DEVELOPMENT OF ARTIFICIAL INTELLIGENCE-BASED AUTISM SPECTRUM DISORDER DETECTION FRAMEWORK

Abstract

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition characterized by challenges in social communication, restricted interests, and repetitive behaviors. The condition manifests in early childhood and persists throughout life, impacting an individual’s ability to interact with their environment. ASD is highly heterogeneous, with symptoms and severity varying widely across individuals, making its diagnosis and management particularly challenging. Early and accurate identification of ASD is crucial, as timely interventions can significantly improve developmental outcomes and enhance the quality of life for those affected. The traditional approach to diagnosing ASD primarily relies on clinical observations, caregiver reports, and standardized behavioral assessments. While effective in many cases, these methods are often time-consuming, subjective, and dependent on the expertise of clinicians. This reliance on subjective evaluation introduces variability and delays in diagnosis, particularly in regions with limited access to specialized healthcare services. Consequently, there is a growing need for innovative solutions to improve the efficiency and accuracy of ASD detection. Artificial Intelligence (AI) has emerged as a transformative force in healthcare, offering powerful tools for analyzing complex and diverse datasets. By leveraging AI techniques, it is possible to identify patterns and relationships within data that might not be readily apparent through traditional analysis. Machine learning and deep learning, subsets of AI, have demonstrated significant potential in various domains, including image analysis, natural language processing, and predictive modeling. These capabilities make AI particularly well- suited for addressing the challenges associated with ASD diagnosis. AI-driven approaches offer the advantage of objectivity, scalability, and the ability to integrate multiple data modalities, such as behavioral data, clinical records, and imaging studies. Furthermore, AI can facilitate early detection by identifying subtle patterns indicative of ASD, even in cases that might be missed by conventional diagnostic methods. As a result, AI-based tools have the potential to complement existing clinical practices, enhance diagnostic precision, and expand access to reliable ASD detection in underserved areas. The application of AI in ASD detection is an evolving field, with ongoing research aimed at developing innovative methods to tackle the complexities of the disorder. By combining advances in AI with insights from neuroscience and psychology, researchers aim to create solutions that not only improve diagnostic accuracy vi but also offer interpretable results that can support clinical decision-making. Through such interdisciplinary efforts, AI holds promise in transforming the landscape of ASD diagnosis and care, ultimately contributing to better outcomes for individuals and their families. Therefore, this study represents a structured and methodical effort to assess the effectiveness, potential, and applicability of deep learning and computational intelligence techniques in the identification and analysis of ASD. Objectives: The objectives of this study are structured into four key segments: • The first objective of the study is to perform a systematic literature review on Autism Spectrum Disorder (ASD), which aims to critically evaluate the existing research, methodologies, and advancements in ASD detection. • The second objective focuses on developing an intelligent diagnostic model for ASD using deep learning and computational intelligence techniques, aiming to improve the accuracy and efficiency of diagnosis. • The third objective is to design a multi-modal framework for ASD detection, incorporating various data sources/modalities to enhance the overall performance of the diagnostic model. • The final objective is to conduct a comparative analysis of the proposed ASD detection model with existing techniques, evaluating its effectiveness, accuracy, and applicability in real-world clinical settings. Methodology: To accomplish the stated objectives, this study leverages advanced machine learning and deep learning methods, such as evolutionary algorithms, neural networks, attention mechanisms, and transformer-based architectures, due to their significant potential in addressing complex challenges in healthcare. The strategies employed to meet these objectives are as follows: • To accomplish the first objective, a systematic literature review was conducted, focusing on machine learning techniques applied to Autism Spectrum Disorder (ASD) detection. This review analyzed various studies to identify the most effective models, methodologies, and data sources used in ASD diagnosis. • For the second objective, two diagnostic models were developed, each utilizing different deep learning and evolutionary approaches. The first model incorporated an adaptive feature fusion technique to enhance the diagnosis process by combining vii various data features obtained from particle swarm optimization (PSO) and the Bat algorithm effectively. The second model integrated a white shark optimization algorithm with a deep learning framework utilizing Bi-LSTM to improve the overall accuracy and robustness of ASD detection. • To address the third objective, a multimodal diagnostic framework was designed that combines various data modalities, such as clinical features and imaging data. This framework employs advanced deep learning techniques, including a multi-head CNN architecture with channel and spatial attention (CBAC) and BERT, to extract and integrate features from diverse modalities for enhanced ASD detection. • For the fourth objective, a comparative analysis was conducted, evaluating the performance of the above-developed models against existing ASD detection techniques. Key performance metrics, such as accuracy, sensitivity, specificity, and F1- score, were used to compare the effectiveness of the proposed models with current state-of-the-art methods. Results: The outcomes of the study are as follows: • A comprehensive review of machine learning techniques for Autism Spectrum Disorder (ASD) detection was conducted. This review analyzed current trends and identified potential future directions in the field, providing valuable insights into existing methodologies and areas for future research. • A study is conducted to explore bio-inspired techniques for improving ASD diagnosis, with a focus on evolutionary algorithms. The study highlighted the promising potential of these algorithms in enhancing diagnostic accuracy for ASD detection. • An adaptive feature fusion technique was developed for ASD diagnosis. This hybrid model combined bio-inspired optimization algorithms with feature fusion to effectively integrate various data features, enhancing the accuracy and robustness of the diagnostic process. • A model was developed by integrating an optimization algorithm with the Bi-LSTM approach. This strategy aimed to improve feature selection, ultimately leading to improved overall performance in the ASD detection system. • A multi-modal framework was created, integrating sequential (phenotypic information) and visual data (brain MRI) using convolutional block attention component and BERT- viii based deep learning architectures. This framework significantly improved ASD detection accuracy and robustness by effectively combining different data types. • A new approach employing facial images of autistic and non-autistic children combining convolutional networks and vision transformers was developed for the diagnosis of ASD. This model enhanced the processing of visual data, leading to improved diagnostic performance. • A self-supervised and self-distillation learning approach was explored for ASD classification using facial images. This innovative method aimed to leverage unsupervised learning to improve the classification accuracy in ASD detection. • A multi-modal diagnostic framework was designed, incorporating various data sources such as clinical and imaging data. This framework leveraged advanced deep learning techniques, including LSTM and transformer-based architectures, to extract and integrate relevant features, improving the diagnostic performance.