Scaling behaviours of deep learning and linear algorithms for the prediction of stroke severity.
Scaling behaviours of deep learning and linear algorithms for the prediction of stroke severity.
复制标题
DOI：
10.1093/braincomms/fcae007
复制
发表时间：
2024
影响因子：
4.8
通讯作者：
中科院分区：
文献类型：
--
作者：
深度学习在许多医疗场景中取得了显着进展。深度学习预测模型通常需要105-107个示例。目前尚不清楚深度学习是否也可以增强对中风患者真实样本中中风后症状的预测，这些样本通常要小几个数量级。然而，这种中风结果预测在指导急性临床和康复护理决策方面可能特别有用。我们在这里比较了经典使用的线性和新型深度学习算法在预测中风严重程度方面的能力。我们的分析依赖于总共1430例患者，这些患者来自MRI-遗传学界面探索合作和马萨诸塞州总医院的研究。研究结果是基于美国国立卫生研究院卒中量表的缺血性卒中发作后急性期卒中严重程度，我们通过MRI得出的病变位置进行预测。我们从弥散加权临床MRI扫描中自动获得病变分割，进行空间归一化并包括主成分分析步骤，保留原始数据95%的方差。然后，我们反复分离训练集、验证集和测试集，以研究样本量的影响;我们将训练集子采样为100、300和900，并训练算法，以利用正则化线性回归和八层神经网络预测每个样本量的中风严重程度评分。我们在验证集上选择了超参数。我们根据测试集中的解释方差（R2）评估模型性能。虽然线性回归在100名患者的样本量下表现得更好，但当对900名患者进行训练时，深度学习开始明显优于线性回归。当样本量增加9倍时，平均预测性能提高了20%[100例患者的最大值：0.279 ± 0.005（R2，95%置信区间），900例患者：0.337 ± 0.006]。总之，对于900名患者的样本量，深度学习显示出比通常采用的线性方法更高的预测性能。这些发现表明，病变部位和卒中严重程度之间存在非线性关系，可用于更大样本量的改善预测性能。Bourached等人根据训练集样本大小对比了基于线性和深度学习的算法对中风严重程度的预测性能。他们发现，对于包含100名患者病变位置信息的较小训练样本，线性回归的性能优于基于深度学习的算法，而深度学习在较大样本（N = 900）的情况下表现出色。
Deep learning has allowed for remarkable progress in many medical scenarios. Deep learning prediction models often require 105–107 examples. It is currently unknown whether deep learning can also enhance predictions of symptoms post-stroke in real-world samples of stroke patients that are often several magnitudes smaller. Such stroke outcome predictions however could be particularly instrumental in guiding acute clinical and rehabilitation care decisions. We here compared the capacities of classically used linear and novel deep learning algorithms in their prediction of stroke severity. Our analyses relied on a total of 1430 patients assembled from the MRI-Genetics Interface Exploration collaboration and a Massachusetts General Hospital–based study. The outcome of interest was National Institutes of Health Stroke Scale–based stroke severity in the acute phase after ischaemic stroke onset, which we predict by means of MRI-derived lesion location. We automatically derived lesion segmentations from diffusion-weighted clinical MRI scans, performed spatial normalization and included a principal component analysis step, retaining 95% of the variance of the original data. We then repeatedly separated a train, validation and test set to investigate the effects of sample size; we subsampled the train set to 100, 300 and 900 and trained the algorithms to predict the stroke severity score for each sample size with regularized linear regression and an eight-layered neural network. We selected hyperparameters on the validation set. We evaluated model performance based on the explained variance (R2) in the test set. While linear regression performed significantly better for a sample size of 100 patients, deep learning started to significantly outperform linear regression when trained on 900 patients. Average prediction performance improved by ∼20% when increasing the sample size 9× [maximum for 100 patients: 0.279 ± 0.005 (R2, 95% confidence interval), 900 patients: 0.337 ± 0.006]. In summary, for sample sizes of 900 patients, deep learning showed a higher prediction performance than typically employed linear methods. These findings suggest the existence of non-linear relationships between lesion location and stroke severity that can be utilized for an improved prediction performance for larger sample sizes. Bourached et al. contrast linear and deep learning–based algorithms in their prediction performances of stroke severity depending on the training set sample sizes. They find that linear regression outperforms deep learning–based algorithms for smaller training samples comprising lesion location information of 100 patients, while deep learning excels in the case of larger samples (N = 900).
影响因子：
4.3
作者：
Bonkhoff, Anna K.;Ullberg, Teresa;Bretzner, Martin;Hong, Sungmin;Schirmer, Markus D.;Regenhardt, Robert W.;Donahue, Kathleen L.;Nardin, Marco J.;Dalca, Adrian, V;Giese, Anne-Katrin;Etherton, Mark R.;Hancock, Brandon L.;Mocking, Steven J. T.;McIntosh, Elissa C.;Attia, John;Cole, John W.;Donatti, Amanda;Griessenauer, Christoph J.;Heitsch, Laura;Holmegaard, Lukas;Jood, Katarina;Jimenez-Conde, Jordi;Kittner, Steven J.;Lemmens, Robin;Levi, Christopher R.;McDonough, Caitrin W.;Meschia, James F.;Phuah, Chia-Ling;Ropele, Stefan;Rosand, Jonathan;Roquer, Jaume;Rundek, Tatjana;Sacco, Ralph L.;Schmidt, Reinhold;Sharma, Pankaj;Slowik, Agnieszka;Sousa, Alessandro;Stanne, Tara M.;Strbian, Daniel;Tatlisumak, Turgut;Thijs, Vincent;Vagal, Achala;Woo, Daniel;Zand, Ramin;McArdle, Patrick F.;Worrall, Bradford B.;Jern, Christina;Lindgren, Arne G.;Maguire, Jane;Wu, Ona;Frid, Petrea;Rost, Natalia S.;Wasselius, Johan
通讯作者：
Wasselius, Johan
DOI：
10.1002/ail2.63
发表时间：
2022-04
期刊：
Applied AI letters
影响因子：
--
作者：
Bourached, Anthony;Griffiths, Ryan-Rhys;Gray, Robert;Jha, Ashwani;Nachev, Parashkev
通讯作者：
Nachev, Parashkev
DOI：
10.1177/1747493020984552
发表时间：
2021-10
期刊：
International journal of stroke : official journal of the International Stroke Society
影响因子：
--
作者：
Weaver NA;Kancheva AK;Lim JS;Biesbroek JM;Wajer IMH;Kang Y;Kim BJ;Kuijf HJ;Lee BC;Lee KJ;Yu KH;Biessels GJ;Bae HJ
通讯作者：
Bae HJ