A serious, persistent and complex disease that affects the human being is autism. Lack of communication and social connection are the key symptoms of this disease (Brun et al., 2016). Children with autism have weak imaginations, limited interests and trouble understanding emotional expressions. Autism is associated with behavioral issues and severe cognitive in the absence of cerebral dysmorphology, so it captured the attention of the scientific and medical communities. Identifying brain areas with aberrant functionality or structure is a key objective of the neurobiological investigation of autism disorder (Uddin et al., 2011). Better methods for the early treatment and detection of autism may be used once the anomalies have been characterized and recognized. Environmental influences, brain function and structure, also the combinations of genetics are the root causes of autism. Most frequently, autism diagnostic interviews and the autism diagnostic observation plan are utilized to make today’s interview-based diagnoses for autism (Mori et al., 2020). These techniques are highly precise, but if the neuroanatomy is unknown, they cannot identify the molecular causes of behavioral problems. However, during the past few years, more research has been done on the functional and structural anomalies of the brain using MRIs (Guo, 2020). Although MRI investigations have offered various involvements of neurodevelopment factors behind ASD, the results often do not hold over the complete collection of ASD cases (Devika et al., 2022). Functional MRI (fMRI) is utilized to know about the connectivity patterns in anomaly brain architecture as well as structural MRI (sMRI) studies typically concentrate on morphometric and volumetric analysis in the abnormal brain (Traut et al., 2022). An essential brain imaging technique that provides high-resolution data (Huang et al., 2023) on the function and structure of brain is MRI (Kuttala et al., 2022). The fMRI of resting state brain has been investigated to learn more about the neurological processes of ASD (Mei et al., 2023). fMRI images are used to measure the Blood Oxygen Level Dependent (BOLD) signals in various regions of brain, which is useful in finding out different neural connections in the brain (Bednarz et al., 2021). The structure, functioning and information processing of brain are thought to be strongly indicated by the spontaneous changes in the BOLD signals in resting conditions (Chen et al., 2023a). By utilizing task-based fMRI investigations, the cause of autism was investigated first (Walsh et al., 2021). The brain connections were also studied in the later part of research by utilizing resting state fMRI images that revealed a significant abnormalities which may be connected to some of the primary symptoms seen in ASD-affected persons (Gomot et al., 2006).

In recent days, for the classification of ASD, deep learning approaches have been widely used. The Multi-Layer Perceptron (MLP) network is one of the most utilized techniques (Safar et al., 2021). However, all the deep learning techniques cannot be able to achieve high accuracy. So, the limitations of these techniques have to be rectified. For the classification phase, only a limited number of subjects has been employed it resulting in unreliable outcomes (Esfahlani et al., 2022). The data needed for the research is collected from a single site for different studies is another drawback (Gordon et al., 2021). So, the resultant outcomes are not effective. Some of the researchers do not focus on structural information, and they consider functional findings only (Bernas et al., 2018). In order to solve these issues, it is necessary to design a new autism detection model utilizing deep learning approaches to address challenges such as noise issues and poor generalization.

The vital contribution of the proposed deep learning-based autism detection in children is mentioned in the following:

To develop a deep learning-based autism detection in children to detect ASD in the initial stage for improving their skills and communication.

To improve the performance of the autism detection approach, an RA-EBO algorithm is introduced. It is used to increase the precision and accuracy of the prediction model by optimizing parameters from SSM-AM and MDA-AUnet.

To locate the abnormal autism region, the MDA-AUnet is developed. In MDA-AUnet, parameters such as filter size, activation function and batch size are optimized by the developed RA-EBO algorithm.

To validate the overall performance of the proposed RA-EBO-SSM-AM model for autism detection is checked among other optimization approaches and prediction models.

The developed RA-EBO-SSM-AM model for detecting autism in children is described in the upcoming sections. In Section 2, the existing autism detection approaches with their benefits and limitations are mentioned. In Section 3, the illustration of the proposed model and the developed optimization algorithm, along with the detection process, are explained. In Section 4, the overall preprocessing steps and the categorization of normal and abnormal images are explained. The region localization process is described in Section 5. The results and discussions are provided in Section 6. The summary of the developed autism detection model is given in Section 7.