The Early Signs of Autism in First Year of Life: Identification of Key Factors Using Artificial Neural Networks

Thursday, May 15, 2014
Atrium Ballroom (Marriott Marquis Atlanta)
H. Alonim1, E. Grossi2, I. Liberman3, G. Schayngesicht4 and D. Tayar5, (1)The Mifne Center and Social Science School, Bar Ilan University, Rosh Pina, Israel, (2)Autism Research Unit, Villa Santa Maria Institute, Tavernerio( Como), Italy, (3)Research Authority, Western Galilee Academic College,Bar Ilan University, Rosh Pina, Israel, (4)The Mifne Center, Rosh Pina, Israel, (5)The Mifne Center and Health Care Unit, Health Ministry, Rosh Pina, Israel

In a previous study we have presented an innovative methodology to  detect early manifestations of autism, using retrospective analysis of parents’ video-recordings of their children's first year of life, filmed before any suspicion concerning defective development arose. Traditional statistics did not allow to handle all the information available due to the high intrinsic non linearity and skewed distribution of symptom frequencies. Similar problems hampered the understanding of natural relationships among factors on study, taking into account simultaneously occurrence, their severity and their precocity in onset.

Objectives: The aim of the study is to assess the natural relationships among variables associated with autism onset.  


This continued data set is composed on 8 variables displayed in 110 infants (76. % boys and 24% girls between the ages of 3-15 months) who were diagnosed with autism at the age of 2-3 years, using retrospective analysis of video-recordings of the infants' first year of life. In addition, interview questionnaires were distributed to the parents. Variables investigated were: Excessive Passivity; Excessive activity; Lack of reaction to voice or presence;  Lack of eye contact;  Aversion to touch; Delayed motor development; Accelerated growth of head circumference; Resistance to eating;  All variables were objectively measured according to a validated evaluation form scoring.

Artificial Neural Networks (Auto-CM system) were applied to highlight the  associations among variables under study. Auto-CM is a special kind of Artificial Neural Network developed at Semeion Research Institute (Rome) and successfully applied in many complex chronic degenerative diseases, able to find out consistent trends and associations among variables creating a semantic connectivity map. The matrix of connections, visualized through minimum spanning tree filter, takes into account nonlinear associations among variables and captures connection schemes among clusters.  


The Semantic  Connectivity Map developed by Auto-Cm system showed  a meaningful scheme of connections. Lack of eye contact resulted a major node in the graph directly linked with autism spectrum diagnosis and coordinating  the other three variables (Lack of reaction to voice o presence; Accelerated growth of head circumference; excessive activity); the other major node resulted to be Lack of reaction to voice o presence, coordinating the other four variables in study.


Findings from this study indicate the utility of a data mining approach based on artificial neural networks in depicting complexity of the variables related to early manifestation of autism.