MULTI-SCALE ROTATION INVARIANT CONVOLUTIONAL NEURAL NETWORKS FOR PREDICTION OF AUTISM SPECTRUM DISORDER

Abstract

Autism spectrum disorder (ASD) is a complex developmental condition that affects communication and behavior. It can manifest itself in a wide range of symptoms and abilities. ASD might be a small issue or a severe condition that necessitates full-time care in a facility. Communication is difficult for people with autism. They have a hard time comprehending what other people are thinking and feeling. This makes it difficult for individuals to communicate, whether through words, gestures, facial expressions, or touch. Learning difficulties may be an issue for people with autism. Their abilities may develop in a haphazard manner. For example, someone may struggle with communication yet excel at art, music, arithmetic, or memory. As a result, individuals may perform particularly well on analytical or problem-solving tests. Autism is currently being diagnosed in greater numbers than ever before. However, the latest figures could be higher due to changes in how the illness is diagnosed, not because more youngsters have it. This paper   presents a method to detect Autism using Multi- Scale Rotation Invariant Convolutional Networks (MSCNN) on the real time dataset. By leveraging multi-scale convolutional layers, MSCNN is employed to identify both detailed and general patterns in the data, which improves the model's capability to effectively detect autism. The results demonstrate the effectiveness of the model by achieving high accuracy, thus proving the capability of deep learning models in medical diagnosis of   brain related diseases. The proposed model’s performance demonstrates its ability to yield reliable predictions that can assist healthcare professionals to detect autism in the early stages, thereby aiding in the enhanced treatment and control of the condition.