[Paper Review] Screening Autism Spectrum Disorder in childrens using Deep Learning Approach : Evaluating the classification model of YOLOv8 by comparing with other models

Subash Gautam, Prabin Sharma+5 more|arXiv (Cornell University)|Jun 25, 2023

Autism Spectrum Disorder Research4 citations

TL;DR

This study proposes a deep learning-based screening method for Autism Spectrum Disorder (ASD) in children using facial images, leveraging the YOLOv8 object detection model adapted for binary classification. Trained on a Kaggle dataset, the model achieved 89.64% accuracy and an F1-score of 0.89, demonstrating strong potential for automated ASD detection using facial feature analysis.

ABSTRACT

Autism spectrum disorder (ASD) is a developmental condition that presents significant challenges in social interaction, communication, and behavior. Early intervention plays a pivotal role in enhancing cognitive abilities and reducing autistic symptoms in children with ASD. Numerous clinical studies have highlighted distinctive facial characteristics that distinguish ASD children from typically developing (TD) children. In this study, we propose a practical solution for ASD screening using facial images using YoloV8 model. By employing YoloV8, a deep learning technique, on a dataset of Kaggle, we achieved exceptional results. Our model achieved a remarkable 89.64% accuracy in classification and an F1-score of 0.89. Our findings provide support for the clinical observations regarding facial feature discrepancies between children with ASD. The high F1-score obtained demonstrates the potential of deep learning models in screening children with ASD. We conclude that the newest version of YoloV8 which is usually used for object detection can be used for classification problem of Austistic and Non-autistic images.

Motivation & Objective

• To develop a practical, automated screening system for Autism Spectrum Disorder (ASD) in children using facial image analysis.
• To evaluate the performance of YOLOv8—originally designed for object detection—when adapted for binary classification of ASD versus typically developing (TD) children.
• To compare YOLOv8’s classification performance against other deep learning models using a public facial image dataset.
• To validate clinical observations of distinct facial features in children with ASD through machine learning.
• To demonstrate the feasibility of using deep learning for early, non-invasive ASD screening in clinical and community settings.

Proposed method

• The study employs the YOLOv8 model, fine-tuned for binary image classification of ASD and TD children, using transfer learning.
• A dataset of facial images from children with ASD and typically developing children was sourced from Kaggle.
• The model was trained and evaluated using standard deep learning metrics, including accuracy, precision, recall, and F1-score.
• Data augmentation techniques were applied to improve model generalization and reduce overfitting.
• Performance was benchmarked against other deep learning models to assess comparative effectiveness.
• The YOLOv8 architecture was repurposed for classification by modifying the final classification head while retaining its feature extraction capabilities.

Experimental results

Research questions

• RQ1 Can YOLOv8, originally designed for object detection, be effectively adapted for binary classification of ASD and TD children from facial images?
• RQ2 How does the classification performance of YOLOv8 compare to other deep learning models in detecting ASD from facial features?
• RQ3 To what extent do the model’s predictions align with clinical observations of facial feature differences in ASD?
• RQ4 What is the optimal balance of precision and recall in ASD screening using deep learning on facial images?
• RQ5 Can a lightweight, fast inference model like YOLOv8 support scalable, real-time ASD screening in low-resource settings?

Key findings

• The YOLOv8-based model achieved a classification accuracy of 89.64% on the test set, indicating strong predictive performance.
• The model recorded an F1-score of 0.89, reflecting a balanced trade-off between precision and recall in detecting ASD cases.
• The results support clinical evidence of distinct facial characteristics in children with ASD compared to typically developing children.
• YOLOv8 outperformed other compared models in terms of both accuracy and F1-score, demonstrating its suitability for this classification task.
• The high F1-score suggests the model is effective at minimizing both false positives and false negatives in ASD screening.
• The study confirms that YOLOv8 can be successfully repurposed for medical image classification tasks beyond object detection.