DESC:ARTIFICIAL INTELLIGENCE AND GAMEPLAY BASED METHOD FOR ASSESSING AUTISM SPECTRUM DISORDER AND DEVELOPMENTAL DELAYS

TECHNICAL FIELD
The present disclosure relates to a method and system for use as an aid or an assessment tool in early evaluation of a subject for neurodevelopmental disorders in particular autism spectrum disorder and developmental delays.
BACKGROUND
The formative years of a child spanning the first 6 years, also known as the “Golden Period” is a crucial time period for their growth and development. To assess a child’s overall growth and development, there are a set of age-specific tasks that act as fundamental marking points further known as Developmental Milestones. Although each milestone has a suggested age limit, the actual milestone age may vary for each child. However, a significant delay or an absence of it may signal a potential risk with a child’s development. It may further indicate the presence of a developmental disorder, such as Autism Spectrum Disorder (ASD). It is a lifelong neurological condition that causes impairment in three major developmental areas of the brain associated with social interaction, communication, and imagination.
Being a spectrum disorder, no two individuals have similar characteristics that makes it a challenge for someone with little or no expertise to distinguish an autistic child amongst neuro-typical children. Although the symptoms start to surface from 18-36 months of age, the barely noticeable signs are often missed or mistaken as delayed milestones. This delay in identification causes the child to miss out on the “golden period”, i.e below 6 years of age which is crucial in maximizing the positive impact of early intervention therapies as the brain is more adaptable at a younger age. There is scope for adapting and integrating digital technologies in assessing and help in identifying the barely noticeable signs that might be missed by an untrained eye.
ASD’s global prevalence is to be estimated at 1 in 160 children. However, reaching an accurate diagnosis, accessing therapy, acquiring epidemiological data and determining true prevalence figures are still major challenges, especially in developing nations. (Onaolapo A Y, Onaolapo O J, 2017) In India, about 1 in 100 children under age 10 has autism, and nearly 1 in 8 has at least one neurodevelopmental condition. (Br JPsychol, 2011).
Autism Spectrum Disorder impacts an individual’s skill to perceive, comprehend situations and participate in social interaction. These challenges are faced by different individuals differently depending on the level of the disability and the combination of symptoms. People with ASD have difficulty with communication and interaction with other people, restricted interests and repetitive behaviours. Symptoms that affect the person’s ability to function properly in school, work, and other areas of life (Nimh.nih.gov, 2019)
Autism is known as a spectrum disorder because there is a wide variation in the types of symptoms people experience. Being on Spectrum refers to a broad range of conditions that measure the severity of the symptoms. Each individual with autism has a distinct set of strengths and challenges. Two individuals with the same diagnosis may look very different when it comes to their behaviours and abilities. The way they learn, think and problem-solve can range from highly skilled to severely challenged. Sometimes, some individuals may require significant support in their daily lives, while some individuals with autism may require less support and can even carry out their daily activities independently. (Mayo Clinic, 2018)
Autism is a vast spectrum and has a lot of similarities with other conditions which makes it a challenge to diagnose an individual with it. There are questionnaire based screening tools available in the market for parents, non-experts to assess the growth and development journey of a child. However, formal diagnosis involves assessment by experienced medical professionals i.e. developmental paediatrician, clinical child psychologist and child psychiatrist, paediatric neurologist (Lisa Jo Rudy,2019). Diagnosis of Autism does not require any medical or genetic test. It is an elaborate and time-consuming assessment consisting of observations on the child’s behaviour, interviews with parents and family members on the child’s developmental stages, and also running through the child’s medical history.
Overall, these diagnosis tests evaluate the child on the basis of developmental milestones, behavioural skills, language acquisition, imitative skills, cognitive skills (Lisa Jo Rudy, 2019). To conduct diagnosis, several standardized assessment tools are available that require significant clinical expertise such as ADI-R (Autism Diagnostic Interview Revised), ADOS (Autism Diagnostic Observation Schedule) and DSM V (Diagnostic and Statistical Manual of Mental Disorder, Fifth Edition).
Socio-cultural factors play a major role in early identification, diagnosis and management of Autism Spectrum Disorder. There is a need to enhance awareness of ASD and its impact on families in order to facilitate early detection and intervention. The first five years of a child’s life is a golden period of their development, contributing to their future learning skills and social and emotional abilities due to rapid gains in physical and development. (Rukmanee Butchon, Tippawan Liabsuetrakul, 2017). Studies suggest that from birth to age 5, a child’s brain develops more than at any other time in life and early brain development has a lasting impact on a child’s ability to learn and succeed in school and life.
Studies show that early diagnosis and introduction of intervention for autism in the formative years can have a major long-term positive impact on the child’s growth and development and can significantly impact cognition, language and adaptive behavior. Early diagnosis is vital to improving long-term outcomes related to cognition, language, adaptive behavior, daily living skills, and social behavior in affected adults.
Correlation between Motor and Cognitive Domains Studies have established a correlation between motor skills and cognitive skills in children. A study showed a strong correlation between visual processing and fine motor skills. (Davis EE, Pitchford NJ, Limback E, 2011) The fine finger movements and balance is associated with the Cerebellum. Since the cerebellum is primarily involved in motor control, neuroimaging has revealed that this structure also plays an important role in nonmotor functions such as language, learning, and memory, which suggests a common neural basis for motor and cognitive skills. (Caroline Leopold, 2018)
Progress in the field of screening or identification using technology utilising sensory-repetitive clinical symptoms and non-social cognitive processing as primary symptoms into consideration, there can be a scope of measuring the spectrum characteristics. According to a recent study, the presence of cues involving motor disturbances in children with autism, the facial expression of interest and affect and attention to other persons' expressions can help in early identification of autism rather than using symptoms involving cognitive or linguistic aspects. (Trevarthen C, 2013) Another research indicated that measuring finger kinematics using tablet-based games can contribute to creating an Autism motor signature at a very early age of growth and development. (Anna A, Krzysztof S, Jonathan T, 2016)
On the other side, progress in the field of designing valid and reliable assessments that measures and captures various parameters to support learning has been developed. The technique called Stealth Assessment based on the evidence-centered design (ECD) enables the development of assessments that are woven directly and invisibly into the fabric of the gaming environment. Evidence centered design uses components such as competency model, evidence model and task model to map tasks/activities to competency to capture using evidence collected during the process of the task. (Valerie S, Matthew V, 2013) Using the Stealth Assessment model and Digital Games that captures fine motor skills using touch interaction of finger movements can contribute to designing innovative tools that detect early signs of autism in an efficient way and by capturing the true nature of the child’s nature.
The present invention pertains to a game-based assessment platform that helps in identifying children who may have a potential risk to autism or any other related developmental disorders within the Golden Period. With the assistance of games on touchscreen devices such as smart tablets and the underlying technology, the platform is able to produce an unbiased analysis based on various patterns identified from the data to distinguish a child with a potential risk to autism amongst neuro-typically developing children. (Figure 1)

OBJECTS OF THE INVENTION
Presently, autism and related neurodevelopmental disorders are identified using questionnaire based screening or observation of the child by a child psychologist or developmental pediatrician. It becomes a challenge for someone with little or no expertise to assess and identify an autistic child amongst typically developing children at early stages. Once the symptoms start becoming obvious, the disability almost sets in, thereby resulting in missing out on the golden period, in particular the first 2-6 yrs. Thus, there is a delay in identification and professional help is sought belatedly after the age of 6-7 years.
This delay in identification causes the child to miss out on the “golden period” which is crucial in maximizing the positive impact of early intervention and therapies as the brain is more adaptable at a younger age. There is scope of integrating digital technologies to screen and help in identifying the non-noticeable signs that might be missed by an untrained eye.
The invention aims to assess children with likelihood of autism within the golden period. It works as an aided tool for an untrained eye to flag children that usually get missed. It is an interactive game-based application that captures the non-obvious traits and strong cues which suggest potential likelihood of autism preferably within 2-6 years of age. Based on the individual child’s gameplay, the tool generates an objective analysis that helps in understanding the child’s developmental progress.
The game-based screening platform that helps in identifying children who may have a potential risk to autism or any other related conditions in the Golden Period. With the assistance of games on smart tablets and the underlying technology, it is able to produce an unbiased analysis based on various patterns gained from the data to distinguish a child with a potential risk to autism amongst neuro-typically developing children.

SUMMARY
The formative years of a child (2-6 years), also known as the “Golden Period” is a crucial time for their growth and development. To assess a child’s overall growth and development, there are a set of age-specific tasks that act as fundamental marking points called Developmental Milestones. Although each milestone has a suggested age limit, the actual age of marking it may vary for each child. However, a significant delay or an absence of it may signal a potential risk with a child’s developmental area. It may also signify characteristics of a hidden developmental disorder, such as Autism Spectrum Disorder. It is a neurological condition that causes impairment in 3 major developmental areas of the brain; social interaction, communication, and imagination. Being a spectrum, no two individuals have similar characteristics that make it a challenge for someone with little or no expertise to distinguish an autistic child amongst neuro-typical children. Although the symptoms start to surface from 18-36 months of age, the non-noticeable signs are missed or mistaken as delayed milestones. Autism spectrum disorder has a wide range of variation in the type and severity of symptoms. Thus, in a child with classic autism, the symptoms are easily noticed and can be identified even by an untrained eye. But for a child with borderline autism, there lies a challenge because it is usually missed or mistaken as delayed developmental milestones or deemed as slow learners. The present invention relates to an assessment tool for identifying children with likelihood of autism within the golden period (below 6 years).
It works as an aid for an untrained eye to flag children that usually get missed. It is an interactive game based application that captures the non-obvious traits and strong cues which suggest potential likelihood of autism within the golden period especially 2-6 years of age. Based on the individual child’s gameplay, the tool generates an objective analysis that helps in understanding the child’s developmental progress.
The present invention relates to a method for use as an aid or an assessment tool in early evaluation of a subject for neurodevelopmental disorders in particular autism spectrum disorder and developmental delays including attention deficit hyperactivity disorder and intellectual disability.
The present invention also relates to an apparatus or device for computation in particular a tablet comprising the gameplay application that extracts features from game data that is subsequently analysed by a machine-learning model or algorithm installed on a computing device for use in early evaluation of a subject for neurodevelopmental disorders in particular autism spectrum disorder and developmental delays.
The invention further relates to a system comprising the gameplay apparatus or device for computation, the gameplay application, the machine-learning model or algorithm and computing device that is used to aid early assessment of a subject for neurodevelopmental disorders in particular autism spectrum disorder and developmental delays. Additionally, the said gameplay application and machine-learning algorithm may be installed within the device or accessed remotely from a server.
The present invention also relates to a method that provides an assessment tool for early evaluation of a subject for neurodevelopmental disorders in particular autism spectrum disorder and developmental delays.
The uniqueness of the method lies in the:
1. Set of features and their derivation or extraction method.
2. The algorithm which understands and compares the features to differentiate patterns.
3. The method/process used to collect, record data and infer final results.

BRIEF DESCRIPTION OF THE DRAWINGS
The accompanying drawings are included to provide a further understanding of the present disclosure, and are incorporated in and constitute a part of this specification. The drawings illustrate exemplary embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure.
FIGURE 1 Illustrates a flow diagram of the method for screening using machine learning model.
FIGURE 2 Stakeholders map.
FIGURE 3 Video of geometric shapes and human faces.
FIGURE 4 Self Recognition Game
FIGURE 5 Coloring Game: Inbuilt instruction prompts the child to color the object. The child can change color by tapping on the crayons. After completion of the game, the “a smiley face pops up” in order to encourage the child to play the game
FIGURE 6 Tracing Game: The child is prompted to trace the car on the specified path. In the end, the reward is shown to the child on completion
FIGURE 7 Selection Game
FIGURE 8 Data analysis process diagram
Figure 9 Process overview
DETAILED DESCRIPTION
The gameplay-based invention is aimed at addressing the challenge of early identification by aiming at early screening of children who may have a potential risk to autism or any other related conditions. This study was conducted under the guidance of medical experts such as clinical child psychologists, rehabilitation psychologists, developmental paediatricians, occupational therapists, and speech therapists. It is a 20-month study conducted to understand the experiences and challenges faced by children with autism, their caregivers and immediate stakeholders. Approaching it as System-level challenge, the goals established were: 1. To study the experiences of children with autistic conditions from the perspective of primary and secondary stakeholders and to identify an impactful opportunity area to intervene. 2. To design a need based digital solution that will help in enhancing the lives of children with autism. 3. To assess and validate the solution. This study includes intensive research on the overall ecosystem, beneficiaries, users, and various on field methodologies to achieve the aim of ScreenPlay.
The present invention relates to an assessment tool for identifying children with likelihood of autism within the golden period (below 6 years). It works as an aid for an untrained eye to flag children that usually get missed. It is an interactive game based application that captures the non-obvious traits and strong cues which suggest potential likelihood of autism within the golden period, especially 2-6 years of age. Based on the individual child’s gameplay, the tool generates an objective analysis that helps in understanding the child’s developmental progress.
The assessment tool is a touchscreen device preferably tablet-based application or system that captures the gameplay of each child and records it under their unique identification that helps in linking the gameplay to each individual child. It is a game based interactive solution that is developed on certain features that are prevalent in a number of autistic children and are non-obvious to untrained eyes to flag within the golden period.
The two main games comprising the application are:

1. Coloring
2. Tracing

The child interacts with the games usually for a period of 7-10 min game-time. These two games help in flagging children who may have autism. The application infers on the basis of certain features and parameters. The features suggest characteristics of autism and the parameters define the difference amongst the typically developing children and children with autism.
The features derived from the games are comprised as follows:
1. Drag Kinematics
2. Prospective Drag
3. Drag
4. Tap
5. Press
6. Response
7. Time
8. Meta

These games are self-intuitive and do not need any further external prompt. In the case of the tracing game, there is an internal prompt suggesting to move the car in the front direction whereas in the coloring game, there is an internal prompt for color selection and coloring activity. Data analysis is conducted by a machine-learning algorithm that helps understand the gameplays and extract differentiating factors which helps us in flagging children at the right age.
Research Methodology
Immersion: The set of research techniques we used to gain an understanding of the strengths, challenges, experiences shared by an autistic child and their care-givers, they are as follows: Ecosystem mapping: To understand the ecosystem for a child with ASD and to recognize the various stakeholders involved, the values that are exchanged, and the direct/indirect effect it has on the beneficiaries, users by creating a Stakeholders Map. (Figure.2)
A child with ASD is the centre of the ecosystem – is identified as the beneficiary. The primary environment would be the parents and family who have a direct impact on the child’s life. Their decisions will help in addressing the challenges faced by the child. The secondary environment is the doctors, therapists, special schools, special educators, mainstream schools, teachers, and friends. Their involvement has a direct impact on the child by providing the right guidance, diagnosis, treatments and an indirect impact on the lives of parents and family members as they are providing them with support. The tertiary environment is the government bodies (policies), NGO’s (creating awareness, working towards their livelihood, treatments, etc.) Outcome: Identifying the involvement of various stakeholders built the team to conduct observational studies, interviews and experiments with them.
Observational method for interaction with Children with ASD Age group: Classroom of Milestones Trust, Nashik a special school and therapy centre for children with special needs. Visit to Tamana Foundation, New Delhi, one of the first rehabilitate centres for autism in India. The observations are as follows: Majority of the children were from the severe category. With the help of the special educators, they were engaged in activities such as writing alphabets, numbers, putting beads into threads, sorting, playing with shapes. Few children were also involved in their repetitive behaviors such as stimming or being extremely attracted to textures. With every interaction with a child with ASD, helped in gaining insights about their routines, interests, and each child had their own unique characteristics. The way they perceive life, may seem atypical to us. They may seem aloof, non-responsive, socially uninterested, they hardly maintain eye contact and have their way own way of playing with objects. They indulge in arranging objects in line. Some children can hold a minimal conversation with other individuals, and some may not be able to even speak. They have an uncanny flair for expressing through and engage in activities such as making paper envelopes, pottery, block printing. There all kind of characteristics for autism and it varies from person to person.
Interviews with stakeholders: Conducted interviews with special educators, developmental Paediatricians, occupational therapists, speech therapists, clinical psychologists. Created an ideal journey map of how a child is evaluated by paediatricians, teachers, parents further involving formal diagnosis undergoing treatments and therapies. Parents shared their experiences about how difficult it was to identify the cues and accept that their child has ASD. Parents undergo a lot of emotional stress when it is conveyed to them that their child is showing characteristics of ASD. Conducted interviews with Pre-schools where teachers, principals, counsellors, shared their experiences of when there are children with ASD in a classroom, for a teacher who is untrained eye and children being at younger age, it is usually neglected as delayed developmental milestones, slow learners or behavioral issues. The school finds it difficult to convey it to the parents about the wellbeing of the child. Parents enrolling their children in mainstream schools, hear complaints about the child’s behavior, not being able to cope up with studies. The school overlooks the condition of the child, looks at as delayed developmental milestones causing a delay in identifying the early symptoms of the relevant condition/challenge faced by the child. Early intervention comes through i) identifying the challenge and acknowledging it ii) acceptance. Based on our secondary and primary research, there emerges a need to identify children who are facing developmental/neurological difficulties in the golden period. The design intent is to develop a screening tool for an untrained eye that will help in identifying children who are at a potential risk to autism or any other related conditions.
Designing of Experiments and Pilots Use of Digital Games: Using games as the medium to screen children, we conducted experiments to develop a strong, reliable and accurate screening tool. The games are to be age appropriate, intuitive and engaging that capture the true nature of the child’s mannerisms through these game-play interactions. This is used to analyze if any child has a potential risk to autism or any other related conditions.
Experiment 1- Designing of activities based on ASD attributes
Intent: The activities have been framed on the basis of some strong non-obvious attributes. This will help in measuring differences in the reactions of children with autism and typically developing children. Further compare them and identify which of the activities are strong differentiators amongst the two sets. In order to do these activities, we selected certain existing videos (in relation to the intent of the activity), existing playstore android applications (used on mobile phones) and experimented with both the sample sets. We also manually marked observations to assess the outcome of the games.
Sample Set: The team conducted these experiments with 9 mild children with autism in the age group of 3-6 years. These children are also attending mainstream schools and have been undergoing therapies for quite some time. The team also conducted these experiments with 12 typically developing children. To assess the activities, the team formatted a standard set of guidelines.
Following are the guidelines:
1. Level 1: Give no instructions – Parameters: Number of games played and till which complexity level
2. Level 2: Verbal Instruction – type of instruction given, how many times, for how many games, child’s response during and after instructions
3. Level 3: Gesture Instruction – type of gesture has been shown, how many times, child’s response during and after instructions.
4. Level 4: Hand-holding instruction – duration of the instruction, how many times given, child’s response during and after instructions.
Basic Parameters to be analyzed: concentrating on one part of the screen, language specific, random tap on screen, tracing slowly or quickly, is the child selecting the right answer or just going by the position (learning or memorizing), responding to awards.
1. Attribute: Social Smile Activity: Video of baby laughing Intent: To observe if the child is able to relate and give a response to the baby’s expression and sound of laughing. Observation: When manually tested with both the sets, the team received a mixed response. Neuro-typically developing children responded but seemed a bit confused. On the other hand, some of the children with ASD were smiling looking at the screen but continued smiling when looked away. Some of the children with ASD did not look at the screen. After doing a background study on the number of children who were undergoing therapy could react to it just as neuro-typically developing child.
2. Attribute: Gaze Patterns, based on research that children with ASD tend to engage with more geometric shapes than children with facial expressions. (Pierce K, Conant D, Hazin R, Stoner R, Desmond J, 2011) Activity: video containing moving geometric patterns on one side and children dancing and doing yoga on the other. (Figure 3) Video of geometric shapes and human faces Intent - To judge if the child likes geometric patterns or children doing yoga (the children with ASD are comfortable with social cues.) Outcome: A lot of typically developing children were also fascinated by the patterns made by the geometric shapes, especially when bright colors were displayed such as yellow, red, pink. The gaze patterns kept shifting between the two screens. With children with ASD, their gaze patterns also suggested spending time with children making facial expressions. Thus, this did not stand as a strong differentiator.
3. Attribute: Self-face Recognition Games to check how quickly the typically developing children and children with autism can spot their faces amongst random ones. (Figure 4) Outcome: The children needed instructions as to know what is to be done. So an external verbal instruction in a familiar language was given only once to both the sets. Based on which neuro-typically developing children could identify themselves until level 3. However, children with ASD could till level 1 and were more tapping randomly on the screen. As this is a language dependent prompt, it was dropped by the team.
4. Attribute: Finger Kinematics
Activity 1- Coloring: the child is able to relate to an object (everyday object) and color it. Here, to manually analyze the gesture patterns of the child. (Figure 5) Coloring Application Outcome: Neuro-typical children could relate to the foreground and the background object, they colored by maintaining the internal boundaries also. Children with ASD were happier to interact with the screen then coloring it. They did not differentiate amongst the object and were randomly scribbling. Neuro-typical children seemed to be confused with a lot of color options available. A child with ASD ignored most of it, and when showed to use the colors, started engaging with the color palate only. This activity stood out as a strong differentiator amongst the two sets.
Activity 2: Tracing: To measure if the child is able to trace the pattern when interacting with a screen in the needed direction of the object, also to judge the gesture patterns of the child. (Figure 6) Outcome: Neuro-typical children followed the instruction of swiping and following the direction of the shape (left to right). Children with autism were moreover tapping and were not able to follow the direction to trace. This activity stood out as a strong differentiator amongst the two sets.
5. Attribute: Recognition Activity: Select same objects based on options - To measure if the child is able to select the correct option by comprehension and not memorizing. (Figure 7) Intent: Outcome: Both neuro-typically developing children as well as children with ASD were randomly tapping on the objects. Neuro-typically developing children were still able to memorize the position of change. However, as the complexity increased they were not able to tap on the correct object. Overall outcome: Out of these experiments, coloring and tracing stood out as strong differentiators amongst the two sets. Also, the need to sit with one child individually and away from the classroom became a must so that we could be able to capture the true nature of the child’s mannerisms through these interactions.

Design Optimization: The intent of the games was to keep it very simple, intuitive, age appropriate, colorful and bright that attract the attention of the child. Further, to make it easy to understand, introduced animations in the beginning of the games as internal prompts:
Coloring Game 1:
Coloring Game Trial 1 (all age groups) Internal Prompt: Hand buzzing over color palate and moving to the object area. In the beginning of the game and if there is no interaction with the screen then again will prompt after 30 seconds of the game. Outcome: After interacting with a few children, I realized that a ball as curve object was causing difficulty for children to color as well as to capture the game-play interaction. This was further progressed to a two option object based on the age. Also, the faint pencils and erasers were acting as distracting mediums and we're confusing the children, hence were removed.
Parameters of the game: The game has a time-limit, 3 minutes that is to be noted by the volunteer during the game-play. The child need not know that the game is clocked. After completion of 3 minutes or 90% of coverage of the object, the replay button and next button are made visible to be used. They are hidden beforehand, to avoid distraction or skipping the game.
Tracing Game 2:
Object to trace Internal Prompt: Car with a bubbling effect (intending to convey to be tapped) and a hand gesture prompting to pull the car ahead. Parameters of the game: If there is no interaction with the screen, the car automatically starts bubbling again after a duration of 30 seconds with the prompt. Once the child is done by driving the car all the way through, the replay and next buttons are made available.

5. Data Collection, Model Training and Result Inference
5.1 Data Collection and Labeling
To collect data and conduct analysis, these games have been designed to be a part of a screening tool that can identify children who may have a potential risk to autism or any other related conditions. (Figure 8)

The team built a controlled environment (away from the classroom to avoid distractions). The children played both the games thrice in a span of 15 minutes. The flagged children (the ones who were not at par with neuro-typically developing children) were asked to play again (after a gap of 4-6 days) in a controlled environment, thrice and in 15 minutes. We did this to eliminate false positives, and external factors such as the mood of the child. The children who were flagged twice were considered at a potential risk.

Finger kinematics data were collected through the games among typically developing children as well as identified children across 6 mainstream schools and 2 therapy centers. The tablet games captured the finger kinematics of the students. This data was collected for a total of 999 subjects. The data collected among mainstream schools was not completely labeled. To label the dataset, the features of those children whose gameplay had few outliers or had few peculiarities in school’s observation about the child was picked and taken through quick assessment by a child developmental pediatrician in further labeling our dataset. This active learning approach helped in making sure that the data was correctly labeled. (Table 1)

Table.1 Statistics of labelled dataset

The children play both the games (coloring and tracing) thrice in a span of 15 minutes. After the game play interaction, the game administrator answers a few basic questions about their child’s behaviour based on their observation. The raw data collected through the games consists of time series data of touch screen coordinates and has structure as shown in Table 2:

Table 2: Raw data structure

5.2 Features Selection:
A total of 73 features were derived from the finger kinematics data of both the games that were further narrowed down to a total of 35 features (17 in game 1 and 18 in the game 2) with the help of feature selection techniques like feature permutation and features importance. Also, the features with low variance were not further selected in building a classification model.

The total feature set can be broadly classified into 9 buckets:
1. Drag Kinematics
2. Prospective Drag
3. Drag
4. Tap
5. Press
6. Response
7. Time
8. Others
9. Questionnaire Responses

Further, those features were validated using 2 sided Student’s t-test and a significant difference between the 2 sets was observed. Table 3 shows the results of the t-test between the identified children and typically developing children across the features bucket.

Table. 3 T-test result of features between ASD Children and Typically Developing Children.
The following features set were derived from the raw data as shown in Table 4:
Bucket Name Code Name Descriptive1 Per Zone2
Drag Kinematics Velocity drag_velocity Yes Yes
Acceleration drag_acceleration Yes
Jerk drag_jerk Yes
Longest Drag Kinematics longest_drag_velocity
Deacceleration drag_deacceleration Yes
Prospective Drag Movement Units movement_units Yes
Peak Velocity peak_velocity Yes
Time to Deacceleration deacceleration_time Yes
Drag Distance drag_distance Yes Yes
Area drag_area Yes
Duration drag_duration Yes Yes
Count drag_count Yes
Drag height drag_height Yes
Drag width drag_width Yes
Tap Tap Gesture Count tap_count Yes
Unique TapCount tap_count_unique Yes
Consecutive Taps tap_count_continious Yes
Press Duration press_duration Yes
Press Count press_count
Response Touch Delay touch_delay Yes
Initial Touch Delay touch_delay_starting
Time Total Time total_time yes
Others Number of colors colors_count
Boundary Maintenance Index bmi
Questionnaire

Table 4: Features set derived from the raw data. Descriptive: For every descriptive feature, mean, median, min, max is calculated. 2 Per Zone: Features are calculated for each predefined zones of the game

With the help of feature selection techniques like feature permutation and features importance, the following features are shortlisted as shown in Table 5.

Sr. No. Top Performing features
1 acc_median_t
2 bmi
3 deacc_median
4 drag_area
5 jerk_max_t
6 jerk_mean
7 jerk_mean_t
8 longest_drag_duration
9 outside_distance
10 press_count
11 tap_count_t
12 total_time
13 touch_delay
14 Touch_delay_t

Table 5: Selected features set

5.3 Classification Model and Training
In order to build a classification model, the data of the ASD set was upsampled using the SMOTE algorithm. After training various algorithms, a random forest with 1500 estimators gave optimal results. To train and validate the model, KFold validation technique was used and hyper params were hyper tuned using GridSearch. Other than the random forest algorithm, ADA boosting gave the second-best optimal results.
The performance of the classification was assessed using 5 performance metrics: accuracy, recall, precision, ROC/AUC curve, and Matthew Correlation Coefficient. Accuracy is measured as the percentage of correctly classified subjects within all subjects. The recall is the ability to correctly identify ASD subjects within an ASD set, whereas precision measures the accuracy of correct classification of ASD subjects. ROC/AUC reflects the diagnostic ability of a binary classifier system when its discrimination cutoff varies. The Mathew correlation coefficient measures how much more effective the binary classifier is than random classification.
5.4 Screening Results
Using the best-selected model i.e: Random forest models performed better than all the models employed during the screening process and ADA boosting displayed second optimal results.
Using this method the screening accuracy of 88%, recall of 87%, and a precision of 88% was achieved in identifying children with developmental disorders and developmental delays as shown in Table 6. Also, game 2 showed the best result with ROC/AUC (Area under the curve) of 0.88 and MCC (Matthew Correlation Coefficient) of 0.77.
These screening results also show that the pattern that emerged during the gameplay is significant enough to identify the likelihood of developmental disorders at an early age by just capturing specific fine motor skills through tablet-based games.
Table.6 Model Evaluation Results

5.5 Inference and Report Generation
The best performing machine learning model is saved, and the next time when a new gameplay data is recorded, the saved model is used to derive results and classify the gameplay data that includes gameplay and questionnaire responses for a child. (Figure 9) The model output is then saved in the database and a report describing the potential of developing neurodevelopmental disorders is generated.

6. Conclusion
Our research using digital technologies along with the assistance of games on smart tablets can provide an unbiased and accurate analysis to identify children with potential risk in the golden period. Our tool can screen and identify the children who are developing differently from other neuro-typically children and through a quick assessment conducted by an expert we are able to validate any potential risks. The child is suggested to seek formal diagnosis and receive the required interventions. Screenplay has screened over 919 children in mainstream Pre-schools (age group of 3- 6 years) and the outcome has been validated by a quick assessment conducted by a medical expert- Developmental Pediatrician. Out of 919 children, screenplay has flagged 82 children out of which, 40 underwent quick assessment (remaining 42 are yet to be evaluated) by a developmental pediatrician. And 10 have confirmed to have a potential risk to conditions, 4 to have a potential risk to Autism, 3 potential risk to Learning issues, 1 potential risk Social Pragmatic Disorder, 1 potential risk to Sensory issue and 1 potential risk to ADHD. And these 10 children were further suggested to undergo a formal diagnosis by a specialist. We believe that Screenplay can have a significant impact on the lives of the children with ASD by screening them in the golden period. There are existing checklists and available questionnaires but can be best filled with the right type of expertise. For an untrained eye, game-based that requires very little or no expertise screening can help in identifying mild categories of children with developmental conditions and help in minimizing the time between when children can receive a diagnosis of ASD and when they are diagnosed.

The uniqueness of the algorithm lies in the:
1. Set of features and their derivation method.
2. The algorithm which understands and compares the features to differentiate patterns.
3. The method/process used to collect data and infer final results.

The present invention relates to a method for use as an aid or an assessment tool in the early evaluation of a subject for neurodevelopmental disorders in particular autism spectrum disorder and developmental delays including attention deficit hyperactivity disorder and intellectual disability.
The present invention also relates to an apparatus or device in particular a tablet comprising the gameplay application that extracts features from data that is subsequently analysed by a machine-learning algorithm installed on a computing device for use in early evaluation of a subject for neurodevelopmental disorders in particular autism spectrum disorder and developmental delays.
The invention further relates to a system comprising the gameplay apparatus or device, the gameplay application, the machine-learning algorithm and the computing device that is used to aid early assessment of a subject for neurodevelopmental disorders in particular autism spectrum disorder and developmental delays. Additionally, the said gameplay application and machine-learning algorithm may be installed within the device or accessed remotely from a server.
The present invention also relates to a method that provides an assessment tool for early evaluation of a subject for neurodevelopmental disorders in particular autism spectrum disorder and developmental delays.
Advantages over the previous techniques:
1. Language Agnostic and can be deployed to even culturally diverse regions despite the language barrier.
2. Provides evidence driven results.
3. Employs non observational methods to infer child’s behavior and removes human bias involved in earlier methods.
4. Removes the need for expert involvement in the earlier stages of assessment.
5. Stealth Assessment using intuitive and fun game play.
6. Provides Quantitative analyses of the child’s behavior and captures true nature.
7. Using machine learning integrated into the mobile games we are able to reduce the time required for a child to be diagnosed with ASD by half. Using this process we can make this process faster.
8. Using data captured through our games we are able to identify and capture true behaviour of the child, which further opens opportunity for evidence driven assessment and diagnosis of developmental disorders using machine learning + Digital Games.
9. The method will reduce the time and cost of early screening of neurodevelopmental disorders and developmental delays, and enable channelization of early intervention at the right age, hence increasing effectiveness of intervention.

,CLAIMS:We Claim:

1. A method for screening neurodevelopmental disorders by machine learning model, the said method comprising steps of:
recording raw data through screenplay tool played by children;
extraction and selection of features from the raw data;
imputation of result by a pre-trained machine learning model;
generation of results describing the likelihood of neurodevelopmental disorders.
2. The method of claim 1, wherein the neurodevelopmental disorders include autism spectrum disorders, developmental delays, attention deficit hyperactivity disorder and intellectual disability.
3. The method of claim 1 wherein the machine learning model is selected from Random Forest or ADA.
4. The method of claim 1 wherein the screenplay tool comprises of coloring game, tracing game and questionnaire.
5. The method of claim 1 wherein the raw data comprises of finger kinematics data and questionnaire data.
6. The method of claim 5 wherein the finger kinematics data comprises touch screen coordinates and time interval data.
7. The method of claim 1 wherein features are classified into following buckets:
a. Drag Kinematics
b. Prospective Drag
c. Drag
d. Tap
e. Press
f. Response
g. Time
h. Others
i. Questionnaire Responses

8. The method of claims 1 and 7 where the features, in particular are classified as:
acc_median_t;
bmi;
deacc_median;
drag_area;
jerk_max_t;
jerk_mean;
jerk_mean_t;
longest_drag_duration;
outside_distance;
press_count;
tap_count_t;
total_time;
touch_delay;
Touch_delay_t.

9. A computation device comprising a touch screen hosting Screenplay tool for screening of developmental disorders.
10. An application based on Screenplay tool for screening of developmental disorders.
11. A system comprising a computation device hosting a Screenplay tool, a data storage server, hosting a features extraction module and machine learning model for screening of developmental disorders.

12. The system of claim 11 wherein the computing device records data through games consisting of time series data of touch screen coordinates, questionnaire based observations, and uploads it to data storage server