没有合适的资源?快使用搜索试试~ 我知道了~
从实验室到病床:深度学习在医疗保健中的旅程.docx
1.该资源内容由用户上传,如若侵权请联系客服进行举报
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
版权申诉
0 下载量 19 浏览量
2024-04-23
07:18:58
上传
评论
收藏 124KB DOCX 举报
温馨提示
试读
11页
从实验室到病床:深度学习在医疗保健中的旅程.docx
资源推荐
资源详情
资源评论
November 2017, IDC #US43185617
White Paper
From Bench to Bedside: Deep Learning's Journey in Healthcare
Sponsored by: NVIDIA
Cynthia Burghard
November 2017
IDC OPINION
Current global efforts to improve the clinical and financial outcomes in healthcare require innovation to
be successful. Technology, and in the case of this research artificial intelligence (AI) and deep
learning, can play a key role in that innovation — from improving the timeliness and accuracy of
diagnoses to informing the treatment of diseases to helping address issues of physician shortages. AI
and deep learning can have significant impact on healthcare organizations and the ecosystem at large
as they are incorporated into ongoing digital transformation initiatives.
IDC defines digital transformation as the continuous process by which organizations adapt to or drive
disruptive changes in their external ecosystem, leveraging digital competencies to improve efficiencies
and performance. By 2019, 40% of digital transformation initiatives will be supported by cognitive/AI
capabilities, providing critical on-time insights.
While deep learning is at the early stage in the adoption cycle, healthcare organizations must
recognize that the maturity of the technology is not the gating factor to adoption; rather, the skepticism
of physicians, the regulatory environment, reimbursement policy, and other issues must be addressed
to move the application of deep learning from bench to bedside.
This white paper explores the application of deep learning in two key areas of healthcare: the
augmentation or assistance to physicians and predicting the onset of an illness or adverse health
event. In all cases, there are initiatives across the broad adoption spectrum, from early stage to
production. You will note that those applications of deep learning that are closest to the delivery of care
are in earlier stages of adoption than other applications as they show promise in improving clinical and
financial outcomes.
WHAT IS DEEP LEARNING?
Deep learning is a type of machine learning where the learning happens in successive layers — each
layer of the neural network adding to the knowledge of the previous layer. Each layer teaches the next
successive layer. Deep learning trains the machine to do what the human brain does naturally.
Feeding unlimited data to an algorithm is successful only when feature extraction is present to guide
the machine about the specific learning objects. In traditional machine learning, the programmer had to
intervene and guide the machine on which features it should be looking for in the learning process.
The burden on the programmer is significant, but more significant is the potential for human error. In
deep learning, programmer intervention is not required, resulting in increased accuracy.
©2017 IDC
#US43185617
2
Deep learning is a type of machine learning that leverages representation learning methodologies to
allow computers to process natural data in its raw form using neural network algorithms. It takes
advantage of both supervised and unsupervised machine learning:
▪
Supervised machine learning begins with examples of training data paired with identifying
labels (e.g., right or wrong and positive or negative) selected from the categories to be
learned. Using these pairs of example data and labels (training data), the system learns
parameters of statistical models that it can then generalize to unlabeled examples of data
items that were not seen in the training data (test data). In most cases, the learned models
improve over time via a feedback loop that adjusts the model parameters to better reflect
additional sets of training or production data. The performance of a learned model can be
measured by simple prediction accuracy or by the business metric the learned model is
designed to support. Performance depends on the degree to which the training data matches
the real world, the choice of algorithm, the algorithm's parameters, and the quantity of data.
▪
Unsupervised machine learning is another variation of machine learning where algorithms
detect and discern attributes and features without the benefit of labeled training data. Some
algorithms cluster data into meaningful groups by finding centers of data density. Other
unsupervised algorithms use dimensionality reduction techniques (such as singular-value
decomposition) to uncover the essential attributes of the data without requiring a human to
define those attributes in advance. This is particularly useful for "unstructured" data, such as
images or text, where an underlying structure can be automatically inferred, enabling other
algorithms to leverage the data.
Chief among the benefits of deep learning is lack of programmer intervention, which improves the
accuracy of results. Healthcare data is diverse, comprising both structured and unstructured data, both
of which can be ingested by deep learning. This ability is driving the growing adoption of deep learning
in healthcare, particularly in the radiology and pathology domains where there is voluminous data in
the form of images, clinician notes, reports, and tissue slides.
APPLICATION OF DEEP LEARNING IN HEALTHCARE
Deep learning in healthcare is augmenting clinical decision making in areas ranging from analyzing
medical research findings and best practices to prioritize and recommend treatment options to
detecting abnormalities in radiology images and pathology slides to identifying genomic markers in
tissue samples. In predictive analytics, deep learning is being applied to the early detection of disease,
the identification of clinical risk and its drivers, and the prediction of future hospitalization. While there
are opportunities for the application of deep learning in other aspects of healthcare, this white paper
focuses on clinician augmentation and prediction of the risk of disease and/or adverse clinical events.
Augment Clinicians
Among the many drivers to augment clinicians' abilities are the overwhelming volume and rapid pace
of new treatment options, the high volume of screenings, the urgency to get results when disease is
suspected and, in some places, physician shortages. These can be addressed with the application of
deep learning. Improving the accuracy of evaluating radiologic images and pathology slides and the
ability to organize fragmented patient data and deliver it in the context of a physician's inquiry are all
benefits that can be realized.
©2017 IDC
#US43185617
3
Reduce Cognitive Burden
One of the key advantages of deep learning is its ability to digest large volumes of data and create
models that can determine the impact on clinical decisions. It is virtually impossible for physicians to
stay abreast of advances in medical care in real time because of an explosion of medical information.
For example, currently, approximately 50,000 oncology research papers are published annually,
1
and
by 2020, medical information is projected to double every 73 days,
2
outpacing the ability of humans to
keep up with the proliferation of medical knowledge. Applying deep learning to voluminous data
enables the model to synthesize millions of data points and draw out correlations that can guide
clinical decisions. This reduces the cognitive burden on physicians, enabling them to keep pace with
current research and treatment protocols.
Improve Accuracy and Speed to Diagnoses and Treatment
Another area where deep learning has shown promising results is in enabling physicians to improve
the accuracy and speed of diagnosis. Medical images, pathology samples, and genomics information
are among the most important tools doctors use in diagnosing conditions. However, analyzing this data
can often be a difficult and time-consuming process. Deep learning can assist by automating some of
the analysis, screening for abnormalities or areas of interest within the medical data rapidly, and
enabling physicians to better focus their expertise. These combined approaches not only can achieve
improved accuracy and speed but could also increase the objectivity of diagnosis as well as
concordance rate.
Researchers at Mayo Clinic concluded that using deep learning could lead to an earlier and more
accurate way to diagnose and treat brain tumors. It is also a way to track progress or response to
treatment without surgery. Using tissue from the tumor and a medical resonance imaging (MRI), the
team identified a type of chromosomal damage that is important for predicting how well patients with a
low-grade type of tumor will respond to chemotherapy and radiation. The research was conducted on
155 patients with glioblastoma multiforme, which is the most common primary brain tumor, accounting
for 45% of all malignant primary central nervous system tumors, and with a median survival of around
14 months. The work, called radiogenomics, reflects the thought that genomic properties of tumors can
be determined in the appearance of images (in this case, medical resonance imaging).
3
Another area where artificial intelligence is being applied is cardiology, where deep learning is being
used with MRIs to measure the volume of blood transported with each pump of the heart's four
chambers. Cardiologists typically need 30–60 minutes to calculate the volume; AI comes up with the
answer within seconds.
Pathology is another area that benefits from the application of deep learning to improve the accuracy
of diagnoses. A pathologist's report is often the gold standard in the diagnosis of many diseases.
Review of pathology slides is a very complex task, requiring years of training to gain the required
expertise and experience. Even with this extensive training, there can be substantial variability in the
diagnoses given by different pathologists for the same patient. For example, agreement in diagnosis
for some forms of breast cancer can be as low as 48% and similarly low for prostate cancer. The lack
of agreement is not surprising given the massive amount of information that must be reviewed to make
an accurate diagnosis. Pathologists are responsible for reviewing all the biological tissues visible on a
slide. However, there can be many slides per patient, the equivalent of a thousand 10MP photos
where every pixel is important. Most biopsies are negative (e.g., 70% of breast biopsies are negative)
and would warrant consideration of the use of technology to evaluate images and tissue samples.
剩余10页未读,继续阅读
资源评论
百态老人
- 粉丝: 1641
- 资源: 2万+
下载权益
C知道特权
VIP文章
课程特权
开通VIP
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
- Python 手写实现 iD3 决策树算法-根据信息增益公式.zip
- 411675952289057车联助手-小窗版(三星)3.5.1.apk
- 三种快速排序方法合并在一个文件中以便直接运行的Python代码示例
- 937712277954201实习5.word
- 2程序语言基础知识pdf1_1716337722703.jpeg
- 简单的Python示例,演示了如何使用TCP/IP协议进行基本的客户端和服务器通信
- 考试.sql
- keil2 + proteus + 8051.exe
- 1961ee27df03bd4595d28e24b00dde4e_744c805f7e4fb4d40fa3f695bfbab035_8(1).c
- mediapipe-0.9.0.1-cp37-cp37m-win-amd64.whl.zip
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功