AUTISM DIAGNOSIS WITH DEEP LEARNING MODEL TRAINED WITH FACE IMAGES


Noğay H. S.

3rd International Congress on Engineering Sciences and Multidisciplinary Approaches, İstanbul, Turkey, 10 - 11 February 2022, pp.552-558

  • Publication Type: Conference Paper / Full Text
  • City: İstanbul
  • Country: Turkey
  • Page Numbers: pp.552-558

Abstract

Autism spectrum disorders (ASD) are among the diseases of our age that cannot be treated, have a long diagnosis period, and have a high rate of disadvantage. Its incidence is estimated to be increasing gradually. Definitive and rapid diagnosis of autism in early childhood can reduce the negative effects of this disease. In this study, a deep learning model was applied for automatic diagnosis of autism with facial images. The deep convolutional neural network (DCNN) model, which is one of the most popular deep learning methods used in many fields in recent years among many deep learning methods, was designed by utilizing the pre-trained Resnet-18 model. Through the transfer learning approach, the designed deep learning model was trained with the new dataset without changing its weights and was also tested with a randomly selected 20% of the whole dataset. As a result of the test process, ASD diagnosis was made with an accuracy rate of 82.3%. The success of the study has been proven both by the test result and by comparing it with other ASD diagnostic studies and methods.