INTRODUCTION
Autism spectrum disorder (ASD) is receiving media attention more than ever. The term “disorder” expresses a sense of differentiation from normal, so just because autism is a neurodevelopmental disorder from birth does not mean that symptoms are immediately noticeable; they typically manifest before a child reaches primary school age (Anagnostopoulou et al., 2020). Early detection can significantly improve a child’s skills and subsequent therapy because an ASD diagnosis can occur at any age (Akter et al., 2021).
Since there are no common medical diagnostic tools for ASD, such as a blood test, diagnosing ASD can be challenging. Children as young as 18 months of age or older can begin to receive an ASD diagnosis, while a formal diagnosis may not come until much later (Thabtah, 2017). A clinical assessment of the patient’s developmental age based on a number of areas, such as behavioral excesses, communication, self-care, and social skills, is also necessary for the diagnosis of ASD (Shahamiri and Thabtah, 2020).
Assessments for ASD have historically been conducted in specialized settings, but the fast increase in prevalence rates over the past 50 years has exceeded the capacity of these facilities and increased the wait times for diagnostic exams (Bridgemohan et al., 2018). The most widely used screening tools, interviews, and clinical observations for diagnosing ASD are the Autism Diagnostic Observation Schedule (ADOS) and the Autism Diagnostic Interview-Revised (ADI-R), which allow clinicians to assess a variety of symptoms and activities (Shahamiri and Thabtah, 2020).
Strong financial investment in artificial intelligence (AI)-powered technology has led to an exponential rise in relevant research in the healthcare industry (Buch et al., 2018). AI in healthcare now has more options, thanks to the availability of enormous and growing volumes of digital healthcare data, as well as improvements in computing, storage, and machine learning (ML) techniques (Shu et al., 2019). With the development of AI, ASD can be predicted at a very early level, which reduces the cost and time associated with diagnosing autistic features through screening tests (Omar et al., 2019). The idea that current autism diagnoses are costly, arbitrary, and time-consuming led to the development of AI for ASD diagnosis (Jiang and Zao, 2017).
By efficiently handling repetitive tasks, storing and manipulating massive amounts of data, and offering support for diagnostic or treatment decisions that may reduce the likelihood of errors, AI-enabled tools have the potential to improve diagnostic accuracy and efficiency. This, in turn, may lead to treatment or intervention and improve scalability (Ahuja, 2019). Researchers intend to create efficient, objective measurements that can aid in quickly diagnosing this disease with little effort by replacing present methods (which focus on the interactions between professionals and patients and/or their friends and family) (Thapaliya et al., 2018).
This inspired scientists to look for fresh approaches to early diagnosis of autism. The treatment is made better as a result (Raj and Masood, 2020). With the development of AI and ML, diagnosis of autistic features can be improved by utilizing a variety of methods from a very young age. As a result, several researchers have found that AI and ML are crucial in the early detection of ASD. These advancements enable doctors to diagnose patients more quickly and accurately (Albahri et al., 2018).
Despite the non-experimental nature of the published results, which makes clinicians generally skeptical of the efficacy of technology-based interventions (Jaliawala and Khan, 2020), there is a need to assist teachers, parents, and therapists in providing the best care possible for students with disabilities in order to address their inclusive needs, particularly in different modalities such as evaluation and therapy (Drigas and Ioannidou, 2012). Therefore, there exists an urgent demand for specialized diagnostic instruments to improve diagnostic capability and streamline ASD testing. The current study seeks to investigate the need for applying AI in the diagnosis and evaluation of ASD as seen by experts in response to this need. It aims to help doctors and decision-makers understand the potential and advantages of AI in supporting the diagnosis of this category.
Our contribution to this work is summarized as follows: The current study is considered the first attempt to study the requirements for the application of AI in diagnosing ASD and the challenges facing its application, as perceived by specialists in the context of the Kingdom of Saudi Arabia (KSA). Data collection from a sample of Arab participants provides meaningful evidence of external validity for the determination of AI requirements in the diagnosis of ASD. Moreover, the results of this study provide additional insights into effective practices as perceived by specialists regarding the requirements for the application of AI in the diagnosis of ASD.
The following part of this paper is categorized as follows: the theoretical background and hypotheses; the aim and research questions; methods and procedures; the results; and finally, the discussion.
THEORETICAL BACKGROUND AND HYPOTHESES
Conventional methods of diagnosing ASD
The most common methods for diagnosing ASD are behavior-based clinical techniques, which make use of well-known instruments, such as the ADOS and the ADI-R (Close et al., 2012). The Checklist for Autism in Toddlers (CHAT) was created by Baron-Cohen et al. (1992). A modified CHAT (M-CHAT) has been developed that promotes reduced sensitivity to the quantitative checklist for young children and is administered by clinical professionals and reported by the child’s parents based on observations about the child’s behavior (Robins et al., 2001).
The M-CHAT was eventually condensed to 10 questions to save time and became known as the questionnaire for children’s autism traits at 10 years old (Q-CHAT-10) (Allison et al., 2012). The Chinese version of the Q-CHAT was created by Wong et al. (2004), who also expanded the screening to cover infants between the ages of 16 and 30 months for each item in the Q-CHAT-10 (Shahamiri and Thabtah, 2020).
The Autism Quotient (AQ) was developed to assess autistic symptoms in elderly people with average intellect (Baron-Cohen et al., 2001). AQ comprises 50 different questions that address social skills, attention shifting, imagination, communication, and attention to detail (Wheelwright et al., 2006). Lord et al. created the semi-structured interview known as the ADI-R in 1994 for use in conjunction with the ADOS (Lord et al., 1994, 2012). Through interviews with parents or other childcare providers, this method is used to identify ASD in youngsters.
The role and contribution of AI in the diagnosis of ASD
In the assessment, diagnosis, and analysis of ASD utilizing AI, several themes have emerged. First, magnetic resonance imaging (MRI) is a medical imaging technique for displaying pathological changes in living tissues that produces a cross-sectional image of the brain (Yin et al., 2021). Furthermore, functional magnetic resonance imaging (fMRI) is a neuroimaging procedure that seeks to assess the functional activity of the brain by detecting changes in blood flow caused by specific stimuli; it also strives to analyze the functional activity of the brain. However, MRI is expensive and not common in hospitals, particularly in nations with limited resources. In addition, the MRI scanner is a constrained, limited area. An MRI scanner is a small, enclosed space. Claustrophobia can cause significant discomfort and anxiety, making it difficult for patients to undergo the procedure, which further complicates the diagnosis, especially for severe cases. Additionally, if metal devices are implanted in the body, the magnetic field may have an impact on them (Alqaysi et al., 2022).
In this context, a narrative review conducted by Chaddad et al. (2021) summarized and discussed the radiomic models used in diagnosing autism spectrum disorder (ASD) with artificial intelligence using magnetic resonance imaging (MRI)/functional magnetic resonance imaging (fMRI). The review highlighted the progress made in ASD-based radiomics, providing a brief description of ASD and the current non-invasive technique used for evaluating between ASD and healthy control (HC) subjects. According to the study by Chen et al. (2020), finding neuroimaging biomarkers that can differentiate between ASD patients and healthy controls based on fMRI brain pictures is more cost-effective. Their study has also demonstrated how data-driven AI technology can be used in therapeutic settings for neuropsychiatric illnesses, which frequently include subtle, often-invisible-to-the-human-eye changes in brain anatomy.
A comprehensive review and meta-analysis of the brain MRI study space by Moon et al. (2019) led to the identification of a structural MRI subset of 43 studies that looked at ML for ASD diagnosis and demonstrated greater diagnostic precision. According to Sen et al. (2018), the support vector machine (SVM) model was 66.8% accurate in predicting photos of people with ASD. To investigate various brain regions and network-level connectivity that may be particular to people with ASD, infrared imaging techniques have been combined with numerous AI algorithms (Mahajan and Mostofsky, 2015). Vaiyapuri et al. (2023) conclude that the Ensemble Learning Driven Computer-Aided Diagnosis Model for Brain Tumor Classification (ELCAD-BTC) technique demonstrates effective performance in classifying brain tumors in MRI scans.
Second, an electroencephalogram (EEG) is a test that uses tiny metal discs (electrodes) connected to the scalp to detect electrical activity in the brain, allowing for an early diagnosis. The biggest drawback of EEG recording is its lackluster spatial resolution. Additionally, according to Heunis et al. (2018), it does not offer the maximum level of accuracy for diagnosing ASD. In the study by Peya et al. (2020), the data were transformed into a two-dimensional model before being evaluated using a convolutional neural network model. The EEG still has limitations in a variety of circumstances (such as signal noise), even if it can be used to diagnose ASD.
Furthermore, the study by Barttfeld et al. (2011) looked at how people with and without ASD differed in terms of functional connectivity and general small brain networks in the delta (D) frequency band. The authors discovered that, on average, children with ASD showed increased short-range connectivity in the lateral–frontal joints and reduced long-range connectivity in the frontal–occipital joints.
The third theme includes algorithms for ML. ML is a field of computer science and, more specifically, a subfield of AI. Its primary characteristic is the capacity to build on prior knowledge by refining fresh results using input from the past. We can examine a vast amount of data using ML approaches (Sharma and Sharma, 2018). As part of an effort to enhance frequently used screening and diagnostic methods for autism, ML was utilized to generate algorithms for ASD. A study by Megerian et al. (2022), for instance, looked at the accuracy of an AI-based program as a medical device designed to assist primary healthcare providers in diagnosing ASD. The sensitivity of the device was 98.4% (91.6-100%). For approximately a third of this primary care sample, the device enabled timely diagnostic evaluation with a high degree of accuracy.
A multidisciplinary systematic review by Alqaysi et al. (2022) examined 40 articles in the literature related to the criteria for diagnosing ASD using AI evaluation; the review also explored the motivations, recommendations, and challenges associated with research on diagnosing ASD in the use of AI techniques and ML algorithms that need synergistic attention. Thus, this systematic review contributed to advancing the recommended solutions for ASD diagnostic research.
Ghosh et al. (2021) also conducted a comprehensive review of 58 research articles on the effectiveness of AI and the internet of medical things (IoMT) in the screening and management of ASD. The study concluded that the IoMT and devices that support AI can help people with ASD to achieve self-sufficiency. ML and AI-enabled devices can help them learn by assessing their condition and enable them to be in a shared environment without the need for a caregiver. The results of a study by Choi et al. (2020) on the applicability of ML in the diagnosis of ASD revealed that the application of AI for rapid diagnosis or screening of people with ASD may be useful. Haque et al. (2023) conclude that the IoMT is a capable technology for enhancing care and management, particularly for contagious diseases such as coronavirus disease 2019 (COVID-19), and that the scope of IoMT-based hospitals has a promising future.
By combining the M-CHAT and the AQ-10, Thabtah and Peebles (2020) suggested a rule-based ML (RML) technique for ASD detection. Compared to other evaluations, the enhanced RML had an error rate of 5.6%. In the adult, adolescent, and child groups, RML attained over 94.5%, 87%, and 90% accuracy, respectively. The ML results from Stevens et al. (2019) demonstrated the ability to identify ASD traits.
Hyde et al.’s (2019) thorough examination of 45 publications utilized supervised ML for ASD, which includes text analysis and rating techniques. The most often used supervised ML algorithms to support ASD diagnosis and screening efforts, investigate the genetic basis of ASD, and discover potential biomarkers for individuals with ASD were SVMs, decision trees, naive Bayes algorithms, and random forests.
The fourth theme covers deep learning (DL). It is a component of cutting-edge ML models that use modern hardware, like the graphics processing unit. In recent times, therapeutic applications of DL have demonstrated outstanding evaluation outcomes (Choi, 2018). Additionally, to enhance performance and combat overfitting, some work has blended conventional approaches with DL (Thomas et al., 2020). Using six distinct ML/AI techniques, including DL, Bahado-Singh et al. (2019) investigated the epigenetic basis of autism and the identification of early biomarkers to predict the illness. The findings suggested that there is a significant genetic component to autism development, and epigenetic markers appeared to be very reliable for predicting ASD in babies. DL algorithms and neural networks were used by Heinsfeld et al. (2018) to detect individuals with ASD, with an accuracy range of 66-71% and an average evaluation accuracy of 70%. The random forest evaluation had an average accuracy of 63%, while the evaluation of support conveying machines had an average accuracy of 65%. In a study conducted by Poonguzhali et al. (2023), it was found that employing advanced imaging methods resulted in a significant improvement in the ability to accurately diagnose medical conditions. This indicates that the use of these advanced imaging techniques led to a marked increase in the precision of identifying and understanding diseases or health issues in patients.
We may conclude that the main objective of studies that used AI and its applications to diagnose autism was to increase the precision of the current assessment scales. The diagnosis aspects that can be used in each diagnostic approach have their limitations, and relying on a single diagnostic strategy may not provide an accurate diagnostic. By focusing on some aspects while ignoring others, these approaches may fail to arrive at the correct diagnosis. Moreover, there is currently no novel unified approach for an AI-based comprehensive diagnostic that connects these different routes. The diagnostic process is bolstered and supported when multiple pathways are integrated. However, research claims that in the context of AI, ML, and DL algorithms, there is a lack of integration between these routes for ascertaining medical features and sociodemographic traits.
THE AIM AND RESEARCH QUESTIONS
According to the World Health Organization (WHO), ASD research has recently attracted increasing attention. Each year, one child out of every 160 is diagnosed with autism, making the present global expansion rate of ASD significant and accelerating rapidly (WHO, 2023). According to Omar et al. (2019), ASD symptoms typically start to manifest in the first 2 years of life and progress with time. Due to the vast variations in the types and degrees of symptoms, including the process of early diagnosis and the study of the aspects associated with this condition, the diagnosis process is difficult and complex (Dutta et al., 2019).
Due to gaps in the cognitive ability of infants aged 24 months or beyond, it is reportedly difficult to detect ASD at a young age (Moon et al., 2019; Kaur et al., 2020). As a result, many hours of waiting are required by current diagnostic instruments for the evaluation of a tiny fraction of cases (Tariq et al., 2018). For enhancing predicted outcomes and lowering financial expenses, accurate and early ASD diagnosis and treatments are crucial (Chen et al., 2020).
There is a need for computational systems that can mimic the abilities of specialists and support current diagnostic techniques by validating professionals’ evaluation judgments (Thabtah and Peebles, 2020). In order to get the highest prediction outcomes for ASD, the researchers have tried to apply AI approaches and ML algorithms, which have significantly aided clinical and healthcare staff (Alqaysi et al., 2022).
Establishing more effective learning settings reduces costs and saves time while extending treatment times and improving the effectiveness of early diagnosis and intervention (Drigas and Ioannidou, 2012).
However, the requirements for the application of AI to be covered in the diagnosis of ASD to improve their preparation and rehabilitation are still vague and unknown. To fill this research gap, the current research aims to identify the most important requirements for the application of AI in diagnosing ASD, as perceived by specialists. It also intends to reveal the most prominent barriers to the application of AI in diagnosing ASD, as perceived by specialists. To achieve this, it will attempt to answer the following question:
METHOD AND PROCEDURES
Research methodology
Based on its suitability for the problem at hand, the authors chose the descriptive survey approach to describe the phenomenon under study in terms of its nature and the degree of its presence in order to determine the conditions under which AI could be used to diagnose ASD. This was done by using a questionnaire to elicit answers to the research questions and enable the use of statistical methods to analyze the data.
Participants
The specialists in the current research include people who hold an academic degree (diploma, bachelor’s, master’s, doctorate) in the field of special education in universities and private sector institutions, schools, programs, and government and private centers of the Ministry of Education in the KSA that register students with ASD. The study sample consisted of 595 specialists in the field of special education at the Ministry of Education, universities, and public and private sector institutions in the KSA. The sample was divided into private and domestic categories in the KSA to calculate the psychometric characteristics, with an average age of 39.71 years and a standard deviation of 4.62. A basic sample consisted of 423 specialists, with an average age of 39.54 years and a standard deviation of 5.24 years, in the field of special education at the Ministry of Education, universities, and public and private sector institutions in the KSA who completed the questionnaire. This is to apply the research tool in its final form and answer the research questions. Table 1 shows the demographic characteristics of the participants in the study according to the variables under study.
Demographic of the participants (n = 423).
| Variable | Variable classes | Region | N | % | ||||
|---|---|---|---|---|---|---|---|---|
| Riyadh | Al Jouf | Hasa | Taif | Jazan | ||||
| Type | Males | 95 | 51 | 75 | 27 | 54 | 302 | 71.39 |
| Females | 41 | 21 | 32 | 9 | 18 | 121 | 28.60 | |
| Employer | Governmental | 80 | 44 | 51 | 21 | 53 | 249 | 58.86 |
| Private | 43 | 32 | 39 | 17 | 43 | 174 | 41.13 | |
| Educational level | Bachelor’s | 78 | 41 | 47 | 16 | 43 | 225 | 53.19 |
| Postgraduate | 48 | 41 | 45 | 13 | 51 | 198 | 46.80 | |
| Job | Faculty member | 29 | 14 | 16 | 9 | 10 | 78 | 18.43 |
| Special education teacher | 88 | 39 | 47 | 12 | 51 | 237 | 56.28 | |
| Special education supervisor | 38 | 17 | 19 | 7 | 27 | 108 | 25.53 | |
| Years of experience | <5 years | 39 | 14 | 17 | 11 | 30 | 111 | 26.24 |
| 5-10 years | 53 | 37 | 39 | 14 | 50 | 193 | 45.62 | |
| >10 years | 36 | 17 | 15 | 12 | 39 | 119 | 28.13 | |
| Total | 423 | |||||||
Source: Statistical analysis of study data.
Ethical considerations
This research project received ethical approval and was given the reference number 4400933928/1444 AH. Informed consent was obtained from all participants (faculty member, special education teacher, special education supervisor) included in the study. The goal of the study was made clear to each participant, and confidentiality was upheld throughout data processing and publication of the findings. The survey ran for 8 weeks.
Instrument
For this study, the authors developed a questionnaire whose items were adapted from previous studies (Al-Bashar, 2020; Al-Habib, 2022). The questionnaire consisted of 2 parts: The first part collected general data, while the second consisted of 35 items. These items were further classified into 4 dimensions: (i) financial requirements for the application of AI in the diagnosis of ASD (8 items; e.g. “allocating funds for the use of technical experts in the field of AI”), (ii) human requirements for the application of AI in the diagnosis of ASD (9 items; e.g. “provide technicians for computer maintenance and network troubleshooting”), (iii) regulatory requirements for the application of AI in the diagnosis of ASD (6 items; e.g. “developing programs and AI application models in diagnosing ASD”), and (iv) barriers to the application of AI in diagnosing ASD (12 items; e.g. “the lack of experience among specialists in diagnosing ASD in the field of AI applications”). Responses to the questionnaire were graded on a five-point Likert scale ranging from 1 (strongly agree) to 5 (strongly disagree). Google Forms were used to design and distribute the questionnaire.
The authors assessed the validity of the content by taking the opinions of five professionals diagnosed with ASD and revised the tool accordingly. The validity and reliability of the instrument were also verified through confirmatory factor analysis using the statistical analysis program Amos (version 26, Analysis of Moment Structures, Seattle, USA). The non-standard regression coefficients, standard regression coefficients, standard error, and critical value equivalent to the value of t and its significance were calculated, as presented in Figure 1. Conformance indicators were also calculated to ensure good conformity of the proposed model, as shown in Table 2. The stability of Cronbach’s alpha tool was 0.958, and the sub-scale financial requirements, human requirements, organizational requirements, and application barriers were 0.959, 0.976, 0.943, and 0.911, respectively.
Confirmatory factor analysis to identify the requirements for employing artificial intelligence.
Indicators of conformity to the confirmatory factor analysis model of the questionnaire for the application of AI in diagnosing ASD among participants.
| M | Matching indicators | Value of the pointer | Ideal range of the indicator | Decision |
|---|---|---|---|---|
| 1 | Root mean square residual | 0.047 | 0 to (0,1) | Acceptable |
| 2 | Good fit index | 0.917 | 0-1 | Acceptable |
| 3 | Corrected GFI | 0.945 | 0-1 | Acceptable |
| 4 | Normative conformity index | 0.909 | 0-1 | Acceptable |
| 5 | Relative conformity index | 0.924 | 0-1 | Acceptable |
| 6 | Increasing conformity index | 0.904 | 0-1 | Acceptable |
| 7 | Tucker–Lewis index | 0.891 | 0-1 | Acceptable |
| 8 | Comparative fit index | 0.952 | 0-1 | Acceptable |
Source: Statistical analysis of study data.
Abbreviations: AI, artificial intelligence; ASD, autism spectrum disorder.
It is clear from Figure 1 that all the non-normative and standard regression coefficients had significant critical values, which indicates the validity of the proposed factorial structure model to identify the requirements for the application of AI in diagnosing ASD among the participants in the questionnaire.
It is clear from Table 2 that all the values of the matching indicators were within the ideal range, which indicates that the confirmatory factor analysis model of the questionnaire was highly compatible with the data of the participants in the preparation of the questionnaire.
The authors grouped the responses based on the weight provided to the options (strongly agree = 5, agree = 4, moderately agree = 3, disagree = 2, strongly disagree = 1) to assess the level of reaction to the tool’s items for interpreting the results. The responses were divided into three levels of equal range by first calculating the range (5-1 = 4), then dividing the result by the number of levels to get the category length (4/3 = 1.33), and finally adding this value to the minimal alternatives, which was 1. The following classification was achieved: 1-2.33 denoted a low level, 2.34-3.67 denoted a medium level, and 3.68-5.00 denoted a high level.
Statistical treatment
The means, standard deviations, and frequencies were compared. Then, the variables were compared using the one-way analysis of variance test, the t-test for one group, and the relative weight of the questionnaires to analyze the results in accordance with the objectives and research questions. Additionally, to determine the validity and reliability of the study instrument, Cronbach’s alpha was used, along with coefficients for apparent validity, confirmatory validity, and reliability.
RESULTS
Requirements for the application of AI in diagnosing ASD, as perceived by specialists
The authors calculated the mean, standard deviation, and relative weight of the requirements for the application of AI in diagnosing ASD, as shown in Table 3.
The mean, standard deviation, and relative weight requirements for the application of AI in the diagnosis of ASD.
| Requirement | Mean | Standard deviation | Relative weight (%) | Arrangement | |
|---|---|---|---|---|---|
| 1 | Financial requirements | 4.00 | 1.23 | 80 | 3 |
| 2 | Human requirements | 4.09 | 1.07 | 81.8 | 2 |
| 3 | Regulatory requirements | 4.16 | 1.01 | 83.2 | 1 |
Source: Statistical analysis of study data.
Abbreviations: AI, artificial intelligence; ASD, autism spectrum disorder.
It is clear from Table 3 that most of the requirements for the application of AI in diagnosing ASD, as perceived by specialists, are regulatory. The mean for regulatory requirements was 4.16, with a standard deviation of 1.01 and a relative weight of 83.2%. Human requirements were ranked second with a mean of 4.09 and a standard deviation of 1.07, representing a relative weight of 81.8%; it was followed by financial requirements with a mean of 4.00 and a standard deviation of 1.23, accounting for a relative weight of 80%. The results of each requirement are discussed below separately.
Financial requirements
It is evident from Table 4 that the items that express the financial requirements for the application of AI in diagnosing ASD, as perceived by specialists, had means ranging between 3.81 and 4.41, and they are all high values. Item 5 had the highest mean (4.41), while item 4 had the lowest mean (3.81).
Frequencies, mean, and standard deviations of the financial requirements for the application of AI in the diagnosis of ASD as perceived by specialists.
| Requirement | M | Item | Mean | Standard deviation | Iterations | Arrangement | Level | ||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Strongly agree | Agree | Moderately agree | Disagree | Strongly disagree | |||||||
| Financial requirements | 1 | Providing sufficient financial support for the purchase of computer hardware, software, and modern applications used in the process of diagnosing ASD | 3.83 | 1.49 | 230 | 62 | 7 | 81 | 43 | 7 | High |
| 2 | Providing appropriate financial support for the maintenance of computer hardware, software, and applications required for the diagnosis of ASD | 3.89 | 1.39 | 218 | 74 | 43 | 46 | 42 | 4 | High | |
| 3 | Providing the necessary financial allocations for the development of programs and applications used in the process of diagnosing ASD | 3.89 | 1.40 | 218 | 75 | 39 | 48 | 43 | 5 | High | |
| 4 | Allocating sums of money for the use of technology experts in the field of AI | 3.81 | 1.38 | 198 | 81 | 52 | 50 | 42 | 8 | High | |
| 5 | Allocating sufficient budget for training and qualification programs for specialists in the diagnosis of ASD internally and externally | 4.41 | 0.84 | 250 | 116 | 42 | 11 | 4 | 1 | High | |
| 6 | Allocating a sufficient budget to introduce academic specializations in Saudi universities to prepare specialists in diagnosing ASD using AI | 4.17 | 1.03 | 217 | 112 | 48 | 43 | 3 | 2 | High | |
| 7 | Allocate appropriate incentives and rewards for outstanding specialists in the application of AI to diagnose ASD | 3.87 | 1.26 | 203 | 63 | 64 | 85 | 8 | 6 | High | |
| 8 | Provide the necessary financial allocations from the concerned authorities for the networking between centers and clinics for diagnosing ASD and the Ministry of Education, for application in the diagnostic process | 4.14 | 1.09 | 225 | 94 | 52 | 46 | 6 | 3 | High | |
Source: Statistical analysis of study data.
Abbreviations: AI, artificial intelligence; ASD, autism spectrum disorder.
Human requirements
It is clear from Table 5 that the items that express the human requirements for the application of AI in diagnosing ASD, as perceived by specialists, had means ranging between 3.80 and 4.37, which are all high values. Item 9 had the highest average (4.37), while item 16 had the lowest mean (3.80).
Frequencies, means, and standard deviations of the human requirements for the application of AI in the diagnosis of ASD as perceived by specialists.
| Requirement | M | Item | Mean | Standard deviation | Iterations | Arrangement | Level | ||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Strongly agree | Agree | Moderately agree | Disagree | Strongly disagree | |||||||
| Human requirements | 9 | Providing technicians for computer maintenance and network fault handling | 4.37 | 0.99 | 264 | 102 | 10 | 44 | 3 | 1 | High |
| 10 | Presence of informed administrative leaders who adopt and support the application of AI in diagnosing ASD | 4.15 | 0.90 | 200 | 97 | 118 | 7 | 1 | 3 | High | |
| 11 | Provide qualified trainers to train specialists in the diagnosis of ASD on the use of AI applications in the diagnosis process | 4.14 | 1.10 | 241 | 53 | 80 | 48 | 1 | 4 | High | |
| 12 | Specialists in the diagnosis of ASD have the ability to reconcile the applications of AI and the human aspects in the diagnosis process | 4.13 | 0.99 | 192 | 138 | 51 | 40 | 2 | 5 | High | |
| 13 | Providing experts for the design and planning of AI applications in the process of diagnosing ASD | 4.05 | 1.15 | 203 | 128 | 5 | 85 | 2 | 7 | High | |
| 14 | Providing specialists trained in the use of systems for diagnosing ASD using AI applications | 4.06 | 1.14 | 206 | 124 | 11 | 80 | 2 | 6 | High | |
| 15 | Finding experts to evaluate the results of the application of AI by specialists in the diagnosis of ASD in the diagnosis process | 3.99 | 1.10 | 199 | 78 | 96 | 46 | 4 | 8 | High | |
| 16 | Presence of administrators familiar with the rules and regulations governing the application of AI in the process of diagnosing ASD | 3.80 | 1.31 | 166 | 136 | 35 | 45 | 41 | 9 | High | |
| 17 | Training specialists diagnosing ASD on the application of AI in the diagnosis process | 4.19 | 1.03 | 225 | 105 | 46 | 45 | 2 | 2 | High | |
Source: Statistical analysis of study data.
Abbreviations: AI, artificial intelligence; ASD, autism spectrum disorder.
Regulatory requirements
It is evident from Table 6 that the items that express the regulatory/administrative requirements for the application of AI in diagnosing ASD, as perceived by specialists, had means ranging between 3.81 and 4.39, all of which are high values. Item 21 had the highest mean (4.39), while item 20 had the lowest mean (3.81).
Frequencies, means, and standard deviations of regulatory requirements for the recruitment of AI in the diagnosis of ASD, as perceived by specialists.
| Requirement | M | Item | Mean | Standard deviation | Iterations | Arrangement | Level | ||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Strongly agree | Agree | Moderately agree | Disagree | Strongly disagree | |||||||
| Regulatory requirements | 18 | Finding systems that require specialists in the diagnosis of ASD to apply AI in diagnosis | 4.05 | 0.96 | 155 | 186 | 34 | 46 | 2 | 5 | High |
| 19 | Issuing laws regulating the application of AI in the diagnosis of ASD | 4.21 | 0.96 | 202 | 162 | 11 | 46 | 2 | 4 | High | |
| 20 | Changing the traditional procedural policies related to ASD diagnosis systems in line with the application of AI | 3.81 | 1.06 | 155 | 87 | 132 | 47 | 2 | 6 | High | |
| 21 | Spreading and consolidating the culture of AI among specialists in diagnosing ASD | 4.39 | 0.98 | 274 | 93 | 9 | 45 | 2 | 1 | High | |
| 22 | Developing programs and models for the application of AI in the diagnosis of ASD | 4.21 | 1.05 | 235 | 94 | 47 | 43 | 4 | 3 | High | |
| 23 | Holding courses and workshops for specialists in the diagnosis of ASD, to disseminate systems for AI in the diagnostic process | 4.30 | 1.08 | 274 | 57 | 42 | 46 | 4 | 2 | High | |
Source: Statistical analysis of study data.
Abbreviations: AI, artificial intelligence; ASD, autism spectrum disorder.
Barriers/challenges in the application of AI in diagnosing ASD, as perceived by specialists
It is clear from Table 7 that the barriers/challenges mentioned in the questionnaire greatly affect the use of AI in diagnosing ASD, as perceived by specialists, as the mean ranged between 4.02 and 4.55, which are all high values. Item 34 had the highest mean (4.55), while item 31 had the lowest mean (4.02).
Frequencies, means, and standard deviations for barriers/challenges in the recruitment of AI in the diagnosis of ASD as perceived by specialists.
| Requirement | M | Item | Mean | Standard deviation | Iterations | Arrangement | Level | ||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Strongly agree | Agree | Moderately agree | Disagree | Strongly disagree | |||||||
| Barriers/challenges in the application of AI | 24 | Lack of readiness of hardware and software in clinics and centers to apply AI in the process of diagnosing ASD | 4.37 | 0.86 | 254 | 87 | 73 | 6 | 3 | 4 | High |
| 25 | Absence of a network link between centers and clinics for diagnosing ASD and the Ministry of Education, to determine an accurate diagnosis | 4.36 | 0.83 | 245 | 97 | 75 | 4 | 2 | 5 | High | |
| 26 | Lack of experience among specialists in diagnosing ASD in the field of AI applications | 4.31 | 0.84 | 218 | 138 | 50 | 16 | 1 | 7 | High | |
| 27 | High financial costs required for the application of AI in the process of diagnosing ASD | 4.34 | 0.81 | 227 | 124 | 63 | 8 | 1 | 6 | High | |
| 28 | Lack of human cadres with computer skills to apply AI in diagnosing ASD | 4.44 | 0.75 | 236 | 153 | 25 | 4 | 5 | 2 | High | |
| 29 | Poor qualification of specialists in the diagnosis of ASD in the field of technology, for the application of AI in diagnosis | 4.39 | 0.86 | 247 | 120 | 34 | 19 | 3 | 3 | High | |
| 30 | The large number of assignments and administrative burdens placed on the shoulders of specialists diagnosing ASD | 4.14 | 1.04 | 200 | 146 | 20 | 53 | 4 | 10 | High | |
| 31 | Weakness of the infrastructure of ASD clinics and diagnostic centers for the application of AI software in diagnosis | 4.02 | 1.04 | 181 | 123 | 70 | 46 | 3 | 12 | High | |
| 32 | Weakness of the legal protection that controls the use of AI in the diagnosis of ASD | 4.12 | 0.96 | 197 | 110 | 91 | 22 | 3 | 11 | High | |
| 33 | Resistance to change by ASD diagnosticians in developing diagnostic methods for many reasons such as their ignorance of AI | 4.17 | 1.00 | 210 | 119 | 53 | 38 | 3 | 9 | High | |
| 34 | Weak awareness of the importance of AI applications in diagnosing ASD | 4.55 | 0.70 | 276 | 114 | 25 | 7 | 1 | 1 | High | |
| 35 | Absence of regulations governing the process of using and applying AI in diagnosing ASD | 4.28 | 1.02 | 243 | 110 | 18 | 51 | 1 | 8 | High | |
Source: Statistical analysis of study data.
Abbreviations: AI, artificial intelligence; ASD, autism spectrum disorder.
DISCUSSION
The major goal of the current study was to determine the prerequisites for using AI to diagnose ASD and the difficulties that can arise in doing so, as seen by experts in the KSA. The study’s findings not only offered insightful analysis of the literature but also strengthened the conclusions of earlier related studies by expanding their scope, demonstrating the value of using AI in diagnosing ASD.
The study’s findings revealed that organizational requirements, followed by human needs and financial requirements, are deemed by experts to be the most crucial for the application of AI in diagnosing ASD. It demonstrates the significance of the two key organizational and human needs, which are prerequisites for the use of AI in the diagnosis of ASD. This may be attributable to the participants’ understanding of the importance of regulatory standards in the diagnosis process, whose absence or inaccuracy causes several issues. It also shows that despite the specialists’ need to fulfill the requirements for applying AI to the diagnosis of ASD, the material requirements do not receive enough attention. This may be because of inadequate financial allocation or a lack of funding for the purpose of offering material and moral awards to outstanding and innovative specialists in the diagnosis of ASD.
The results shown in Table 4 suggest that a high number of respondents were concerned about the financial requirements for the application of AI in diagnosing ASD. Item 5 had the highest mean (4.41), indicating that it is a very important financial requirement. It indicates that the specialists believe that the necessary budget should be allocated for training and qualifying programs for specialists in the diagnosis of ASD in a better way, and they seek to prioritize such initiatives in the future. Item 4 had the lowest mean (3.81), which indicates that funds are not allocated for the assistance of technology experts in the field of AI. This result is in agreement with the findings of Alqaysi et al. (2022), who found that MRI is expensive and not available in many hospitals, especially in resource-poor countries. The financial aspect of using AI to diagnose ASD is not solely determined by the availability of financial resources, but it is also dependent on the financing strategy chosen by the ministries of education and health. This strategy depends on government spending to provide the necessities for diagnosing ASD, which is what attracted the participants. The respondents believe that without investing adequate funds for training and rehabilitation programs for professionals with ASD, achieving effective diagnosis and care for individuals with ASD can be challenging.
The authors also credit this outcome to professionals’ growing recognition of the value of financial requirements, which are crucial to ensuring sufficient funding for the acquisition of new applications, computer hardware, and software utilized in the process of diagnosing ASD and offering financial and moral incentives to eminent and inventive experts in the field of ASD diagnosis.
The results presented in Table 5 suggest that specialists perceive the value of human requirements for the application of AI in diagnosing ASD to be high, as the mean ranged between 3.80 and 4.37. Item 9 had the highest mean (4.37), which indicates that adequate technical support must be provided to maintain computers and deal with network faults. It also points to the urgent need for AI experts to be present in ASD diagnostic centers and clinics. Furthermore, item 16 had the lowest mean (3.80), which indicates that a radical structural change is required in terms of eliminating outdated jobs and creating new ones that are compatible with AI applications. This result can be explained by the fact that the availability of experts for designing and planning the application of AI in the ASD diagnostic process is less prioritized compared to the presence of administrators familiar with the rules and regulations governing such applications.
All of these results indicate that the human requirements for the application of AI in the diagnosis of ASD pertain to the availability of experts for designing and planning the application of AI in the process of diagnosing ASD, in addition to the presence of aware administrative leaders who adopt and support the application of AI in the diagnosis of ASD.
The regulatory requirements had an important role in the application of AI in diagnosing ASD, as perceived by specialists, as the means ranged between 3.81 and 4.39, which are all high values. It indicates that the specialists confirm their support for the items on the regulatory requirement axis. These results confirm that the fulfillment of regulatory requirements can help achieve better performance of AI in diagnosing ASD. Item 21 had the highest mean (4.39), which indicates that the most important of these requirements, as perceived by the participants, is to spread and consolidate the culture of the application of AI in diagnosing ASD among specialists. Other important items included conducting courses and workshops for specialists in the diagnosis of ASD to disseminate knowledge about AI systems and their applications in the diagnostic process. Item 20 had the lowest mean (3.81); this indicates that specialists believe that regulatory requirements call for a change in the traditional procedural policies related to ASD diagnosis systems in line with recent developments in the field. This result is consistent with the findings of Peya et al. (2020), who found that although EEG can be used to diagnose ASD, it still has limitations in several conditions (such as signal noise).
The results also indicate that the participants agreed on the existence of barriers that prevent the fulfillment of the requirements for the application of AI in diagnosing ASD. The authors believe that this represents a major problem in the process of diagnosing ASD in general, and solutions must be found to address these barriers.
Item 34 had the highest mean (4.55), indicating that the biggest obstacle is people’s ignorance of the value of AI application in the diagnosis of ASD. It suggests that experts in the diagnosis of ASD are not fully convinced of the value of using AI in the diagnosis procedure. This result is in line with Jaliawala and Khan’s (2020) assertion that the non-experimental nature of the published data makes clinicians less likely to be persuaded of the efficacy of technology-based therapies.
This result, according to the authors, is due to the specialists’ lack of experience in diagnosing ASD using AI applications, as well as the absence of human cadres with computer skills. These cadres must have a clear desire to use AI in diagnosing ASD, as well as the necessary experience and awareness of the importance of technology.
The infrastructure of ASD clinics and diagnostic facilities was doubled for the deployment of AI software in diagnosis, even though item 31 had the lowest mean (4.02). This is a surprising outcome that the authors explain by pointing out that the participants had no prior exposure to the AI tools used to diagnose ASD. The lack of an integrated infrastructure for clinics and centers diagnosing ASD, in which all the necessary equipment and technical services are available, leads to difficulty in the use of AI applications and software. It negatively affects the motivation of specialists in the field to employ this technique for diagnosing ASD. This finding is in line with that of Drigas and Ioannidou (2012), who found that teachers, parents, and therapists must be supported in providing adequate care for students with disabilities, particularly in terms of techniques of evaluation and treatment and time and money savings through the use of AI. Additionally, it highlights the need for permanent teams of experts within ASD diagnosis clinics and centers so that experts in the field can comprehend and assimilate the applications of AI used in diagnosis, find solutions when issues arise, and hold training sessions for others to eliminate the anticipated resistance to change.
CONCLUSION
The results of the research showed that the most important requirements for the application of AI in diagnosing ASD, as perceived by specialists, are organizational requirements, followed by human requirements and financial requirements. Therefore, specialists and agencies concerned with diagnosing people with ASD must fulfill these requirements to facilitate the use of AI. The results also showed that there are many barriers to the application of AI in diagnosing ASD, the most important of which is the lack of awareness of the importance of AI applications in diagnosing ASD.
RECOMMENDATIONS
In light of the study’s findings, the authors suggest using AI to speed up and simplify the diagnosis of ASD while also taking conceptual and methodological considerations into account. Additionally, modern research requires an interdisciplinary approach. Thus, the technologists who develop the algorithms must be well versed in ASD, and experts need to be digitally savvy because transdisciplinary research collaboration can prove successful in the study of ASD. Furthermore, the appropriate government agencies must provide adequate funding for the use of AI applications in diagnosing ASD.
In addition to assigning a sufficient budget for the purchase of computers, projectors, and the contemporary software required for the use of AI in the process of diagnosing ASD, the budget should also include funds for the routine maintenance of devices and internet-based communication networks. Finally, it is essential to provide seminars for administrative leaders to help them adopt an AI-centric mindset while diagnosing ASD; this will have a significant impact on the accuracy and speed of the diagnostic process.
The authors aim to raise awareness among readers of the theoretical insights provided in this study about the conditions that strengthen the use of AI for the diagnosis of ASD. Ultimately, the relevant ministries must oversee the purposeful utilization of AI in diagnosing ASD. They should align the objectives of the specialized centers with the ministry’s overall objectives, support the integration of specialists into their workplaces, use cutting-edge diagnostic tools including AI techniques, and foster the professional growth of these individuals and enable them to diagnose ASD with greater speed and accuracy.
LIMITATIONS AND DIRECTION FOR FUTURE RESEARCH
Despite providing valuable practical insights for professionals and family members of children with ASD, the current study has several limitations. Self-report data, which were gathered via survey methods, should be analyzed with care because they could be biased on the part of the participants. Moreover, it cannot be assured that cross-sectional studies are representative of the population because they represent a single point in time rather than longitudinal monitoring. This issue also restricts the generalizability of our findings due to the very low response rate of the survey among professionals. To corroborate our findings, additional research using different independent samples is required, as we are aware that the sample selection procedure itself has limitations.
Future research should focus on the difficulties faced by ASD diagnostic centers and clinics when applying AI in the diagnosis process. In addition, studying the impact of using AI on early diagnosis of ASD could have promising results. Future work should test a training program based on AI applications to develop the skills of specialists in the early diagnosis of ASD. It could also be interesting to study the effectiveness of using AI applications and models in predicting ASD. While our study did not investigate the effect of age, gender, educational level, and duration of experience on the specialists’ awareness of the requirements for the application of AI in diagnosing ASD, further studies are recommended to explore this aspect further. Finally, conducting a study on the competencies of relevant stakeholders in applying AI in the diagnostic process of ASD could offer promising insights.