Automated FISH Imaging and Analysis for Screening Cervical Cancer

荧光原位杂交（FISH）成像已被证明是检测癌症、预测癌症预后和监测治疗效果的强大分子成像工具。然而，人工FISH分析方法非常繁琐，并且引入了很大的读者间差异。使用人工FISH分析方法从一个样本上描绘的压倒性的正常细胞中检测少量的异常染色体细胞是困难的或不切实际的（即，使用巴氏涂片筛查早期宫颈癌）。为了解决这个困难，我们提出了开发和测试的原型系统的自动FISH成像分析。该系统包括光学成像仪和计算机辅助检测（CAD）方案。与目前可用的技术相比，我们提出的系统提供了独特的优势。首先，它提高了成像效率。其次，它允许CAD处理从测试样本中检测比当前手动FISH图像分析方法显著更多的可分析或异常细胞。我们的假设是，如果在一个样本上描绘的所有细胞和FISH信号都可以自动成像并正确计数，则可以降低遗漏异常或癌细胞的早期癌症迹象的概率，从而提高癌症诊断的灵敏度。为了验证这一假设，我们将进行一些具体的研究任务。首先，我们将收集和组装一个初步的图像数据库，其中包括细胞学和FISH图像的每次巴氏涂片检查。大约一半的病例描述了导致癌症发展的宫颈异常。其次，我们将建立一个原型荧光图像扫描系统，并测试其质量和可靠性。第三，我们还将开发并初步测试我们的CAD方案在从临床图像中选择和分割可分析的间期细胞、识别和计数独立的FISH标记的生物标志物信号（斑点）以及在正常和异常（癌症）病例之间进行分类方面的准确性。在初步整合我们的初步成像系统和CAD方案后，我们将进行表征研究，以测试该技术的性能。本研究将有助于我们初步评估开发和应用这种自动化FISH图像分析系统和方法的可行性。该项目的成功可以产生更多过去无法做到的后续临床研究。

Fluorescent in situ hybridization (FISH) imaging has been proved as a powerful molecular imaging tool to detect cancers, predict cancer prognosis, and monitor therapy efficacy. However, manual FISH analytic method is very tedious and introducing large inter-reader variability. It is difficult or unpractical to use manual FISH analytic method to detect small number of abnormal chromosome cells from the overwhelming normal cells depicted on one specimen (i.e., screening for early cervical cancer using Pap smear). To solve this difficulty, we proposed to develop and test a prototype system of automated FISH imaging analysis. The system includes both an optical imager and a computer-aided detection (CAD) scheme. Comparing to the currently available techniques, our proposed system provides unique advantages. First, it improves imaging efficiency. Second, it allows CAD processing to detect significantly more analyzable or abnormal cells from a testing specimen than the current manual FISH image analytic method. Our hypothesis is that if all cells and FISH signals depicted on one specimen can be automatically imaged and correctly counted, the probability of missing early cancer sign of abnormal or carcinoma cells can be reduced and thus improving sensitivity of cancer diagnosis. To test this hypothesis, we will conduct a number of specific research tasks. First, we will collect and assemble a preliminary image database that includes both cytology and FISH images for each Pap smear examination. Approximately half of cases depict cervical abnormalities that lead to the development of cancer. Second, we will build a prototype fluorescent image scanning system and test its quality and reliability. Third, we will also develop and preliminarily test the accuracy of our CAD schemes in selecting and segmenting analyzable interphase cells from clinical images, identifying and counting independent FISH labeled biomarker signals (spots), and classifying between the normal and abnormal (cancer) cases. After initial integration of our preliminary imaging system and CAD scheme, we will conduct characterization study to test the performance of the technique. This pilot study will help us to preliminarily assess the feasibility of developing and applying this automated FISH image analysis system and method. The success of this project can generate more follow-up clinical studies that are unable to do in the past.