基于贝叶斯分析的癌症诊断中临床变量因果关系挖掘

67 浏览量更新于2024-07-15 收藏 1.81MB PDF 举报

本文是一篇发表在《BioDataMining》（2015年，第8卷，DOI:10.1186/s13040-015-0046-4）的研究论文，主要探讨了利用贝叶斯分析挖掘癌症诊断中临床变量之间的因果关系。癌症作为全球第二大死因，其预后预测是医学研究的重要课题。传统数据挖掘方法虽然在癌症结果预测上取得了一定进展，但关于临床变量之间因果关系的探讨相对较少，这在理论指导癌症诊断和治疗方面具有重要意义。研究者李明（Li Min Wang）及其团队提出了一个新颖的贝叶斯学习策略，目标是预测特定原因导致的死亡类别。他们注意到，许多领域已经应用了受限的贝叶斯分类器来发现有价值的信息。然而，本文的重点在于设计一种能够揭示这些临床变量内在联系的图形结构，即通过构建因果图谱，以便清晰地展示各变量之间的潜在因果关系。这种方法旨在通过贝叶斯网络这一统计模型，结合临床数据，分析各变量对疾病发生、发展和预后的潜在影响。通过计算各变量之间的条件概率和依赖性，可以更深入地理解变量间的相互作用机制，从而提升癌症诊断的准确性和个性化治疗的针对性。具体来说，论文可能涉及以下几个关键步骤： 1. 数据收集与预处理：收集包含多种临床特征的数据集，如年龄、性别、肿瘤类型、病程阶段等。 2. 构建贝叶斯网络：运用贝叶斯定理，根据先验知识和观察数据来估计各变量之间的概率关系。 3. 网络结构学习：通过算法（如最大似然估计或贝叶斯分数法）确定变量间的因果路径和强度。 4. 预测与评估：使用构建的网络进行疾病分类，同时分析其在预测疾病结局方面的性能，如敏感度、特异度和AUC值。 5. 结果解释与讨论：基于贝叶斯网络的推断，解释哪些变量对疾病进程影响最大，以及这些因果关系如何有助于优化临床决策。这篇研究论文通过引入贝叶斯分析来探索临床变量之间的因果关系，为癌症诊断提供了一种更为深入且可能具有实用价值的方法。它不仅提高了我们理解疾病机制的能力，而且可能在实际临床实践中推动个性化医疗的发展。

Wang BioData Mining

(2015) 8:13

Page 3 of 15

To minimize user effort, Vukicevic et al. [6] applied genetic algorithms to achieve the

best prognostic performances relevant for clinicians (i.e., correctness, discrimination and

calibration). The only 2 use r dependent tasks were data selection (input and output vari-

ables) and the evaluation of the ANN threshold probability with respect to regret theory

(RT). After optimally configuring ANNs with respect to these criteria, the clinical use-

fulness was evaluated by the RT Decision Curve Analysis. Tsao e t al. [7] develop ed an

ANN model to predict prostate cancer pathological staging in patients prior to receiv-

ing radic al prostatectomy. This exper imental study examined the cases of 299 patients

undergoing retro-pubic radical prostatectomy. In this investigation, the validation was

assessed by using the current Partin Tables for the Taiwanese population. ANN induced

larger AUCs and provided a more accurate pre diction of the pathologic stage of prost ate

cancer.

Bayesian networks (BNs) are characterised by the use of the probabilistic approach in

solving problems and encompass the uncertainty of specific occurrences. The origin of

BNs is ba sed on probability distribution, which can be graphically depicted. Alexander et

al. [8] applied the SEER database (1969 to 2006) to form a clinical dec ision support system

for the real-time estimation of the overall survival (OS) rate of colon cancer patients. The

BN model accurately estimated OS with an area under the receiver-operating character-

istic curve of 0.85. They significantly improved upon the existence of AJCC stage-specific

OS estimates. Furthermore, they determined the significant differences in OS b etween

low- and high-risk cohorts. Khan et al. [9] used Bayesian method to derive the poste-

rior density function for the parameters and the predictive inference for future survival

times from the exponentiated Weibull model, assuming that the observed breast can-

cer survival data follow such type of model. The Markov chain Monte Carlo method

was used to determine the inference for the parameters. They found that the exponen-

tiated Weibull model fits the male survival data. Mean predictive survival times, 95%

predictive intervals, predictive skewness and kurtosis were obtained. Jong et al. [10]

introduced a hybrid model that combined ANN and BN to obtain a good estimation

of prognosis and a good explanation of the results. In this research, the SEER database

(1973 to 2003) was employed to construct and evaluate the proposed models. Nine

clinically acceptable var iables were selected to be incorporated into the nodes of the pro-

posed models. Consequently, the hybrid model achieved the highest area under the curve

value of 0.935, and the corresponding values of ANN and BN were 0.930 and 0.813,

respectively.

Other machine learning models have also been applied to solve the problems in pre-

dicting cancer survivability. Molina e t al. [11] suggested that an incremental learning

ensemble of a suppor t vector machine (SVM) must be implemented to adapt to the work-

ing conditions in medical applications and to improve the effectiveness and robustness

of the system. These studies calculated the probability estimation of cancer structures

by using SVM and performed the corresponding optimisation with a heuristic method

together w ith a three-fold cross-validation methodology. Mahmoodian et al. [12] de vel-

ope d a new algorithm on the basis of fuzzy association rule mining to identify fuzzy

rules and significant genes. In this study, different sub sets of gene s that have been

selected by different me thods were used to separately generate primary fu zzy classifiers.

Subsequently, the researchers administered their proposed algorithm to mix the genes

associated with the primary classifiers and to generate a novel classifier.

剩余14页未读，继续阅读

weixin_38551837

粉丝: 4
资源: 922

基于贝叶斯分析的癌症诊断中临床变量因果关系挖掘

Big Data in Omics and Imaging_Association Analysis-CRC (2018).pdf

causal analysis_impact evaluation and machine learning with R.

Bayesian inference for causal effects

Failed to build installable wheels for some pyproject.toml based projects (causal_conv1d)

We additionally performed genome-wide MR analyses using Causal Analysis Using Summary Effect estimates (CAUSE)，孟德尔随机化安装CAUSE包

ERROR: Could not build wheels for causalml, which is required to install pyproject.toml-based projects

All of Statistics

a causal inference based approach

causal inference for statistics social and biomedical sciences 下载

Could not build wheels for causal-conv1d, which is required to install pyproject.toml-based projects

最新资源