Autinosis

An algorithm to assist the screening process for autistic spectrum disorders using Machine Learning: accessible solution to identify ASD in all ages.

The problem of the research is the lack of a satisfactory and accessible diagnosis for ASD, considering that from 2020 to 2022 the cases had an increase of 22% [8]. With this sporadic growth of the disorder, there is a huge lack of professionals and specialized centers able to make a proper diagnosis, in addition to the high cost and difficulty in obtaining treatment due to these factors, besides not being 100% accurate given the lack of specialization of some professionals in the area.

According to the research "Portrait of Autism in Brazil", about 2 million Brazilians are autistic. Even so, the carriers find it difficult to perceive the presence of ASD due to the lack of means of information dissemination about diagnostic medicine.

The relevance of this disorder has become increasingly evident in the global context, since it is the fastest growing developmental (Centers for Disease Control and Prevention, 2020). This organization also released 2014 data indicating that 1% of the world's population has ASD and that its presence increased from 6% to 15% per year from 2002 to 2010 (CDC biennial data).

In addition to the challenges faced among families with ASD, there are difficulties that society itself faces related to the continuous growth in demand for consultations, which is much higher than the maximum capacity of pediatric clinics in the country (Thabtah, 2019). In addition to the adversities related to the expenses of treatment and education of the carriers. In the US the cost of lifelong services for Autism is as much as $2.4 million for a person with an intellectual disability and $1.4 million for a person without (Bloomberg, 2014). 

The consequences of the characteristics on the individual's development vary according to severity. According to the Mayo Clinic, these differences can include problems getting a job. The Chicago Tribune shows that 35% of young adults (19-23 years old) with Autism could not get a job or receive higher education after leaving high school (2012 data). Another consequence that those with ASD may face is social isolation and dependence on others to perform their activities, which increases the possibility of mental disorders in the future. More than 15% of young people with ASD think about or attempt suicide during adolescence, making them 30 times more at risk than typically developing children (Downs, 2018). 

About the expenses related to treatment, the cost in the U.S. is approximately $236 to $262 billion annually And most of this spending related to autism is on adult services (Bloomberg, 2014).

Therefore, the identification of Autistic Spectrum Disorder behaviors by artificial intelligence would be facilitated and make diagnosis more effective. Thus, the research is based on observing data from autistic people and their behavioral patterns through Machine Learning so that it is possible to recognize people with this syndrome in a faster and more accessible way to the population.

Based on the problems pointed out, the identification of behaviors of Autistic Spectrum Disorder through artificial intelligence would be facilitated and would make the diagnosis more effective. Thus, the research focuses on observing data from autistic people and their behavioral patterns through Machine Learning so that it is possible to recognize people with this disorder in a faster and more accessible form to the population. Thus, the research hypothesizes that Machine Learning methods would not only help evaluate the risk for Autism quickly and accurately but would also be essential to simplify the entire diagnostic process, making it less costly and reducing the stress patients and families deal with during too many clinical consultations. 

A Machine Learning algorithm was developed in Python programming to construct the model, using the scikit-learn library for statistical calculations, NumPy for vector operations, and pandas for data processing. Regarding the application, the flask library with the API feature created the back end. The life cycle methodology for building the Machine Learning model was CRISP-DM, divided into business understanding, data understanding, data preparation, modeling, model evaluation, and deployment.

The front-end screens were made on the Figma platform, with the main windows: profile tests, features, results, and community. In profile testing, the autistic patterns and behaviors question form will be available. The answers to the likelihood of having Autism will appear in the results window. The resources will have support materials about the disorder, and the community will be a place for diagnosed individuals, specialists, researchers, and their family members to talk.

Who does your solution serve?

Those families who have children and suspect they may have ASD and want to check this possibility, and those adolescents/adults who suspect ASD. The application is for people of all socio-economic levels but will be mainly for those who cannot afford a diagnosis from a clinic.

 In what ways will the solution impact their lives?

It intends to increase the effectiveness of diagnosis, especially in children. It is also proposed with this research to democratize access to autism diagnosis through the "Autinosis" application so that people of various ages can make an adequate diagnosis. 

Describe the target population whose lives you are working to directly and meaningfully improve. Who are they, and in what ways are they currently underserved?

Clinics that perform autism diagnosis are increasingly crowded with the demand increases, the high prices demanded, and a diagnosis that can be flawed because it is often given by only one health professional.

Matheus started his career as a software developer scientist. And today, he works as a Machine Learning engineer. He has advanced proficiency in the Python programming language and experience working with tools for data processing and statistical modeling using structured and unstructured data in a parallel or distributed fashion. In his job, Matheus develops resources that support companies to create and operate Machine Learning solutions. He constructed a Facial Recognition System: a biometric system with registration, verification, and identification steps. This project aimed to provide a way to perform facial recognition independent of the device used. It was developed over a neural network architecture (FACENET) considered state of the art at the time of development. He worked with both the planning and development phases, choosing the data set and the metrics the system would be evaluated, considering the National Institute of Standards and Technology (NIST) standards, and working on improvements such as Anti Spoofing techniques using RGB-D data and Data Cleaning phase at the registration stage. All code was written in Python, using libraries such as Tensorflow, Numpy, and Realsense sensors.

Fernanda is a mechatronics technician at the Federal Institute Sul Rio-grandense. She has been studying the intersection of mechanics, electronics, and computer science since 2018. During high school, Fernanda became the developer of SmartLeg, an active prosthesis project for transfemoral amputation, research for which she received a scientific initiation scholarship. This kind of prosthesis simulates the human knee providing greater naturalness for the user's walking process. However, it has a high cost for most people worldwide. Therefore, SmartLeg aims to simulate the human gait — adapting to the user's biotype — at a low price. Since her involvement in this project, the student has defined it as a goal to continue creating accessible medical innovations — especially projects related to neuroscience, her prospective major in university. Therefore, Fernanda is joyful about being part of Autinosis because it joins her passion for neuroscience with her goal of creating affordable medical inventions.

Millena gave the idea and led the group that stood out among 1200 projects from all over Brazil in the Liga Jovem Challenge, with the Olimpigram app. In addition, she has contributed more than 4 app ideas to various programming competitions. She was selected to be one of the 8 national ambassadors of the Technovation Girls Global Challenge. She was the 1st Brazilian woman to receive a 100% scholarship for an Extension Program at the University of Toronto (2021), as well as a program for girls in physics and programming at Stanford (2022). She also founded and coordinates one of the largest science Olympiad NGOs in Brazil, which has already reached +10k subscribers on social media.

Firstly, the team contacted an Autistic person who told her experience with the formal diagnosis. She talked about the expensive, lengthy, and tedious process of diagnosis and how implementing a Machine Learning algorithm could facilitate it. Then, we started the bibliographic research to understand more precisely the traditional diagnostic process and its limitations. After that, we also researched similar ASD projects using Machine Learning as a reference.