人工智能线性判别分析算法建立喉癌5年生存状态预测模型

Establishment of a 5-Year Survival Prediction Model for Laryngeal Cancer by Linear Discriminant Analysis

目的:采用人工智能线性判别分析(lineardiscriminantanalysis,LDA)算法建立判断喉癌5年生存状态的预测模型,为临床喉癌的诊治及预后判断提供参考。方法:回顾性分析我院124例喉癌患者生存预后数据,采用LDA算法,以患者的性别,年龄,是否抽烟、饮酒,TNM分期,肿瘤的临床分期,有无复发,有无放化疗及病理分级等临床参数作为特征,以60%数据为训练集,40%数据为测试集建立模型,以准确率、灵敏度、特异度、曲线下面积(areaunder thecurve,AUC)、F1值和Cohen’skappa系数对模型进行评估。此外,将LDA与支持向量机等其他机器学习算法的模型结果进行对比。结果:我们成功使用LDA算法建立了基于上述临床特征的喉癌5年生存状态预测模型,模型的准确率、灵敏度、特异度、AUC、F1值及Cohen’skappa系数分别为:0.86、0.86、0.85、0.86、0.86和0.66,与其他机器学习算法相比,LDA算法建立的模型性能最佳。结论:采用LDA算法可以建立可靠的喉癌5年生存状态的预测模型,本模型可为喉癌临床诊治提供一个新的预后评估手段。

Objective: To establish a model for predicting the 5-year survival of laryngeal cancer patients by using linear discriminant analysis(LDA), an artificial intelligence algorithm, so as to provide evidence for clinical diagnosis, treatment and prognosis of laryngeal cancer patients. Methods: Prognosis data of 124 laryngeal cancer patients in our hospital were analyzed retrospectively. 60% of the data were used as the training set and 40% of the data as the test set. LDA was used to establish the model with clinical parameters such as sex, age, tobacco and alcohol use, TNM stage, metastasis, recurrence, chemoradiotherapy and pathological grade were used as input features of the model. Sensitivity, specificity, area under the curve(AUC), F1 value and Cohen’s kappa coefficient were used to evaluated the model. In addition, the model results of LDA are compared with other machine learning algorithms such as support-vector machines. Results: The model for predicting the 5-year survival of laryngeal cancer patients was successfully established based on the above clinical characteristics by using LDA. The accuracy, sensitivity, specificity, AUC, F1 value and kappa coefficient of the model were 0.86, 0.86, 0.85, 0.86, 0.86 and 0.66, respectively. Compared with other machine learning algorithms, LDA was the best in performance. Conclusion: A reliable model for predicting the 5-year survival of laryngeal cancer patients can be established using LDA. This model can provide a new method for evaluating the clinical diagnosis and treatment of laryngeal cancer.