Page 182 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

Use Case 27: Multi-Cancer Early Screening with Non-contrast CT

and Deep Learning�

Organization: Alibaba Damo (Beijing) Technology Co., Ltd.

Country: China
Contact Person:

Primary: Yi Chen, elaine.cy@ alibaba -inc .com
Secondary: Jianfei Guo, guojianfei.gjf@ alibaba -inc .com

1 Use Case Summary Table

Item Details
Category Artificial Intelligence, Healthcare

Problem Addressed Cancer is a worldwide leading cause of death. Early or incidental cancer
detection is associated with prolonged survival and improved patient
outcomes. However, early screening of asymptomatic individuals for a
wide range of cancers using a single test remains unfeasible due to the
low prevalence and potential harms of false positives. Despite the fact that
non-contrast computed tomography (CT) offers the potential for large-
scale screening, the identification of multi-cancer using non-contrast
CT has long been considered impossible, as the low contrast of tumors
renders them indistinguishable in non-contrast CT images. Therefore,
a deep learning-based early cancer screening solution is introduced
to detect and diagnose early multi-cancer from chest and abdominal
non-contrast CT, enabling timely treatment with the intent to cure.

Key Aspects of Solu- The deep learning-based early cancer screening solution leverages
tion deep learning models to detect and diagnose cancers for eight chest
and abdominal organs, including lung, breast, pancreas, stomach, kidney,
esophagus, liver and colon, on non-contrast CT scans. Collaborated with
clinical experts, we develop deep learning models can help to diagnose
multi-cancer with high accuracy and can be readily utilized for oppor-
tunistic screening in large-scale asymptomatic patient populations in a
fully automatic manner, resulting in safe, low-cost, and effective detec-
tion of early-stage malignancies missed by standard-of-care diagnostic
techniques and circumventing the toxicity of contrast agent. The deep
learning-based early cancer screening solution can help radiologists and
medical experts with fast and accurate organ localization, lesion detection
and advanced differential diagnosis of lesions, showing great potential for
accurate detection of multi-cancer for which no guideline-recommended
screening tests are available for average-risk individuals and opening
up an exciting possibility of non-invasive, low-cost and universal early
multi-cancer detection at both high sensitivity and high specificity levels
in which cancers can be detected before symptoms appear.

146