Chilakapati Lab, ACTREC, Navi Mumbai, India.
J Biomed Opt. 2011 Feb;16(2):025003. doi: 10.1117/1.3548303.
Cancer is now recognized as one of the major causes of morbidity and mortality. Histopathological diagnosis, the gold standard, is shown to be subjective, time consuming, prone to interobserver disagreement, and often fails to predict prognosis. Optical spectroscopic methods are being contemplated as adjuncts or alternatives to conventional cancer diagnostics. The most important aspect of these approaches is their objectivity, and multivariate statistical tools play a major role in realizing it. However, rigorous evaluation of the robustness of spectral models is a prerequisite. The utility of Raman spectroscopy in the diagnosis of cancers has been well established. Until now, the specificity and applicability of spectral models have been evaluated for specific cancer types. In this study, we have evaluated the utility of spectroscopic models representing normal and malignant tissues of the breast, cervix, colon, larynx, and oral cavity in a broader perspective, using different multivariate tests. The limit test, which was used in our earlier study, gave high sensitivity but suffered from poor specificity. The performance of other methods such as factorial discriminant analysis and partial least square discriminant analysis are at par with more complex nonlinear methods such as decision trees, but they provide very little information about the classification model. This comparative study thus demonstrates not just the efficacy of Raman spectroscopic models but also the applicability and limitations of different multivariate tools for discrimination under complex conditions such as the multicancer scenario.
癌症现在被认为是发病率和死亡率的主要原因之一。组织病理学诊断是金标准，但它被证明是主观的、耗时的、容易受到观察者之间意见分歧的影响，并且常常无法预测预后。光学光谱方法被认为是传统癌症诊断的辅助手段或替代方法。这些方法最重要的方面是它们的客观性，多元统计工具在实现这一目标方面发挥着重要作用。然而，严格评估光谱模型的稳健性是前提条件。拉曼光谱在癌症诊断中的应用已经得到了很好的证实。到目前为止，光谱模型的特异性和适用性已经针对特定的癌症类型进行了评估。在这项研究中，我们使用不同的多元测试，从更广泛的角度评估了代表乳腺、宫颈、结肠、喉和口腔的正常和恶性组织的光谱模型的实用性。我们之前的研究中使用的极限测试具有很高的灵敏度，但特异性较差。其他方法（如因子判别分析和偏最小二乘判别分析）的性能与决策树等更复杂的非线性方法相当，但它们提供的分类模型信息很少。因此，这项比较研究不仅展示了拉曼光谱模型的有效性，还展示了不同多元工具在复杂条件（如多癌情况）下进行区分的适用性和局限性。
