An Advanced Deep Learning Framework for Video-based Diagnosis of ASD

This repository contains the implementation for the paper: Miaomiao Cai, Mingxing Li, Zhiwei Xiong, Pengju Zhao, Enyao Li and Jiulai Tang, "An Advanced Deep Learning Framework for Video-based Diagnosis of ASD"

Dataset

To protect the rights and privacy of our research participants, please do not repost our dataset.

A total of 82 children (ASD: 57; TD: 25) finish the experiments with their parents at home, and they all consent to use their data for our research.

All diagnostic labels (ASD or TD) are provided by the ADOS-2, which have been done independently of our experiments.

We provide the face features extracted by the Openface 2.0 toolbox.

Framework

Prerequisites

• numpy 1.17.4
• torch 1.3.1
• torchvision 0.4.2

commands

Training stage

• In /code/main.py, please change rootpath.
• For 2 gpus:

python python /code/main.py --mynum 100 --numsegments 4 --model_name attention

Test stage

• In /code/test.py, please change rootpath.

python python /code/test.py --mynum 100 --numsegments 4 --model_name attention

Citing

If you use our code in your research or applications, please consider citing our paper.

@inproceedings{cai2022advanced,
  title={An Advanced Deep Learning Framework for Video-Based Diagnosis of ASD},
  author={Cai, Miaomiao and Li, Mingxing and Xiong, Zhiwei and Zhao, Pengju and Li, Enyao and Tang, Jiulai},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  pages={434--444},
  year={2022},
  organization={Springer}
}