ABCD神经认知预测挑战2019:基于概率分割与核岭回归从结构MRI预测个体流体智力分数
神经元与认知
2019-05-28 v1 应用统计
摘要
作为2019年ABCD神经认知预测挑战(ABCD-NP-Challenge)的一部分,我们应用了若干回归与深度学习方法,从T1加权MRI扫描预测流体智力分数。我们使用体素强度及由此导出的概率组织类型标签作为特征训练模型。预测性能最佳(均方误差最低)的是核岭回归(KRR;λ=10),其在验证集上均方误差为69.7204,测试集上为92.1298。这使我们的小组在验证排行榜上位列第五,在最终(测试)排行榜上位列第一。
引用
@article{arxiv.1905.10831,
title = {ABCD Neurocognitive Prediction Challenge 2019: Predicting individual fluid intelligence scores from structural MRI using probabilistic segmentation and kernel ridge regression},
author = {Agoston Mihalik and Mikael Brudfors and Maria Robu and Fabio S. Ferreira and Hongxiang Lin and Anita Rau and Tong Wu and Stefano B. Blumberg and Baris Kanber and Maira Tariq and Maria Del Mar Estarellas Garcia and Cemre Zor and Daniil I. Nikitichev and Janaina Mourao-Miranda and Neil P. Oxtoby},
journal= {arXiv preprint arXiv:1905.10831},
year = {2019}
}
备注
Winning entry in the ABCD Neurocognitive Prediction Challenge at MICCAI 2019. 7 pages plus references, 3 figures, 1 table