ASD Detection from Facial Images – Dataset, Code and Results

This repository contains all materials associated with the study on Autism Spectrum Disorder (ASD) detection using facial image analysis and deep transfer learning techniques, as presented in our manuscript submitted to Scientific Reports (2025). 📁 Contents The repository is structured as follows: 1. Dataset (original Autism by Face Images from Piosenka Dataset) Folder: ASD Data/ This folder contains the facial image dataset used in our experiments, including training, validation, and test sets. 2. Code File: ASD Detection (Scientific Reports 2025).ipynb This Jupyter Notebook contains all the code used in our experiments, including: Image preprocessing Feature extraction using pre-trained networks Dimensionality reduction Classification model training and evaluation 3. Extracted Features Folder: result/autismo/face/ Contains feature vectors extracted from facial images using different pre-trained deep learning models. File naming convention: face_<network_code>_<benchmark_subset>.csv <network_code>:swin, vit, affectnet, alexnet, convnext, resnet50, vgg16, vgg19, vitfer <benchmark_subset>:train, valid, or test 4. Model Evaluation Reports Folder: reports/autismo/face/ Contains CSV files with classification results for each network configuration. File naming convention: <network_code>_<file_save_date>.csv Example: vit_2025-03-25.csv Each file includes metrics such as Accuracy, F1-Score, Recall, Precision, AUC, among others. 📌 Notes All scripts are organized for reproducibility. The proposed framework allows a modular configuration, enabling individual analysis of preprocessing techniques and classification stages. 💡 Citation If you use this repository in your research, please cite our work accordingly (citation details will be provided upon acceptance of the manuscript).