SwarmDeepSurv: swarm intelligence advances deep survival network for prognostic radiomics signatures in four solid cancers.
SwarmDeepSurv: swarm intelligence advances deep survival network for prognostic radiomics signatures in four solid cancers.
复制标题
DOI：
10.1016/j.patter.2023.100777
复制
发表时间：
2023-08-11
期刊：
影响因子：
6.5
通讯作者：
Wu, Jia
中科院分区：
文献类型：
--
作者：
Al-Tashi, Qasem;Saad, Maliazurina B.;Sheshadri, Ajay;Wu, Carol C.;Chang, Joe Y.;Al-Lazikani, Bissan;Gibbons, Christopher;Vokes, Natalie I.;Zhang, Jianjun;Lee, J. Jack;Heymach, John, V;Jaffray, David;Mirjalili, Seyedali;Wu, Jia
生存模型的存在是为了研究生物标志物和治疗效果之间的关系。深度学习驱动的生存模型取代了经典的 Cox 比例风险 (CoxPH) 模型，但由于不相关/冗余信息，在高维特征上观察到性能大幅下降。为了填补这一空白，我们通过将群体智能算法与深度生存模型相结合，提出了 SwarmDeepSurv。此外，设计了四个目标函数来优化预后预测，同时规范所选的特征数。当对四种不同癌症类型的多中心集（n = 1,058）进行测试时，与流行的生存建模算法相比，SwarmDeepSurv 不太容易过度拟合，并实现了最佳的患者风险分层。引人注目的是，与经典特征选择算法相比，SwarmDeepSurv 选择了不同的特征，包括最小绝对收缩和选择算子 (LASSO)，并且这些模型之间几乎没有特征重叠。总而言之，SwarmDeepSurv 提供了一种替代方法来建模放射组学特征和生存终点之间的关系，该方法可以进一步扩展到研究包括基因组学在内的其他输入数据类型。 SwarmDeepSurv 将群体智能与深度生存模型融合在一起 提出了四种成本函数来平衡模型性能和过度拟合风险 对四种癌症类型的多中心组（n = 1,058）进行了评估 SwarmDeepSurv 通过改进分层，优于当前的生存模型 预后模型通过对某些生物标志物（包括放射组学特征和治疗效果）之间的关系进行建模，对于癌症患者的战略干预至关重要。在这里，我们将群体智能与深度生存建模相结合，创建了一个名为 SwarmDeepSurv 的新平台，以从高维特征中发现预后模型。与现状相比，SwarmDeepSurv 展示了更稳健的患者分层和对生存结果的准确预测，这有可能为个性化治疗提供信息。特别是，SwarmDeepSurv 选择一组不同的放射组学特征来构建预后模型。从模型的可解释性来看，除了当前的预后生物标志物模型之外，我们的管道还提供了产生新见解和假设的机会。我们预计该平台通常不仅适用于放射组学特征，还可以研究其他临床应用的丰富临床基因组学特征。现有基于高维特征的深度学习生存模型的一个主要弱点是，由于维数灾难，其性能大幅下降。为了填补这一空白，SwarmDeepSurv 被提出，将群体智能算法与深度生存模型相结合，可以对患者的生存结果进行稳健的分层。此外，它权衡了现有算法未选择的特征，从而导致独特的模型可解释性和假设生成。
Survival models exist to study relationships between biomarkers and treatment effects. Deep learning-powered survival models supersede the classical Cox proportional hazards (CoxPH) model, but substantial performance drops were observed on high-dimensional features because of irrelevant/redundant information. To fill this gap, we proposed SwarmDeepSurv by integrating swarm intelligence algorithms with the deep survival model. Furthermore, four objective functions were designed to optimize prognostic prediction while regularizing selected feature numbers. When testing on multicenter sets (n = 1,058) of four different cancer types, SwarmDeepSurv was less prone to overfitting and achieved optimal patient risk stratification compared with popular survival modeling algorithms. Strikingly, SwarmDeepSurv selected different features compared with classical feature selection algorithms, including the least absolute shrinkage and selection operator (LASSO), with nearly no feature overlapping across these models. Taken together, SwarmDeepSurv offers an alternative approach to model relationships between radiomics features and survival endpoints, which can further extend to study other input data types including genomics. SwarmDeepSurv fuses swarm intelligence with deep survival model Four cost functions proposed to balance model performance and overfitting risk Evaluation was carried out on a multicenter set (n = 1,058) of four cancer types SwarmDeepSurv outperformed current survival models with improved stratification Prognostic models are crucial for strategic intervention in cancer patients, by modeling the relationship between certain biomarkers including radiomics features and treatment effects. Here, we integrate swarm intelligence with deep survival modeling to create a new platform termed SwarmDeepSurv to discover prognostic models from high-dimensional features. Compared with the status quo, SwarmDeepSurv demonstrates a more robust stratification of patients and accurate prediction for survival outcomes, which has the potential to inform personalized treatment. Particularly, SwarmDeepSurv selects a different set of radiomics features to build the prognostic models. From the view of the model’s interpretability, our pipeline offers an opportunity to generate new insights and hypotheses besides current prognostic biomarker models. We anticipate that this platform will generally be applicable beyond radiomics features to also study rich clinicogenomics features for other clinical applications. One major weakness of existing deep learning survival models that are built on high-dimensional features is its observed substantial performance drops because of the curse of dimensionality. To fill this gap, SwarmDeepSurv is proposed by integrating swarm intelligence algorithms with the deep survival models, which can robustly stratify patients’ survival outcomes. Furthermore, it weighs on the features that are not selected by existing algorithms, leading to distinctive model interpretability and hypothesis generation.
登录
查看更多内容
影响因子：
10
作者：
Bach Hoai Nguyen;Xue, Bing;Zhang, Mengjie
通讯作者：
Zhang, Mengjie
影响因子：
3.9
作者：
Al-Wajih, Ranya;Abdulkadir, Said Jadid;Talpur, Noureen
通讯作者：
Talpur, Noureen
DOI：
10.1038/s41568-021-00408-3
发表时间：
2022-03
期刊：
Nature reviews. Cancer
影响因子：
--
作者：
Boehm KM;Khosravi P;Vanguri R;Gao J;Shah SP
通讯作者：
Shah SP
影响因子：
3.9
作者：
Al-Tashi, Qasem;Abdulkadir, Said Jadid;Alqushaibi, Alawi
通讯作者：
Alqushaibi, Alawi
一种与极值优化相结合的新型粒子群优化器
影响因子：
8.7
作者：
Chen, Min-Rong;Li, Xia;Lu, Yong-Zai
通讯作者：
Lu, Yong-Zai