A Tensor Approach for Diagnosing Autism Spectrum Disorder Using fMRI Data

Autism Spectrum Disorder (ASD) is a mental disease which effects on social skills of the person. ASD has many sub-types in which each of them has its own markers, But most of the time a person with ASD is recognized by repetitive behaviours, speech and nonverbal communication. Based on recent report of Center for Disease Control, ASD is growing. Indicators of ASD usually become visible by age 2 or 3. Moreover, some of its markers can appear even earlier. Researches indicate that early treatment has many positive effects in the life of people with ASD [1]. So early diagnosing the ASD is very important. On the other hand, fMRI is one of the most applicable and accurate instruments of the imaging. Despite of the fMRI benefits, we have some challenges in working with its data. The biggest challenge that we are dealing with is the volume of the data. The volume of fMRI data is very huge and it makes the procedure of the data analyzing, time consuming.Method Recently, machine learning(ML) is used to analyze fMRI data and it got good results. The ML algorithms can establish a trade off between running time and accuracy. In this research, we are going to present a tensor decomposition algorithm [2] for identifying ASD from fMRI data. The recent work on this topic was done in 2018 [3].Results The researchers obtained the average accuracy of 70% using a deep neural network method [2]. But by applying tensor method on ABIDE data set the obtained average accuracy is about 80%.Conclusions As the results show, the presented method is more efficient than the state-of-the-art methods in characterizing the ASD and we expect that this method can be employed for other fMRI data analysis tasks.