Infrared Spectroscopic Imaging and Machine Learning for Risk Stratification of Oral Epithelial Dysplasia

项目摘要/摘要 口腔粘膜病变的成功治疗和管理依赖于明确、准确和及时的 诊断。尽管口腔很容易进入，但口腔鳞状细胞癌(OSCC)是最常见的 口腔癌，通常直到晚期才被诊断出来，导致预后很差。口腔上皮异型增生(OED) 是一种经显微镜诊断的癌前病变，与口腔鳞癌转化的风险增加有关。 根据世界卫生组织的组织学分级，OED可分为轻度、中度或重度 三级分类体系。不幸的是，组织病理学诊断的黄金标准依赖于主观 活检组织的形态评估，无法识别最有可能发生的高危OED 经历恶变。缺乏客观和定量的OED风险分层方法 预防了口腔癌前病变的有效处理，延误了口腔鳞癌的诊断。我们建议 一种利用傅里叶变换红外光谱成像和机器学习的新方法 解决客观的OED风险评估的医学差距。FTIR光谱提供了定量的 以特征吸收光谱的形式表示的样品的生化信息。用显微镜 与FTIR光谱仪相结合，FTIR成像可以对一种 样本，每个像素包含一个完整的FTIR光谱。机器学习是高光谱研究的有力工具 FTIR图像分析和诊断模型的开发。使用机器学习辅助的FTIR成像，我们 成功地训练了三个机器学习分类器，识别口腔鳞癌和口腔鳞癌的准确率为95%-100% 我们对口腔良性组织进行了初步研究。更令人兴奋的是，我们的结果展示了一种创新的分层 根据上皮性FTIR指纹将重度OED分为良性和OSCC样亚组。灵感 根据早期的发现，这一提议的中心假设是FTIR成像辅助机器学习 提供客观和定量的OED风险分层。为了检验这一假设，我们提出了以下两个假设 具体目标：1)开发基于上皮和间质FTIR指纹的口腔鳞癌良性分类器，以及2) 目的：评价基于FTIR图像的方法在OED危险分层中的可行性。中国的长期目标是 研究是开发一种人工智能辅助精密成像系统，使用FTIR成像或在 与其他形态和功能成像方式相结合，如数字病理学和 免疫组织化学用于口腔癌的早期诊断、治疗和预防。

PROJECT SUMMARY/ABSTRACT Successful treatment and management of oral mucosal lesions depend on a definitive, accurate, and timely diagnosis. Despite easy accessibility to the oral cavity, oral squamous cell carcinoma (OSCC), the most common oral cancer, is often not diagnosed until late stages, leading to a poor prognosis. Oral epithelial dysplasia (OED) is a microscopically diagnosed precancerous lesion associated with an increased risk of OSCC transformation. An OED can be histologically graded as mild, moderate, or severe based on the World Health Organization’s three-tier classification system. Unfortunately, the gold standard histopathological diagnosis relies on subjective morphological evaluation of the biopsy tissue and is unable to identify high-risk OEDs that are most likely to undergo malignant transformation. The lack of an objective and quantitative OED risk stratification approach has prevented effective management of precancerous oral lesions and delayed the diagnosis of OSCC. We propose a novel approach using Fourier transform infrared spectroscopic (FTIR) imaging and machine learning to address the medical gap of objective OED risk assessment. FTIR spectroscopy provides quantitative biochemical information of a sample in the form of characteristic absorption spectrum. With a microscope coupled to an FTIR spectrometer, FTIR imaging allows detailed and spatially resolved biochemical analysis of a sample, with each pixel containing a full FTIR spectrum. Machine learning is a powerful tool for hyperspectral FTIR image analysis and diagnostic model development. Using FTIR imaging aided by machine learning, we successfully trained three machine learning classifiers with 95–100% accuracy in discriminating OSCC from benign oral tissues in our preliminary study. More excitingly, our results demonstrated an innovative stratification of severe OEDs into Benign-like and OSCC-like subgroups based on their epithelial FTIR fingerprints. Inspired by the early finding, the central hypothesis of this proposal is that FTIR imaging aided by machine learning provides objective and quantitative OED risk stratification. To test the hypothesis, we propose the following two specific aims: 1) to develop OSCC-Benign classifiers based on epithelial and stromal FTIR fingerprints, and 2) to evaluate the feasibility of the FTIR image-based approach in OED risk stratification. The long-term goal of the research is to develop an artificial intelligence aided precision imaging system using FTIR imaging or in combination with other morphological and functional imaging modalities such as digital pathology and immunohistochemistry for early oral cancer diagnosis, treatment, and prevention.