Information-Theoretic Based CAD in Mammography

描述(由申请人提供)： 本研究的目的是开发和评估一种新的计算机辅助决策(CAD)方案，以提高乳腺肿块筛查的临床诊断水平。该方案将信息论相似性度量与基于知识的决策算法相结合。它将帮助放射科医生仔细检查乳房X光照片，提供基于证据的决策支持。给定查询的乳房X光照相区域，CAD系统将询问存档的乳房X光照片的数据库，检查类似的病例，并指定关于潜在恶性肿块的存在的可能性测量。 这项研究提出了信息论度量的公式来量化两个乳房X光检查区域的相似性。相似性度量基于沙龙的熵；这是一种对图像中包含的复杂性(或信息)的度量。从理论上讲，如果两个乳房X光照相区域描绘的结构相似，它们应该包含彼此的诊断信息。相关诊断信息量可以通过基于熵的相似性度量来测量，该相似性度量是直接从图像计算的，而不需要分割或特征提取。利用相似性度量和已知真相的乳房X光检查病例图像库，将实施知识总线式CAD方案，用于筛查乳房X光检查中的肿块检测。初步研究证实，标准互信息(MI)是一种有效的任务相似性度量。 这项研究的具体目标是：(1)充分利用衡量两个乳房X光照相区域相似内容的信息论度量标准，(2)优化它们在基于证据的决策算法中的贡献，以早期发现潜在的恶性肿块，以及(3)对CAD系统进行初步临床评估。 由于数字图像库是放射学的一个即将到来的趋势，拟议的CAD系统将利用连续存储的具有既定基本事实的乳房X光照片。该系统旨在减少与筛查乳房X光检查相关的解释错误和/或提示CAD方案产生的假阳性，总体而言，本研究旨在提高敏感性，同时保持或提高筛查乳房X光检查对肿块的特异性。

DESCRIPTION (provided by applicant): The purpose of the study is to develop and evaluate a novel computer-assisted decision (CAD) scheme for improving the clinical detection of breast masses in screening mammograms. The CAD scheme combines information-theoretic similarity metrics with knowledge-based decision algorithms. It will help radiologists scrutinize mammograms providing evidence-based decision support. Given a query mammographic region, the CAD system will interrogate a database of archived mammograms, examine similar eases, and assign a likelihood measure regarding the presence of a potentially malignant mass. The study proposes the formulation of information-theoretic metrics to quantify the similarity of two mammographic regions. The similarity metrics are based on Sharmon's entropy; a measure of complexity (or information) contained in an image. Theoretically, if two mammographic regions depict similar structures, they should contain diagnostic information for each other. The amount of relevant diagnostic information can be measured by entropy-based similarity metrics that are computed directly from the images without requiring segmentation or feature extraction. Using the similarity metrics and an image databank of mammographic cases with known truth, a knowledge-bussed CAD scheme will be implemented for the detection of masses in screening mammograms. Preliminary studies have established that standard mutual information (MI) is an effective similarity metric for the task. The specific aims of the study are: (1) To fully exploit information-theoretic metrics that measure the similar content of two mammographic regions, (2) To optimize their contributions in an evidence-based decision algorithm for the early detection of potentially malignant masses, and (3) To perform preliminary clinical evaluation of the CAD system. As digital image libraries are an upcoming trend in radiology, the proposed CAD system will take advantage of continuously deposited mammograms with established ground truth. The system aims to reduce the interpretation error associated with screening mammograms and/or the false positives generated by cuing CAD schemes, Overall, the study aims to improve the sensitivity while maintaining or improving the specificity of screening mammography for masses.