Multi-task weak supervision enables anatomically-resolved abnormality detection in whole-body FDG-PET/CT.
Multi-task weak supervision enables anatomically-resolved abnormality detection in whole-body FDG-PET/CT.
复制标题
DOI：
10.1038/s41467-021-22018-1
复制
发表时间：
2021-03-25
影响因子：
16.6
通讯作者：
Lungren MP
中科院分区：
文献类型：
--
作者：
Eyuboglu S;Angus G;Patel BN;Pareek A;Davidzon G;Long J;Dunnmon J;Lungren MP
计算决策支持系统可以在全身 FDG-PET/CT 工作流程中提供临床价值。然而，标记数据的可用性有限，加上 PET/CT 成像检查的规模较大，使得应用现有的监督机器学习系统面临挑战。利用自然语言处理的最新进展，我们描述了一个弱监督框架，该框架从自由文本放射学报告中提取不完美但高度精细的区域异常标签。我们的框架会自动在解剖区域的自定义本体中标记每个区域，从而在每次成像检查中提供病理的结构化轮廓。然后，我们使用这些生成的标签来训练基于注意力的多任务 CNN 架构，以检测和估计全身扫描中异常的位置。我们凭经验证明，我们的多任务表示对于训练数据有限的罕见异常情况的出色表现至关重要。该表示还有助于根据成像数据进行更准确的死亡率预测，这表明我们的框架除了异常检测和位置估计之外还有潜在的用途。计算决策支持系统可以在全身 FDG PET/CT 工作流程中提供临床价值，但标记数据稀缺，PET/CT 成像检查很麻烦。在这里，作者描述了一个弱监督框架，该框架从自由文本放射学报告中提取区域异常标签。
Computational decision support systems could provide clinical value in whole-body FDG-PET/CT workflows. However, limited availability of labeled data combined with the large size of PET/CT imaging exams make it challenging to apply existing supervised machine learning systems. Leveraging recent advancements in natural language processing, we describe a weak supervision framework that extracts imperfect, yet highly granular, regional abnormality labels from free-text radiology reports. Our framework automatically labels each region in a custom ontology of anatomical regions, providing a structured profile of the pathologies in each imaging exam. Using these generated labels, we then train an attention-based, multi-task CNN architecture to detect and estimate the location of abnormalities in whole-body scans. We demonstrate empirically that our multi-task representation is critical for strong performance on rare abnormalities with limited training data. The representation also contributes to more accurate mortality prediction from imaging data, suggesting the potential utility of our framework beyond abnormality detection and location estimation. Computational decision support systems could provide clinical value in whole-body FDG PET/CT workflows, but labeled data is scarce and PET/CT imaging exams are cumbersome. Here, the authors describe a weak supervision framework that extracts regional abnormality labels from free-text radiology reports.
登录
查看更多内容
影响因子：
16.6
作者：
Fries, Jason A.;Varma, Paroma;Priest, James R.
通讯作者：
Priest, James R.
DOI：
10.1145/3368555.3384468
发表时间：
2020-04
期刊：
Proceedings of the ACM Conference on Health, Inference, and Learning
影响因子：
--
作者：
Oakden-Rayner L;Dunnmon J;Carneiro G;Ré C
通讯作者：
Ré C
影响因子：
9.3
作者：
Niederkohr, Ryan D.;Greenspan, Bennett S.;Rohren, Eric M.
通讯作者：
Rohren, Eric M.
影响因子：
64.8
作者：
Esteva A;Kuprel B;Novoa RA;Ko J;Swetter SM;Blau HM;Thrun S
通讯作者：
Thrun S
影响因子：
82.9
作者：
Titano, Joseph J.;Badgeley, Marcus;Oermann, Eric K.
通讯作者：
Oermann, Eric K.