A computer aided chromosome imaging technique for cancer diagnosis

描述（由申请人提供）：识别复发性染色体畸变对大多数血液系统恶性肿瘤的诊断、预后和治疗都很重要。由于肿瘤细胞培养困难，有丝分裂指数低，染色体形态差，患病率低，细胞遗传学临床医生需要花费大量的精力和时间才能在显微镜下获得足够数量的可分析的中期细胞，才能做出准确的临床诊断。这个过程不仅效率低下，而且容易出现人为错误。为了提高白血病诊断的效率和准确性，我们建议发展一种计算机辅助染色体成像技术。具体而言，我们将开发一种基于延时积分技术的创新高速显微成像系统。该系统可对整个载玻片进行高倍扫描，获得临床诊断所需的中期染色体的高分辨率数字图像。我们还将开发一种新的计算机辅助诊断（CAD）方案，包括四个特定模块，以(1)检测可分析的中期染色体细胞，(2)片段重叠染色体，(3)识别和分类与癌细胞相关的畸变染色体，以及(4)预测癌症预后。在对可分析染色体进行识别和分割后，我们将计算和搜索有效且鲁棒的图像特征。遗传算法将用于训练和优化人工神经网络和贝叶斯信念网络，分别用于分类和预测任务。利用集成的CAD工作站，我们将进行一项观察员性能研究，以评估该技术的性能及其临床可行性。总之，所提出的成像技术是非常高效的，从最初的幻灯片扫描到CAD结果的呈现，不需要或只需要很少的人为干预。有了这样一个新的计算机化临床工具，细胞遗传学家可以有效地集中精力分析/验证染色体异常模式并做出最终诊断决策。因此，期望所提出的技术能够显著提高癌症（即白血病）诊断的效率和准确性。该技术在监测肿瘤治疗效果方面也具有重要的临床潜力。

DESCRIPTION (provided by applicant): Identification of recurrent chromosomal aberrations is important for diagnosis, prognosis, and therapy of most hematological malignancies. Due to difficulties with culture of tumor cells, low mitotic index, poor chromosomal morphologies, and low prevalence, it takes tremendous effort and time for a cytogenetic clinician to obtain a sufficient number of analyzable metaphase cells under microscope before he/she can make an accurate clinical diagnosis. This process is not only very inefficient but also subject to human errors. In order to improve the efficiency and accuracy of leukemia diagnosis, we propose to develop a computer aided chromosome imaging technique. Specifically, we will develop an innovative high-speed microscopic imaging system based on a time-delay-integration technique. The system can scan the entire sample-slide at high magnification to obtain high resolution digital images to reveal metaphase chromosomes as required by clinical diagnosis. We will also develop a novel computer aided diagnosis (CAD) scheme including four specific modules to (1) detect analyzable metaphase chromosome cells, (2) segment overlapped chromosomes, (3) identify and classify distorted chromosomes associated with cancer cells, and (4) predict the cancer prognosis. After identification and segmentation of analyzable chromosomes, we will compute and search for the effective and robust image features. Genetic algorithm will be used to train and optimize an artificial neural network and a Bayesian belief network for the classification and prediction tasks, respectively. Using the integrated CAD workstation, we will conduct an observer performance study to assess the performance of the technique and its clinical feasibility. In summary, the proposed imaging technique is highly efficient, and no or only minimal human interventions are required from initial slide-scanning up to the presentation of CAD results. With such a new computerized clinical tool, cytogeneticists can effectively focus their efforts on analyzing/verifying chromosomal abnormal patterns and making final diagnostic decisions. It is therefore expected that the proposed technology can significantly improve the efficiency and accuracy of cancer (i.e., leukemia) diagnosis. The proposed technique has significant clinical potentials in monitoring therapeutic efficacy of cancer treatment as well.