Silver D L, Hurwitz G A
Department of Nuclear Medicine, Victoria Hospital, Canada.
J Investig Med. 1997 Feb;45(2):99-108.
This paper compares two machine learning systems, an inductive decision tree (IDT) and a back-propagation neural network (ANN), in the noninvasive assessment of coronary artery disease given a set of diagnostic input attributes. A collection of 490 patient cases were accumulated from the reference of diagnostic stress myocardial scintigraphy performed in a nuclear medicine department. All cases had correlating angiography, the results of which were used to derive the target diagnoses. Input attributes included 4 baseline clinical characteristics, 4 nonimaging stress components, and 3 scintigraphic findings.
We chose 4 possible angiographic criteria for coronary artery disease and assessed the ability of each learning system to develop a diagnostic model. The 2 machine learning systems were compared on the basis of predictive performance and explanatory power.
Cross-validation experiments showed the 2 machine learning systems to have equivalent predictive power at the same level as the clinical scan reading. For the 70% stenosis criterion, the IDT had a sensitivity of 94 +/- 3% (mean +/- 95% confidence interval) and a specificity of 59 +/- 8%, and the ANN had a sensitivity of 97 +/- 2% and a specificity of 51 +/- 13%. However the IDT system exhibited excellent explanatory power; producing simple representations of the diagnostic models which agree with previous research.
In comparison with the more widely used ANNs, the IDT learning system may bring advantages to certain problems in diagnostic classification.
本文比较了两种机器学习系统，即归纳决策树（IDT）和反向传播神经网络（ANN），在给定一组诊断输入属性的情况下对冠状动脉疾病进行无创评估。从核医学科进行的诊断性应激心肌闪烁显像参考资料中积累了490例患者病例。所有病例均有相关的血管造影，其结果用于得出目标诊断。输入属性包括4项基线临床特征、4项非成像应激成分和3项闪烁显像结果。
我们选择了4种可能的冠状动脉疾病血管造影标准，并评估了每个学习系统建立诊断模型的能力。基于预测性能和解释力对这两种机器学习系统进行了比较。
交叉验证实验表明，这两种机器学习系统在与临床扫描读数相同的水平上具有同等的预测能力。对于70%狭窄标准，IDT的敏感性为94±3%（均值±95%置信区间），特异性为59±8%，ANN的敏感性为97±2%，特异性为51±13%。然而，IDT系统表现出出色的解释力；生成的诊断模型简单表示与先前研究一致。
与使用更广泛的ANN相比，IDT学习系统可能在诊断分类的某些问题上带来优势。
