally mounted onto a lower foot portion of the subject, preferably on the sole of the subject's foot, the third data flow being indicative of a pressure exerted by the foot of the subject during movement. Measuring the pressure exerted by the foot on the ground offers a valuable advantage, providing insights into weight distribution during both standing and movement. This offers objective observations of motor behaviour, particularly beneficial in identifying atypical gait patterns, which are frequently observed in individuals with a neurodevelopmental disorder.

[0078] In embodiments, the third data flow is sampled during a time window of at least 1 second to at most 30 seconds, preferably 10 to 20 seconds, more preferably about 15 seconds.

[0079] In embodiments, the third data flow is collected for a time period of at least 1 hour, at least 2 hours; at least 4 hours, at least 5 hours, during which time period atypical movement of the subject is determined. In embodiments, the third data flow is collected for a time period of at most 12 hours, at most 11 hours; at most 10 hours, at most 9 hours, at most 8 hours, at most 7 hours, at most 6 hours, at most 5 hours during which time period atypical movement of the subject is determined. In embodiments, the third data flow is collected for a time period of between about 1 hour and 12 hours, about 2 hour and 11 hours, about 3 hour and 10 hours, about 4 hour and 9 hours, about 5 hour and 8 hours during which time period atypical movement of the subject is determined. As used herein, the term "pressure measurement unit", preferably a pressure sensor, refers to a transducer device detecting and measuring the force or pressure applied by the foot onto a surface, such as the ground or a footpad, during various movements. This pressure measurement unit can detect and quantify the pressure distribution across the foot, converting it into an electrical signal. This signal can then be analysed to provide information into foot biomechanics, gait analysis, and weight distribution on the foot.

[0080] In embodiments, the method further comprises generating a trigger signal responsive to detection of the atypical movement event in the first data flow and the second data flow, said trigger signal being configured to activate an actuator configured to provide a sensory alert to the subject, indicating the occurrence of the atypical movement event.

[0081] Preferably, the method comprises generating a trigger signal responsive to detection of tiptoe behaviour in the first data flow and the second data flow, said trigger signal being configured to activate an actuator configured to provide a sensory alert to the subject, indicating the occurrence of the tiptoe behaviour.

[0082] Preferably, the method comprises generating a trigger signal responsive to detection of spinning behaviour in the first data flow and the second data flow, said trigger signal being configured to activate an actuator configured to provide a sensory alert to the subject, indicating the occurrence of the spinning behaviour.

[0083] Alerting the subject to the occurrence of an atypical movement event serves as an educational tool to increase awareness and discourage further engagement in such movements. This tool can be utilized within a low-impact intervention therapy, following guidance from healthcare professionals. By using this method, subjects can learn to halt atypical movements and prevent potential physiological injuries, thereby supporting corrective neurodevelopmental growth.

[0084] In further embodiments, the actuator comprises a haptic feedback device configured to provide haptic feedback to the subject upon receiving the trigger signal. Preferably, the haptic feedback device comprises an eccentric rotating mass motor (ERM). Alternatively, or in combination, the actuator may implement other forms of sensory alerts, such as an auditory signal.

[0085] In embodiments, the report may comprise an overview of the time period for detecting the atypical movement (e.g. the time period during which the device was used), the frequency and / or duration of the atypical movement, such as tiptoe behaviour and / or spinning behaviour, occurring during one or more predetermined time windows.

[0086] In embodiments, the report may comprise an overview of tiptoe behaviour occurring at a frequency of at least 1, 2, 3, 4 or 5 times per 5 minutes, 10 minutes, 15 minutes, 20 minutes, 25 minutes, 30 minutes, 35 minutes, 40 minutes, 45 minutes, 50 minutes, 55 minutes or 60 minutes, demonstrating an increased likelihood of predicting a kinetic biomarker as defined herein In embodiments, the report may comprise an overview of spinning behaviour occurring at a frequency of at least 1, 2, 3, 4 or 5 times per 5 minutes, 10 minutes, 15 minutes, 20 minutes, 25 minutes, 30 minutes, 35 minutes, 40 minutes, 45 minutes, 50 minutes, 55 minutes or 60 minutes demonstrating an increased likelihood of predicting a kinetic biomarker as defined herein

[0087] In embodiments, the report may comprise an overview of both tiptoe behaviour and spinning behaviour each occurring at a frequency of at least 1, 2, 3, 4 or 5 times per 5 minutes, 10 minutes, 15 minutes, 20 minutes, 25 minutes, 30 minutes, 35 minutes, 40 minutes, 45 minutes, 50 minutes, 55 minutes or 60 minutes demonstrating an increased likelihood of predicting a kinetic biomarker as defined herein.

[0088] In embodiments, the report may comprise an overview of tiptoe behaviour detected during at least 20%, at least 25%, at least 30%, at least 40%, at least 45%, at least 50, at least 55% of the detection time show an increased likelihood of a predicting a kinetic biomarker as defined herein .

[0089] In embodiments, the report may comprise an overview of spinning behaviour detected during at least 20%, at least 25%, at least 30%, at least 40%, at least 45%, at least 50, at least 55% of the detection time show an increased likelihood of predicting a kinetic biomarker as defined herein

[0090] In embodiments, the report may comprise an overview of tiptoe behaviour and spinning behaviour detected during at least 20%, at least 25%, at least 30%, at least 40%, at least 45%, at least 50, at least 55% of the detection time show an increased likelihood of predicting a kinetic biomarker as defined.

[0091] In this context, the term "increased likelihood" refers to a higher probability of predicting a kinetic biomarker as defined herein or exhibiting features typically associated with an atypical movement event . It can indicate the presence of tiptoe behaviour and / or spinning behaviour at the above specified frequencies and / or percentages is associated with the possibility that the subject has a neurodevelopmental disorder, or exhibits features typically associated with a neurodevelopmental disorder. Nonetheless, it should be understood that a diagnosis of a neurodevelopmental disorder, should be performed by a clinical professional qualified to make a clinical diagnosis of neurodevelopmental disorders. Accordingly, aspects of the herein described invention may function as an assistive tool for assessing a likelihood of a neurodevelopmental disorder , optionally in combination other methods known in the art, rather than a diagnostic tool.

[0092] In embodiments, the report may further comprise a timestamp during which spinning behaviour and / or tiptoe behaviour was detected.

[0093] In embodiments, the report may comprise an overview of the pressure exerted by the foot of the subject on the pressure measurement unit during movement. In embodiments, the report further comprises pressure detected during at least 20%, at least 25%, at least 30%, at least 40%, at least 45%, at least 50, at least 55% of the detection time show an increased likelihood of predicting a kinetic biomarker as defined herein. In embodiments, the report may comprise a pressure measurement occurring at a frequency of at least 1, 1, 3, 4 or 5 times per 5 minutes, 10 minutes, 15 minutes, 20 minutes, 25 minutes, 30 minutes, 35 minutes, 40 minutes, 45 minutes, 50 minutes, 55 minutes or 60 minutes show an increased likelihood of predicting a kinetic biomarker as defined herein.

[0094] In embodiments, the method may further comprise timestamping the occurrence of an atypical movement event; and wherein the report comprises a timeline of the timestamped atypical behaviour event, preferably with the same frequency of the measurement rate of the original data flows. An advantage of this timeline is that it enhances the accuracy of behavioural analysis and supports more precise monitoring.

[0095] In embodiments, receiving the first data flow and receiving the second data flow, and optionally the third data flow, comprises wirelessly sending the first data flow and the second data flow, and optionally the third data flow, to a mobile device such as a telephone in wireless communication with a data link coupled to the first and second IMU, and optionally the pressure measurement unit.

[0096] Another aspect of the present disclosure relates to a kinetic biomarker analysis system, comprising

[0097] - a first inertial measurement unit adapted to be externally mounted onto a lower leg portion of the subject, and configured to generate data indicative of a movement and orientation of the lower leg of the subject;

[0098] - a second inertial measurement unit adapted to be externally mounted onto an upper foot portion of the subject, and configured to generate data indicative of a movement and orientation of the foot of the subject;

[0099] - at least one data processing unit comprising a memory configured to store computer-executable instructions, and at least one processor configured to access the memory and execute the computerexecutable instructions to:

[0100] - receive a first data flow generated by the first inertial measurement unit, the first data flow indicative of a movement and orientation of the lower leg of the subject;

[0101] - receive a second data flow generated by the second inertial measurement unit, the second data flow indicative of a movement and orientation of the foot of the subject;

[0102] - determine using at least one machine learning model that the first data flow and second data flow correspond to an atypical movement event, wherein the atypical movement event includes at least tiptoe behaviour and spinning behaviour;

[0103] - determine from a frequency of the atypical movement events over a period of time that the first data flow and second data flow correspond to kinetic biomarker associated with a neurodevelopmental disorder;

[0104] - generate a report comprising the at least one kinetic biomarker. As used herein, the term "data processing unit" refers to a component or system within a computing device responsible for handling data-related tasks. It may encompass various hardware components such as processors, memory modules, input / output interfaces, and specialized circuits.

[0105] In embodiments, the system comprises at least two data processing units embedded within a sock, each connected to a read-out system. In embodiments, this readout system can be attached to the sock as a removable wearable, facilitating convenient data retrieval and transmission. It retrieves data from both the data processing unit and transmits it wirelessly to a smartphone or other compatible devices. Equipped with its own data processing unit, for example a central processing unit (CPU), the read-out system incorporates a wireless module for Bluetooth signal transmission.

[0106] In certain embodiments the report can be used for assessing a likelihood of subject having a neurodevelopmental disorder, which refers to the assessment of the probability that a subject displays a kinetic biomarkers associated with the neurodevelopmental disorder or at least exhibits atypical movement features associated with the neurodevelopmental disorder based on the displayed movement characteristics being consistent with those typically observed in individuals with having the neurodevelopmental disorder. Assessing likelihood may entail analysing data collected from sensors and comparing it to established neurodevelopmental disorder-associated patterns or profiles. A higher likelihood can indicate a stronger correlation between the observed movement patterns and characteristics in individuals with a neurodevelopmental disorder. The report can provide insights into the subject's movement patterns and kinetic biomarkers associated with a neurodevelopmental disorder. It may serve as an assistive tool for (healthcare) professionals and caregivers to identify individuals who may benefit from additional assessment, intervention, or tailored support according to their specific needs. Thus, the report can function as an assistive report, not a diagnostic report.

[0107] In embodiments, the system may further comprise a garment wearable on the foot and at least a lower leg portion of a subject; wherein the garment wearable is a sock having a lower leg portion and a foot portion; wherein the first inertial measurement units is embedded within a fabric at a frontal part of the sock's lower leg portion; and wherein the second inertial measurement units is embedded within a fabric at an upper part of the sock's foot portion. In embodiments, the sock comprises a pressure measurement unit, preferably externally, mounted onto a lower foot portion of the subject, preferably on the sole of the subject's foot. Incorporating a pressure measurement unit onto the lower foot portion of the sock improves the system's capabilities by providing supplementary data on weight distribution and pressure exertion during both standing and movement for a more precise assessment.

[0108] As further shown in Figure 1, the system can comprise a garment 1, preferably a sock, having a lower leg portion 2 and a foot portion 3. The system also comprises the first inertial measurement unit which is secured onto a lower leg portion of the subject, preferably externally secured onto a lower leg portion of the subject. The first IMU is configured to generate data flow (the first data flow) 4 indicative of a movement and orientation of the lower leg of the subject. Furthermore, the system comprises the second inertial measurement unit which is secured onto an upper foot portion of the subject, preferably externally secured onto an upper foot portion of the subject. The second IMU is configured to generate data flow (the second data flow) 5 indicative of a movement and orientation of the foot of the subject. The first 4 and second data flow 5 preferably comprise data obtained from the accelerometer, the magnetometer and the gyroscope. The system can further comprise a pressure measurement unit (10) which secured onto a lower foot portion of the subject, preferably on the sole of the subject's foot. The pressure measurement unit is configured to generate data flow (the third data flow 11) indicative of a pressure exerted by the foot of the subject during standing and / or movement. The first 4 and second 5 data flow, and optionally the third data flow 11, is collected during a time period 6. The first and second data flow, and optionally the third data flow, can be used as training data flow 7 for training the machine learning model. This first and second data flow, and optionally the third data flow, can be send to the machine learning model 8, directly, after collecting the data flow or after training. Preferably, the machine learning model is a random forest classifier. The machine learning model can determine from a frequency of the atypical movement events, such as tiptoe behaviour and spinning behaviour, within a period of time 9.

[0109] In embodiments, the garment wearable is seamless to ensure a smooth and uninterrupted surface, minimizing friction against the skin and enhancing overall comfort of the patient. In embodiments, the garment wearable can be made of a foldable technical fabric such as, but not limited to, Acrylonitrile Butadiene Styrene (ABS), High-Impact Polystyrene (HIPS), High-Density Polyethylene (HDPE), Polyvinyl Chloride (PVC), Polyethylene Terephthalate (PET), Thermoplastic Polyolefin (TPO), and the like. Additionally, the garment can be adorned in dull and uniform colours, imparting a discreet appearance. Preferably, the garment is made of soft fabric, to provide a gentle and tactile experience, prioritizing comfort and minimizing irritation.

[0110] In embodiments, the system may further comprise at least one first processing unit integrated into the garment wearable and configured to perform at least a portion of the steps of the method; and at least one second processing unit external to the garment wearable and configured to perform at least the remaining portion of the steps of the method.

[0111] In embodiments, the system may further comprise a power supply for supplying power to the first and second inertial measurement unit. In embodiments, the power supply can also supply power to the pressure measurement unit and / or the data processing unit. In embodiments, the power supply is electrically connected to provide power to the first and second inertial measurement unit and optionally the pressure measurement unit and / or the data processing unit. In a preferred embodiment, the power supply comprises a battery, such as an ultra-thin rechargeable lithium polymer battery, preferably located at the readout system in the sock.

[0112] In embodiments, the system may further comprise a sensory alert unit configured to provide a sensory alert to the subject responsive to detection of the atypical movement event in the first data flow and the second data flow. The sensory alert unit may comprise an actuator that is configured to receive a trigger signal generated by the processing unit detecting the occurrence of the atypical movement event in the first data flow and the second data flow. Preferably, the atypical movement event includes the occurrence of tiptoe behaviour and / or spinning behaviour.

[0113] In further embodiments, the sensory alert unit comprises a haptic feedback device configured to provide haptic feedback to the subject upon receiving the trigger signal. Preferably, the haptic feedback device comprises an eccentric rotating mass motor (ERM). The use of a haptic feedback device provides low- impact feedback to the subject, thereby increasing awareness and discourage further engagement in such movements without causing discomfort or annoyance. Alternatively, or in combination, the sensory alert unit may comprise other forms of sensory device, such as an auditory device (e.g. speaker).

[0114] In embodiments, the system may further comprise a datalink, preferably a readout system, configured to transmit the first and second data flow, and optionally the third data flow, from the first and second inertial measurement units, and optionally the pressure measurement unit, to the data processing unit; preferably wherein the datalink comprises a wireless transmitter in wireless communication with the data processing unit. In embodiments, a datalink is connected to the first and second inertial measurement units, and optionally the pressure measurement unit, through connections embedded in the sock. In embodiments, the system comprises at least two data processing units attached to the sock, each connected to a read-out system.

[0115] In embodiments, this readout system can be attached to the sock facilitating convenient data retrieval and transmission. The readout system can retrieve data from the first and / or second inertial measurement units, and optionally the pressure measurement unit, and transmit it the first and / or second data flow, and optionally the third data flow, wirelessly to a smartphone or other compatible devices. The smartphone or other compatible device can process the first and second data flow, and optionally the third data flow, received from the readout system(s) using a machine learning model and provide a report (e.g. digital) on the detected atypical movement event(s) and kinetic biomarker(s) as defined herein. Advantageously, smartphones and other compatible devices offer greater processing power and memory capacity compared to the components mounted onto the sock, facilitating efficient execution of machine learning algorithms. Moreover, these devices provide ample storage capacity, allowing for storage of larger datasets, and long-term tracking. Alternatively or in combination, the readout system can be equipped with its own data processing unit, for example a central processing unit, for processing the first and / or second data flow, and optionally the third data flow, prior to transmitting it to a smartphone or other compatible devices. The data processing unit of the readout system can process the first and / or second data flow, and optionally the third data flow, using a machine learning model and transmit a report (digital) on the detected atypical movement event(s) and kinetic biomarker(s) as defined herein to a smartphone or other compatible devices.

[0116] In embodiments, the read-out system can incorporate a wireless module for Bluetooth signal transmission to the smartphone or other compatible devices. Advantageously, local processing on the data processing unit on the garment wearable streamlines the user experience and enhances convenience. Additionally, processing data on the data processing unit is more power-efficient than transmitting raw sensor data wirelessly to an external device for processing. This efficiency can extend battery life.

[0117] Another aspect of the present disclosure relates to a method of training a machine learning model, wherein the determining comprises:

[0118] - using the one or more test parameters to generate (by pre-processing) a reference dataset,

[0119] - applying the test dataset to a trained model algorithm to generate one or more (inferred) test predictions, wherein the trained model algorithm has been generated by:

[0120] - training an untrained model algorithm using multiple training records, wherein a training record comprises:

[0121] - a training dataset determined from (by pre-processing) one or more record parameters measured from a reference subject during a record session, wherein said training dataset comprises a first data flow generated by a first inertial measurement unit externally mounted onto a lower leg portion of the subject, the first data flow indicative of a movement and orientation of the lower leg of the subject; and a second data flow generated by a second inertial measurement unit externally mounted onto an upper foot portion of the subject, the second data flow indicative of a movement and orientation of the foot of the subject;

[0122] - an outcome of atypical movement event, wherein said atypical movement event includes at least tiptoe behaviour and spinning behaviour, of the same reference subject for the same record session,

[0123] - outputting said one or more (inferred) atypical movement event predictions.

[0124] Pre-processing typically refers to the initial stage of data flow preparation in which raw data undergoes various operations and transformations to make it suitable for further analysis or modelling. As used herein, the term "pre-processing" involves manually assigning specific behaviours to predetermined time windows. Preprocessing can involve manually assigning specific behaviour to creation annotated files that can link behavioural observations with raw measurements, facilitating the training of machine learning models by cleaning and filtering data flow aimed at enhancing the quality and utility of the data flow for subsequent analysis or modelling tasks.

[0125] In some embodiments, the method may comprise processing the raw reference data to obtain a set of feature vectors, wherein the feature vectors comprise one or more statistical or kinematic parameters derived from the first data flow and / or the second data flow; and annotating the feature vectors with corresponding labels indicating one or more atypical movement events, wherein the atypical movement events include at least tiptoe behaviour and spinning behaviour, thereby generating the ground truth dataset.

[0126] As used herein, the term "feature vector" refers to an ordered set of numerical values that represent measurable characteristics extracted from raw sensor data, such as the data generated by one or more inertial measurement units. Each feature vector may corresponds to a defined time segment of sensor input and is used as input to a machine learning model for classification or prediction tasks.

[0127] In embodiments, the feature vector comprises one or more statistical, temporal, or kinematic parameters derived from the data flow generated by the IMUs. These features may be extracted from one or more axes of the accelerometer, gyroscope, and / or magnetometer signals as described herein, and may be calculated over sliding or fixed time windows. The feature vectors serve as structured, model-ready inputs that allow the machine learning model to distinguish between typical and atypical movement behaviours. Examples of feature vector may include, but are not limited to statistical metrics such as mean, median, variance, standard deviation, peak values or zero-crossing rates; frequency-domain features, temporal features such as step duration, stride frequency, or rotation periodicity; derived quantities such as jerk (rate of change of acceleration), pitch, roll, or yaw angles; symmetry measures or inter-limb coordination metrics.

[0128] In some embodiments, the method may comprise inputting the feature vectors into an untrained machine learning model; and training the machine learning model by iteratively adjusting internal parameters of the machine learning model based on a comparison of the model output to the ground truth dataset, thereby obtaining a trained machine learning model.

[0129] Reference throughout this specification to "one embodiment" or "an embodiment" means that a particular feature, structure or characteristic described in connection with the embodiment is included in at least one embodiment of the present disclosure. Thus, appearances of the phrases "in one embodiment" or "in an embodiment" in various places throughout this specification are not necessarily all referring to the same embodiment.

[0130] As used herein, the terms "comprising", "comprises" and "comprised of" as used herein are synonymous with "including", "includes" or "containing", "contains", and are inclusive or open-ended and do not exclude additional, non-recited members, elements or method steps. The terms "comprising", "comprises" and "comprised of" when referring to recited members, elements or method steps also include embodiments which "consist of" said recited members, elements or method steps. The singular forms "a", "an", and "the" include both singular and plural referents unless the context clearly dictates otherwise.

[0131] As used herein, relative terms, such as "left," "right," "front," "back," "top," "bottom," "over," "under," etc., are used for descriptive purposes and not necessarily for describing permanent relative positions. It is to be understood that such terms are interchangeable under appropriate circumstances and that the embodiment as described herein are capable of operation in other orientations than those illustrated or described herein unless the context clearly dictates otherwise.

[0132] Objects described herein as being "adjacent" to each other reflect a functional relationship between the described objects, that is, the term indicates the described objects must be adjacent in a way to perform a designated function which may be a direct ( / .e. physical) or indirect ( / .e. close to or near) contact, as appropriate for the context in which the phrase is used.

[0133] Objects described herein as being "connected" or "coupled" reflect a functional relationship between the described objects, that is, the terms indicate the described objects must be connected in a way to perform a designated function which may be a direct or indirect connection in an electrical or nonelectrical ( / .e. physical) manner, as appropriate for the context in which the term is used.

[0134] As used herein, the term "substantially" refers to the complete or nearly complete extent or degree of an action, characteristic, property, state, structure, item, or result. For example, an object that is "substantially" enclosed would mean that the object is either completely enclosed or nearly completely enclosed. The exact allowable degree of deviation from absolute completeness may in some cases depend on the specific context. However, generally speaking the nearness of completion will be so as to have the same overall result as if absolute and total completion were obtained. The use of "substantially" is equally applicable when used in a negative connotation to refer to the complete or near complete lack of an action, characteristic, property, state, structure, item, or result.

[0135] As used herein, the term "about" is used to provide flexibility to a numerical value or range endpoint by providing that a given value may be "a little above" or "a little below" said value or endpoint, depending on the specific context. Unless otherwise stated, use of the term "about" in accordance with a specific number or numerical range should also be understood to provide support for such numerical terms or range without the term "about". For example, the recitation of "about 30" should be construed as not only providing support for values a little above and a little below 30, but also for the actual numerical value of 30 as well.

[0136] The recitation of numerical ranges by endpoints includes all numbers and fractions subsumed within the respective ranges, as well as the recited endpoints. Furthermore, the terms first, second, third and the like in the description and in the claims, are used for distinguishing between similar elements and not necessarily for describing a sequential or chronological order, unless specified. It is to be understood that the terms so used are interchangeable under appropriate circumstances and that the embodiments of the disclosure described herein are capable of operation in other sequences than described or illustrated herein.

[0137] Reference in this specification may be made to devices, structures, systems, or methods that provide "improved" performance (e.g. increased or decreased results, depending on the context). It is to be understood that unless otherwise stated, such "improvement" is a measure of a benefit obtained based on a comparison to devices, structures, systems or methods in the prior art. Furthermore, it is to be understood that the degree of improved performance may vary between disclosed embodiments and that no equality or consistency in the amount, degree, or realization of improved performance is to be assumed as universally applicable.

[0138] In addition, embodiments of the present disclosure may include hardware, software, and electronic components or modules that, for purposes of discussion, may be illustrated and described as if the majority of the components were implemented solely in hardware. However, one of ordinary skill in the art, and based on a reading of this detailed description, would recognize that, in at least one embodiment, the electronic based aspects of the present disclosure may be implemented in software (e.g., instructions stored on non-transitory computer-readable medium) executable by one or more processing units, such as a microprocessor and / or application specific integrated circuits. As such, it should be noted that a plurality of hardware and software-based devices, as well as a plurality of different structural components may be utilized to implement the technology of the present disclosure. For example, "servers" and "computing devices" described in the specification can include one or more processing units, one or more computer-readable medium modules, one or more input / output interfaces, and various connections connecting the components.

[0139] EXAMPLES

[0140] Examples of an implementation of the technology according to the present disclosure is given hereinbelow. The provision of examples is meant to aid the reader in understanding the technological concepts more easily, but it is not meant to identify the most important or essential features thereof, nor is it meant to limit the scope of the present disclosure.

[0141] Example 1: Experimental set up

[0142] To set up the classification model, seven children performed different walking patterns with movement sensors incorporated in a sock and attached to three well-specified foot locations i.e., shin, and instep. During the data acquisition using the I MU -attached system, the behaviour of the children has also been recorded using a video system. In order to generate a reliable dataset for training and testing the model, the time series are annotated by manually assigning a given behaviour to a specific time interval.

[0143] After collecting the experimental data, RF models were generated to predict the classification performance as a function of the sensor outputs (given 3 axes of the 3 sensors), the optimized location or locations on the foot, and different sampling time intervals for each child. The high accuracy is already obtained by means of visual inspection and cross-validation at the initial phase of the experimental setup. Based on the highest classification accuracy, the sensors were installed on two locations on the foot, more specifically the instep and the shin. For further data acquisition, both sensors simultaneously generated data.

[0144] For further validation of the system and generated model, a statistical analysis was carried out to validate the optimal foot location and sampling time interval for reaching (i) maximal and (ii) robust classification accuracy; robust performance means a set of location and time, so that predicted accuracy is independent of movement variation among children. Linear multiple regression models were generated with output parameters (i) accuracy and (ii) accuracy variance among children. Besides foot location / combination & time interval, also the factor "child" was an input factor, and, as it is a random effect, mixed linear regression models must be used. The statistical assessment confirmed the earlier hypothesis i.e., that the instep-shin sensor combination results in maximal classification accuracy. In addition, it is also shown that performance variance among children is minimal, so the instep-shin location also is a robust setting.

[0145] The final model allows the automatic detection of tiptoe behaviour and spinning behaviour using data originating from two IMU sensors positioned on the instep and shin.

[0146] Example 2: Training machine learning model

[0147] Example 2 outlines the development, parameters, and outcomes of Random Forest models designed to predict specific behavioural patterns based on annotated data. The models aimed to classify behaviours such as Typical Toe-to-Heel (TTB) walking, heel / toe line walking, spinning, BW / FW walking / running, typical line walking, and a comparison between TTB, spinning, and typical behaviours.

[0148] Parameters

[0149] For the model fitting, using sci-kit-learn in Python, the following parameters are taken into account:

[0150] Random Forest classifier

[0151] Number of estimators (= The number of trees in the forest) = 100

[0152] Random state = 46

[0153] Criterion (= The function to measure the quality of a split) = entropy or log loss

[0154] Maximum features (= The number of features to consider when looking for the best split) = Iog2 Class weights (=Weights associated with classes) = balanced, stratification as a function of the annotated behaviour

[0155] Training versus testing versus validation ratio = 70% / 15% / 15%

[0156] In addition, a time kernel of 15 seconds is considered, as well as a behavioural smoother of + / - 5 measurements. The latter is used as a noise-reduction tool.

[0157] Results

[0158] The tables presented below are a representation of the performance of various Random Forest classifiers trained to identify specific behavioural patterns using both typical development (TD) and autism spectrum disorder (ASD) pilot study data. The performance metrics given are precision, recall, fl-score, and support:

[0159] • Precision measures the accuracy of positive predictions (i.e., the proportion of identified positives that are correct).

[0160] • Recall (or sensitivity) indicates the ability of the model to find all relevant cases (i.e., the proportion of actual positives that are identified correctly).

[0161] • Fl-score is the harmonic mean of precision and recall, providing a single score that balances both the concerns of precision and recall in one number.

[0162] • Support is the number of actual occurrences of the class in the specified dataset.

[0163] For TD (Figure 2a) and ASD (Figure 2b), different models were assessed:

[0164] 1. Spinning versus Non-Spinning: The classifiers' ability to distinguish between spinning and nonspinning behaviour.

[0165] 2. TTD versus Non-TTD: The models' performance in differentiating tiptoe walking (TTB) from other walking patterns.

[0166] 3. Original: A multiclass classification where the models predict various behaviours, including typical line walking, spinning, TTB, typical walking, heel / toe line walking, and BW / FW walking / running.

[0167] 4. Typical versus TTB versus Spinning: A three-class classification task to differentiate between typical behaviour, TTB, and spinning.

[0168] Performance tables

[0169] The metrics showed high precision and recall rates in some categories, indicating good model performance. For example, in the ASD dataset (tables 5 to 8), the precision for non-spinning behaviours is very high (0.99), with perfect recall (1.00), leading to an fl-score of 0.99, which indicates excellent model performance. However, lower recall rates were seen in some behaviours, like spinning in the ASD dataset (0.48), which suggests that the model had difficulty in correctly identifying all true spinning behaviours. Typical development (TD) dataset:

[0170] Table 1: Spinning versus non-spinning

[0171] Table 2: TTD versus non-TTD

[0172] Table 3: Original Table 4: Typical versus TTB versus Spinning

[0173] Autism spectrum disorder (ASD) dataset:

[0174] Table 5: Spinning versus non-spinning

[0175] Table 6: TTD versus non-TTD Table 7: Heel / toe line / TTB / Typical / Spinning

[0176] Table 8: Typical versus TTB versus Spinning ROC curves training / validation / test for dichotomic combinations

[0177] The receiver operating characteristic (ROC) curves graphically display the performance of the classification models by plotting the true positive rate against the false positive rate at various threshold settings. These curves provide a powerful visual comparison of classifiers' performance. In this case, ROC curves were only presented for dichotomic combinations such as spinning versus non-spinning in TD and ASD patients (Figures 4, 5,8 and 9) and TTB versus non-TTB in TD and ASD patients (Figures 6, 7,10, and 11).

[0178] In an ROC curve, the true positive rate (sensitivity) is plotted on the y-axis, and the false positive rate (1 - specificity) is plotted on the x-axis. The closer the curve follows the left-hand border and then the top border of the ROC space, the more accurate the test. The Area Under the Curve (AUC) provides a single metric summarizing the performance across all thresholds; an AUC of 1 indicates perfect classification, whereas an AUC of 0.5 indicates no classification ability better than random chance.

[0179] For all sets of curves, the smoothness and steep ascent towards the upper-left corner are indicative of both high sensitivity and specificity, suggesting the models are both accurate and reliable in their predictions for these particular classifications. The tightness of the curve to the top-left corner of the plot also suggests a high overall accuracy across different threshold levels.

[0180] Furthermore, a very high AUC is calculated, indicating excellent model performance in differentiating both spinning versus non-spinning behaviour and TTB versus non-TTB behaviour. This statement holds for both typical and ASD children.

[0181] Behaviour contributions

[0182] Table 10: TD

[0183] Table 11: ASD

[0184] (1) Non-TTB; (2) Non-Spinning; (3) Typical; (4) TTB; (5) Typical walking; (6) Spinning; (7) Typical line; (8) BW / FW walking / running; (9) Heel / toe line.

[0185] Data training set

[0186] Figures 12 and 13b display sequences of behaviours over time, differentiated by colours. The behaviours are categorized into three types: spinning (black), TTB (light grey), and typical (dark grey). In the upper graphs of Figures 12 and 13, data acquired within the ASD series (Table 11) were used for the prediction of a specific person, using the ASD-model. In the lower graphs of Figures 12 and 13, the data from the same person were used for behavioural predictions using the TD-model (Table 10). Based on these Figures 8 to 11, it is stated that behavioural modelling indeed requires appropriate training of the data, using manual annotation from the same series of persons.

[0187] Deployment

[0188] The deployment of the developed Random Forest models implemented in Python involves several steps. First, a new model was trained using previously acquired and annotated data, as well as newly acquired data and manually annotated data. This model can be saved and then loaded for prediction in a production environment. Python's 'pickle' module was used for serializing and deserializing these models, demonstrating that it's suitable for saving a trained model.

Claims

CLAIMS1. A computer-implemented method for predicting a kinetic biomarker associated with a neurodevelopmental disorder the method comprising the steps of: receiving a first data flow, generated by a first inertial measurement unit externally mounted onto a lower leg portion of the subject, the first data flow being indicative of a movement and orientation of the lower leg of the subject; receiving a second data flow, generated by a second inertial measurement unit externally mounted onto an upper foot portion of the subject, the second data flow being indicative of a movement and orientation of the foot of the subject; determining using at least one machine learning model that the first data flow and the second data flow correspond to an atypical movement event, wherein the atypical movement event includes at least tiptoe behaviour and spinning behaviour; determining, based on a frequency of occurrence of the atypical movement event during a predetermined time window, that the first data flow and the second data flow correspond to at least one kinetic biomarker associated with the neurodevelopmental disorder; and, generating a report comprising the at least one kinetic biomarker.

2. The method according to any one of the preceding claims, wherein the first inertial measurement unit is mounted onto a lower portion of the shin of the subject's leg, preferably adjacent to an ankle of the subject, and / or wherein the second inertial measurement unit is mounted on the instep of the subject's foot, preferably around a middle region of the instep of the subject's foot.

3. The method according to any one of the preceding claims, wherein the first and / or second inertial measurement units comprise: an accelerometer configured to generate linear acceleration data, a gyroscope configured to generate angular velocity data, and a magnetometer configured to generate magnetic field strength data used to determine an orientation, in three directions.

4. The method according to any one of the preceding claims, wherein the first and / or second data flows are indicative of one or more of: a rate, a rate of change, a frequency, and / or an amplitude of rotations of the foot of the subject.

5. The method according to any one of the preceding claims, wherein the machine learning model comprises a classifier model configured to classify the first data flow and the second data flow into categories representing 'tiptoe behaviour' and 'non tiptoe behaviour', and 'spinning behaviour' and'non spinning behaviour'; preferably wherein the classifier model is selected from the group consisting of decision trees, random forests, gradient boosting and / or neural networks; more preferably random forests.

6. The method according to any one of the preceding claims, further comprising the step of receiving a third data flow generated by a pressure measurement unit externally mounted on a lower foot portion of the subject, preferably on a sole of the subject's foot, the third data flow being indicative of a pressure exerted by the foot of the subject during movement.

7. The method according to any one of the preceding claims, wherein the neurodevelopmental disorder includes one or more of: autism spectrum disorder, attention-deficit / hyperactivity disorder, and / or developmental coordination disorder.

8. The method according to any one of the preceding claims, wherein the report comprises an overview of the frequency of occurrence of the atypical movement event during one or more predetermined time windows and / or a timestamped timeline of the detected atypical movement events within one or more predetermined time windows.

9. The method according to any one of the preceding claims, wherein the method further comprises wirelessly transmitting the first data flow and the second data flow via a data link communicatively coupled to the first and second inertial measurement units to a processing unit configured for performing the steps of the method; preferably, wherein the processing unit comprises a portable electronic device, such as a smartphone, tablet, or wearable computing device.

10. A computer program comprising instructions which, when executed by a computer, cause the computer to perform the method according to any one of claims 1-9.

11. A kinetic biomarker prediction system, comprising a first inertial measurement unit adapted to be externally mounted onto a lower leg portion of a subject, and configured to generate data indicative of a movement and orientation of the lower leg of the subject; a second inertial measurement unit adapted to be externally mounted onto an upper foot portion of the subject, and configured to generate data indicative of a movement and orientation of the foot of the subject;at least one processing unit comprising a memory configured to store computer-executable instructions, and at least one processor configured to access the memory and execute the computerexecutable instructions to: receive a first data flow generated by the first inertial measurement unit, the first data flow being indicative of a movement and orientation of the lower leg of the subject; receive a second data flow generated by the second inertial measurement unit, the second data flow being indicative of a movement and orientation of the foot of the subject; determine using at least one machine learning model, that the first data flow and the second data flow correspond to an atypical movement event, wherein the atypical movement event includes at least tiptoe behaviour and spinning behaviour; determine, based on a frequency of occurrence of the atypical movement event during a period of time, that the first data flow and second data flow correspond to at least one kinetic biomarker associated with a neurodevelopmental disorder; and, generate a report comprising the at least one determined kinetic biomarker.

12. The system according to claim 11, further comprising a garment wearable on the foot and at least a lower leg portion of a subject; wherein the garment includes at least a sock comprising a lower leg portion and a foot portion; wherein the first inertial measurement units is embedded within a fabric at a frontal part of the sock's lower leg portion; and wherein the second inertial measurement units is embedded within a fabric at an upper part of the sock's foot portion.

13. The system according to claim 12, further comprising at least one first processing unit integrated into the wearable garment and configured to perform at least a portion of the computer-executable instructions stored in the memory of the processing unit, and at least one second processing unit external to the wearable garment and configured to perform at least the remaining portion of the computer-executable instructions.

14. The system according to any one of claims 11-13, further comprising a data link configured to wirelessly transmit the first data flow and the second data flow from, respectively, the first inertial measurement unit and the second inertial measurement unit to the processing unit; preferably, wherein the data link comprises a wireless transmitter in wireless communication with the processing unit.

15. A method of training a machine learning model for use in the method according to any one of claims1-9, the method comprising the steps of: externally mounting a first inertial measurement unit on a lower leg portion of a reference subject, the first inertial measurement unit being configured to generate a first data flow indicative of a movement and orientation of the lower leg of the reference subject; externally mounting a second inertial measurement unit on an upper foot portion of the reference subject, the second inertial measurement unit being configured to generate a second data flow indicative of a movement and orientation of the foot of the reference subject; acquiring, during a movement of the reference subject, the first data flow and the second data flow, thereby obtaining reference data; processing the reference data to obtain a set of feature vectors, wherein the feature vectors comprise one or more statistical or kinematic parameters derived from the first data flow and / or the second data flow; annotating the feature vectors with labels indicating one or more atypical movement events, wherein the atypical movement events include at least tiptoe behaviour and spinning behaviour, thereby obtaining a ground truth dataset; inputting the feature vectors into an untrained machine learning model; training the machine learning model by iteratively adjusting internal parameters of the machine learning model based on a comparison of the machine learning model output to the ground truth dataset, thereby obtaining a trained machine learning model; and, optionally, applying the trained machine learning model to determine whether a first data flow and a second data flow of a subject correspond to an atypical movement event.

Images