Deep interpretation of mammographic images in breast cancer screening

项目总结/摘要 筛查性乳房X光检查已被证明在早期发现乳腺癌和减少乳腺癌的发生方面是有效的。 mortality.然而，争议和挑战仍然存在，主要关注个人乳腺癌 乳腺X线实质标志物的风险预测，高召回率和良性活检率，以及改善 放射科医师的临床阅读实践。在这些方面已经开发了计算机化方法， 目标是为放射科医生提供计算机辅助，帮助他们做出临床决策。虽然成功， 这些方法的准确性取决于适当的数据表示（即，图像特征），需要 强大的特征工程。一种新出现的人工智能技术，称为深度学习， 这是机器学习范式的突破，并彻底改变了计算机图像分析和许多 在过去的几年里，其他应用。乳腺癌筛查产生了大量的乳房X线照片数据 这需要深入的解释，以改善目前的临床检查。本研究的目的是开发和 优化基于卷积神经网络（CNN）的计算方法，以改善乳腺X射线摄影 成像特征识别，分析和解释，并使用这种方法来解决准确的乳腺癌 癌症风险预测和降低错误回忆率。这项研究将是第一个检查的影响， 革命性的深度学习技术，用于对大型筛查乳房X线照片数据进行深入解读， 旨在改善临床实践。新的风险生物标志物将有助于提供更准确的风险 比目前的预测。召回决策模型将有助于减少错误召回（与 潜在的良性活检结果），并更好地了解放射科医师的阅读行为。总的来说，基于CNN的 该方法将优化筛查性乳腺X射线摄影的临床效用，并且很有可能转化为 乳腺癌筛查诊所

Project Summary/Abstract Screening mammography has been shown effective in early detection of breast cancer and in reducing mortality. However, controversies and challenges still remain, with primary concerns on personal breast cancer risk prediction from mammographic parenchymal markers, high recall and benign biopsy rates, and improving radiologists’ clinical reading practices. Computerized methods have been developed in these regards, with the goal of providing computer assistance to radiologists in making clinical decisions. While successful, the accuracy of these methods is subject to appropriate data representation (i.e., image features) that requires strong feature engineering. A newly emerged artificial intelligence technique, called deep learning, represents a breakthrough in machine learning paradigms, and has revolutionized computer image analysis and many other applications in the past few years. Breast cancer screening yields a huge amount of mammogram data that requires in-depth interpretation to improve current clinical workup. The goal of this study is to develop and optimize a convolutional neural network (CNN)-based computational approach to improve mammographic imaging trait identification, analysis, and interpretation and to use this approach to address accurate breast cancer risk prediction and reduce false recall rates. This study will be the first to examine the effects of the revolutionary deep learning technique on performing in-depth interpretation of big screening mammogram data, aimed at improving clinical practice. The new risk biomarkers will contribute to providing more accurate risk prediction than currently available. The recall-decision model will help reduce false recalls (associated with potential benign biopsy results), and better understand radiologists’ reading behaviors. Overall, the CNN-based approach will optimize the clinical utility of screening mammography and has a high likelihood to translate to the clinic for breast cancer screening.