摘要
机器学习算法在医学肿瘤领域应用广泛,文章针对肺腺癌磨玻璃结节患者,运用Logistic回归、人工神经网络、支持向量机和Ada Boost等4种经典算法,从准确率、灵敏度、特异度、F1值和G值等5个指标进行模型性能的分析和比较,发现4种模型的准确率均高于85%,而临床医师对患者诊断准确率只有78%,并总结出不同模型的特点:BP神经网络的精度最高,具有较强的自我学习能力;支持向量机需要调节参数来优化模型,对小样本训练效果较好;Logistic回归不需要调节参数,但在大样本下训练比较耗时;Ada Boost具有组合多个模型后再做诊断的优良品质,为临床医师提供术前诊断方案.
Machine learning algorithms are widely used in the field of medical oncology. In this paper, the four classic algorithms, including logistic regression, artificial neural network, support vector machine and AdaBoost methods, are applied to the patients with pulmonary adenocarcinoma ground nodules. The comparison and analysis of model performance, on accuracy, sensitivity, specificity, F1-score and G-score parts, shows that the accuracy of the four models is more than 85% while the clinician is only 78% in diagnosis. We obtain the characteristics of different models： the back propagation neural network has the highest accuracy and has strong self-learning ability, support vector machines need to adjust the parameters to optimize the model and the training effect from SVM is better for small samples. Logistic regression does not require parameters adjustment, but it is time-consuming to train under large samples. AdaBoost has the good quality of combining multiple models before making a diagnosis, which provides the clinician with a preoperative diagnosis plan.
作者
吴艇帆
张仁寿
WU Ting-fan;ZHANG Ren-shou(School of Economics and Statistics,Guangzhou University,Guangzhou 510006,China)
出处
《广州大学学报（自然科学版）》
CAS
2018年第3期33-39,共7页
Journal of Guangzhou University:Natural Science Edition
关键词
机器学习
肺腺癌
性能比较
临床医学
machine learning
lung adenocarcinoma
performance comparison
clinical medicine
