首页 > 文章详情

人工智能将如何改变癌症治疗？

2023年08月14日

来源：癌症研究UPDATE

以下内容原文发布于AACR官方博客《Cancer Research Catalyst》, 中文内容仅做参考，请点击文末“阅读原文”，阅览原文内容。

人工智能（artificial intelligence，AI）革命对医学各领域具有重要意义，尤其是癌症的精准治疗。医疗用人工智能工具掀起的第一波浪潮已经改变了多项技术，如放射成像中使用的乳房X线摄影和CT扫描，以及用于观察活检组织和手术切除标本的显微技术。

得益于统计和机器学习算法（包括神经网络和被称为“深度学习”的多层神经网络），模式识别和分类越发精细。这几类图像分析应用于医学成像和病理诊断后，有助于癌症的检测和诊断，以及判定癌症的侵袭性。放射影像采集和病理切片扫描技术的进步使得收集到的数据量大大增加，这也使得人工智能辅助阅片越发重要。

在过去，放射科医生和病理学家依靠X光照片和显微镜进行观察，使用肉眼分辨特征，用大脑来识别通过上述方法获得的低分辨率图像中的（病理）模式。然而，技术的进步提高了图像分辨率和数据丰富度，随之产生的大量数据超出了人眼和大脑的处理极限，使得基于人工智能的辅助阅片格外重要。

建立放射学、病理学相关人工智能辅助工具的基本策略是，使用大量已知结果（如癌症与非癌症、侵袭性癌症与非侵袭性癌症等）的病例图像文件来“训练”计算机算法，然后使用另一组病例来验证算法。随着后续分析的图像文件越来越多，该算法可不断得到改进。

了解更多内容，请阅读以下原文。

How Will Artificial Intelligence Change Cancer Care?

The artificial intelligence (AI) revolution has important implications for all of medicine, and especially for precision cancer care. The first wave of medical AI tools has impacted technologies like mammography and CT scans used in radiographic imaging and microscopy used to examine tissue biopsies and surgical resection specimens.

Statistical and machine-learning algorithms, including neural networks and layers of neural networks termed “deep learning,” have allowed increasingly refined pattern recognition and classification. When applied to medical imaging and diagnostic pathology, these types of image analyses can aid in cancer detection and diagnosis, and in determining cancer aggressiveness. Advances in radiographic image acquisition and pathology slide scanning—leading to much larger amounts of data collected—are making AI-aided interpretations increasingly essential.

William G. Nelson, MD, PhD

Historically, radiologists gazed at X-ray pictures and pathologists looked through microscopes. Both used their eyes to resolve features and their brain to recognize patterns from the lower-resolution images obtained through these methods. However, technological advances have increased the resolution and data richness of the images, producing vast amounts of data beyond what the human eye and brain can process, making AI-based assistance invaluable.

The basic strategy used for building AI tools to aid radiology and pathology is to “train” a computer algorithm using a large collection of image files from cases with known outcomes, such as cancer versus noncancer, aggressive cancer versus nonaggressive cancer, and so on, and then to validate the algorithm using a second set of cases. Subsequently, the algorithm can continue to improve as it analyzes more image files.

These approaches will quickly work their way into common clinical practice. In 2021, the U.S. Food and Drug Administration released an Artificial Intelligence/Machine Learning Action Plan, anticipating a wave of device applications featuring computer-aided approaches to a wide variety of medical indications. By October 2022, 521 approvals of AI-aided medical devices had been made.

So far, AI has not replaced expert radiologists or pathologists. Rather, the refined classification capabilities have been used to direct the human eye to critical image features. Will this improve precision cancer medicine by evolving into a more highly functioning human-machine interface? As the noted sage Yogi Berra once pointed out, “It’s tough to make predictions, especially about the future.”

William G. Nelson, MD, PhD, is the editor-in-chief of Cancer Today, the quarterly magazine for cancer patients, survivors, and caregivers published by the American Association for Cancer Research. Nelson is the Marion I. Knott professor of oncology and director of the Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins in Baltimore. You can read his complete column in the summer 2023 issue of Cancer Today.

阅读更多内容，请点击“阅读原文”

阅读原文

AACR

结尾

责任编辑：肿瘤资讯-Bree
排版编辑：肿瘤资讯-李莹洁

免责声明

本文仅供专业人士参看，文中内容仅代表癌症研究UPDATE立场与观点，不代表肿瘤资讯平台意见，且肿瘤资讯并不承担任何连带责任。若有任何侵权问题，请联系删除。

领新版指南，先人一步>>

查看详情