Video-Audio Neural Network Ensemble For Comprehensive Screening Of Autism Spectrum Disorder in Young Children (Openpose ADOS Dataset)

Here, we share a de-identify subsample of the data used in the preprint, that will allow interested scientists to test the shared code, as well as, develop alternatives for achieving better prediction accuracy. We have prepared a subsample of pose estimation videos for the first 10 minutes of ADOS examination videos for each of the 160 children including in the current study (80 ASD and 80 TD, 80 Training set and 80 Testing set). With this subset of the full dataset, our trained model achieved an accuracy of 68.75% over 80 videos (40 ASD & 40 TD) by training the Visual Geometry Group 16 Long short term memory recurrent neural network (VGG16 LSTM RNN) over 80 training videos (40 ASD & 40 TD).本研究共享预印本所用数据的去标识化子样本，以供感兴趣的科研人员测试共享代码，同时亦可用于开发可实现更优预测精度的替代方案。 本研究纳入160名儿童（其中80名孤独症谱系障碍（Autism Spectrum Disorder, ASD）患儿、80名典型发育（Typical Development, TD）儿童，数据集划分为训练集与测试集各80份），我们为每名儿童的自闭症诊断观察量表（Autism Diagnostic Observation Schedule, ADOS）检查视频的前10分钟片段，制备了姿态估计视频子样本。 针对该完整数据集的子样本，我们通过在80份训练视频（40名ASD患儿、40名TD儿童）上训练视觉几何组16长短期记忆循环神经网络（VGG16 LSTM RNN），最终在80份测试视频（40名ASD患儿、40名TD儿童）上取得了68.75%的预测精度。