Application of deep learning to segmentation and grading of fine needle biopsy breast images

深度学习在乳腺细针活检图像分割和分级中的应用

• 批准号: 538142-2019
• 负责人: Krzyzak, Adam
• 金额: $ 1.82万
• 依托单位: Concordia University
• 依托单位国家: 加拿大
• 项目类别: Engage Grants Program
• 财政年份: 2019
• 资助国家: 加拿大
• 起止时间: 2019-01-01 至 2020-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=680275
• 关键词: Application; deep; learning; segmentation; grading

The effectiveness of cancer treatment largely depends on the early detection of the disease. Important and often used diagnostic methods are x rays, cat scans, MRI scans and ultrasonography imaging. Once early signs of cancer are detected, doctors often recommend biopsy of suspicious mass. For many types of cancers, including breast cancer and liver cancer, fine needle aspiration biopsy (FNB) is a preferred diagnostic technique as it is less invasive than the open and closed surgical biopsy. Determining cancer grade from FNB images may be a challenging task even for pathologists since aspiration process may result in partial destruction of the tissue structure and sometimes even of nuclei. Recent advances in machine learning and high performance data computation have created new possibilities for intelligent processing of medical images with high accuracy. In particular, deep convolutional neural networks have demonstrated the capacity to recognize various objects within an image, and can outperform state-of-the-art object recognition techniques. The proposed research aims at creating an innovative computational system for rapid and reliable cancer grading utilizing computer-expert imaging software developed by the ORS company for automatic segmentation and grading of virtual FNB slides using deep convolutional neural networks combined with the state-of-the-art random forest classifiers.

癌症治疗的有效性在很大程度上取决于疾病的早期发现。重要和常用的诊断方法是X射线，CAT扫描，MRI扫描和超声成像。一旦发现癌症的早期迹象，医生通常会建议对可疑的肿块进行活检。对于许多类型的癌症，包括乳腺癌和肝癌，细针穿刺活检（FNB）是首选的诊断技术，因为它比开放和闭合手术活检的侵入性更小。从FNB图像确定癌症等级可能是一项具有挑战性的任务，甚至对于病理学家来说，因为抽吸过程可能导致组织结构的部分破坏，有时甚至是细胞核的部分破坏。机器学习和高性能数据计算的最新进展为高精度的医学图像智能处理创造了新的可能性。特别是，深度卷积神经网络已经证明了识别图像中各种对象的能力，并且可以超越最先进的对象识别技术。拟议的研究旨在创建一个创新的计算系统，用于快速可靠的癌症分级，利用ORS公司开发的计算机专家成像软件，使用深度卷积神经网络结合最先进的随机森林分类器对虚拟FNB切片进行自动分割和分级。