A Novel Hybrid Vision Transformer–XGBoost Framework for Automated Autism Spectrum Disorder Detection Using Image-Based Data

Autism Spectrum Disorder, Deep Learning, Vision Transformer, DeiT, ImageBased Diagnosis, ViTASD, XGBoost

1Debmitra Ghosh
Computer Science and
Engineering Department, JIS
University, Kolkata, India

2Dr. Kaushik Adhikary
Computer Science and
Engineering Department, JIS
University, Kolkata, India

3Dr. Dharmpal Singh
Computer Science and
Engineering Department, JIS
University, Kolkata, India

4Dr. Sumit Das
Department of Information
Technology, JIS College of
Engineering, Nadia, India

5Kallol Acharjee
Department of Information
Technology, JIS College of
Engineering, Nadia, India

6Subhodip Koley
Department of Computer
Application, JIS College of
Engineering, Nadia, India

7Sudipta Sahana
Department of AI, UEM, Kolkata,
India

Received-12-06-2026
Revised-14-07-2026
Accepted-19-07-20.26

European Journal of Prosthodontics and Restorative Dentistry (2026) 34 (5s), 43-53

A Novel Hybrid Vision Transformer–XGBoost Framework for Automated Autism Spectrum Disorder Detection Using Image-Based Data

Our study focusses on a novel deep learning framework that integrates Vision Transformer-based Autism Spectrum Disorder detection using ViTASD and DeiT along with XGBoost to enhance classification performance using image-based data. ASD is a multifaceted disease and the early diagnosis has been a serious challenge in terms of disease detection. It is a neurodevelopmental disorder in which precise diagnosis is often hampered due to its varied characteristics and dependence on subjective evaluations. Conventional diagnostic methods, such as behavioral checklists and clinical assessments, are labor-intensive, necessitate expert participation, and may lack applicability across different demographic groups. The suggested hybrid model has been pre-trained on a ResNet50 architecture and utilizes the feature extraction strengths of gradient boosted tress (XGBoost) combined with the attention-focused accuracy of transformer architecture of ViTASD and DeiT to detect distinctive patterns that may be linked to ASD. This model has been trained and validated on different datasets, and its efficiency was compared to existing machine learning models. Our findings have exhibited a notable enhancement in accuracy, specificity, and overall robustness highlighting the model’s potential as a reliable, scalable, and serviceable screening instrument. This study also addresses some shortcomings of previous research by providing a new approach for future studies regarding ASD detection, with encouraging prospects for both clinical and remote diagnostic applications. Our study has shown significant results with an accuracy of 97% and good overall performance metrics which suggests a successful predictive analysis of ASD detection.

••••••••••••••••••••••••••••••• ejprd.org - Published by Riset Publication Services LLC

EJPRD

Copyright © 2026 by Riset Publication Services LLC