부산대학교 도서관

The medical and health research domain continues to grapple with ongoing obstacles in promptly diagnosing ASD. Swift recognition of ASD holds paramount importance for proficient condition management and the early deployment of intervention tactics. Encouragingly, advancements in deep learning, notably in CNN, present hopeful paths toward overcoming these diagnostic hurdles. This research concentrates on crafting a tailored dataset for deep learning, specifically leveraging CNN, to improve the early detection of ASD. The dataset underwent thorough curation, incorporating diverse tasks encompassing written and visual components, administered to individuals spanning ASD and non-ASD profiles. The objective was to construct a robust dataset that encapsulates the intricacies of handwriting challenges linked to ASD. Following this, the dataset was utilized to devise an automated ASD diagnosis technique employing transfer learning with the GoogleNet architecture. Each handwritten task within the dataset underwent scrutiny and classification via the trained CNN model. This method endeavors to harness the capabilities of deep learning in identifying patterns and characteristics in the handwriting of individuals with ASD, thus advancing towards more precise and streamlined diagnostic procedures.