Deep Neural Network-based Handheld Diagnosis System for Autism Spectrum Disorder
Vikas Khullar1, Harjit Pal Singh1, Manju Bala2
1I.K.G. Punjab Technical University, Kapurthala; CT Institute of Engineering, Management and Technology, Jalandhar, Punjab, India.
Neurology India
|March 1, 2021
Summary
A new deep neural network (DNN) handheld diagnosis system (HDS) accurately identifies autism spectrum disorder (ASD). The long short-term memory (LSTM) algorithm achieved 100% accuracy, offering a promising alternative to manual ASD diagnosis.
Area of Science:
- Medical Informatics
- Artificial Intelligence
- Neuroscience
Background:
- Autism Spectrum Disorder (ASD) diagnosis and severity assessment are complex and often rely on subjective clinical evaluations.
- There is a need for objective, accurate, and accessible diagnostic tools for ASD.
Purpose of the Study:
- To propose and implement a deep neural network (DNN)-based handheld diagnosis system (HDS) for accurate ASD diagnosis and severity assessment.
- To evaluate the performance of different DNN algorithms for ASD detection.
Main Methods:
- Implemented and compared deep neural network (DNN) algorithms including Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Multilayer Perceptron (MLP) using a dataset based on the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-V).
- Analyzed algorithm performance using accuracy, loss, Mean Squared Error (MSE), precision, recall, and Area Under the Curve (AUC).
- Integrated the optimal DNN algorithm into a handheld diagnosis system (HDS) and validated its stability with a cohort of 20 individuals with ASD and 20 typically developed (TD) individuals.
Main Results:
- Long Short-Term Memory (LSTM) demonstrated superior performance in ASD diagnosis compared to CNN and MLP, exhibiting stabilized results, maximum accuracy, minimal Mean Squared Error (MSE), and reduced loss.
- The LSTM-based HDS achieved 100% accuracy for ASD diagnosis, validated statistically against DSM-V criteria.
Conclusions:
- Advanced artificial intelligence (AI) algorithms, particularly LSTM, are crucial for modern ASD diagnosis.
- The proposed LSTM-based HDS offers a highly accurate and potentially superior alternative to manual methods for diagnosing ASD and determining its severity, aligning with DSM-V criteria.
