Optimizing Oral Cancer Screening and Precision Management of Potentially Malignant Oral Lesions

项目摘要 尽管在过去的几十年里治疗取得了进展，但口腔癌的癌症特异性生存率 情况依然黯淡，主要是因为大多数病例是在晚期诊断出来的。早期检测 癌症(最常见的是口腔鳞状细胞癌(OSCC))可以减少毁容，降低治疗成本 带着治愈的意图。然而，传统的视觉-触觉检查在口腔癌和口腔癌前病变中的局限性 癌症损害阻碍了癌症检测和筛查的支持。分离目视检查 从癌前病变或癌病变中区分良性病变是不准确的，因此标准做法需要 转诊和手术刀活检大多数潜在的口腔恶性病变。此外，约20%的 潜在的恶性口腔病变包含一定程度的上皮不典型增生或癌，因此早期 确诊可以提供根治性治疗，因为大多数口腔鳞癌通常以不典型增生开始，而 不典型增生的程度与恶变率有关。口腔筛查的批评者引用了 口腔良性病变和轻度不典型增生的高患病率使患者面临损害的风险 因为过度检测和过度治疗。因此，辅助诊断可以改变筛查工作。 在护理点提供高度准确的细胞病理信息的测试，如NIDCR支持的 护理点口腔细胞病理学工具。计算机视觉辅助的精密成像测试最近显示 对口腔病变特征的强大诊断性能，但其潜在的陷阱和承诺必须 在临床应用前进行了彻底的调查。同样，机器学习可以支持光学测试 将潜在的恶性病变可视化。如果成功，这些人工智能设备可以帮助决策- 为低风险病变进行、防止不必要的手术刀活检，并实现风险分层监测或 治疗。我们的专家团队在计算机疾病模拟建模、机器学习、口腔医学和 经济评估将改变疾病模拟模型，以在护理点提供分析，以及 评估精确成像诊断的不同潜在用途，以便转化为临床护理。我们会 扩展我们现有的潜在恶性口腔病变的疾病模型，以代表病变特征和 临床风险类别(例如，基于烟草和酒精使用)通过纳入大型纵向 数据集(目标1)，以评估人工智能辅助细胞学检测是否可以改善 低、中、高风险类别筛查的有效性和成本效益(目标2)。最后，我们 将评估用于病变可视化的辅助设备是否具有良好的效果和成本效益 在有或没有人工智能支持的情况下跨风险类别进行筛选，并为 模型(目标3)。这项工作将产生一个分析引擎来指导人工智能的临床翻译-- 用于口腔病变检测和表征的辅助诊断，以克服筛查可靠性不足的问题。

Project Summary Despite treatment advances over the past several decades, cancer-specific survival for oral cancers remains bleak, mostly due to the majority of cases being diagnosed at late stages. Early-stage detection of cancers (most often oral squamous cell carcinoma (OSCC)) would enable less disfiguring, less costly therapy with curative intent. However, limitations of traditional visual-tactile examination for oral cancerous and pre- cancerous lesions have hindered cancer detection and support for screening. Visual inspection for separation of benign from precancerous or cancerous lesions is inaccurate, and therefore standard practice entails referral and scalpel biopsy of most potentially malignant oral lesions. Furthermore, approximately 20% of potentially malignant oral lesions contain some degree of epithelial dysplasia or carcinoma, and therefore early identification could allow curative treatment as the majority of OSCC typically starts as dysplasia, and the degree of dysplasia is correlated with the rate of malignant transformation. Detractors of oral screening cite the high prevalence of benign oral lesions and mild dysplasia as circumstances placing patients at risk of harms from over-testing and over-treatment. Thus, screening efforts could be transformed by adjunctive diagnostic tests that offer highly accurate cytopathologic information at the point of care, such as the NIDCR-supported Point-of-Care Oral Cytopathology Tool. Computer vision-assisted precision imaging tests have recently shown strong diagnostic performance for oral lesion characterization, but their potential pitfalls and promises must be thoroughly investigated before clinical application. Similarly, machine learning could bolster optical tests for visualizing potentially malignant lesions. If successful, these artificial intelligence devices could aid decision- making, preventing unnecessary scalpel biopsies for low-risk lesions and enabling risk-stratified surveillance or treatment. Our team of experts in computer disease simulation modeling, machine learning, oral medicine, and economic evaluation will transform a disease simulation model to provide analysis at the point of care, and evaluate the different potential uses of precision imaging diagnostics for translation to clinical care. We will expand our existing disease model of potentially malignant oral lesions to represent lesion characteristics and clinical risk categories (e.g. based on tobacco and alcohol use) through incorporation of large longitudinal datasets (Aim 1), in order to evaluate whether artificial intelligence-assisted cytologic testing can improve the effectiveness and cost-effectiveness of screening for low, moderate, or high risk categories (Aim 2). Finally, we will evaluate whether adjuncts for lesion visualization render favorable effectiveness and cost effectiveness of screening across risk categories, with or without artificial intelligence support, and develop a user interface for the model (Aim 3). This work will produce an analytic engine to guide clinical translation of artificial intelligence- aided diagnostics for oral lesion detection and characterization, to overcome insufficient screening reliability.