Deep Learning to Estimate RECIST in Patients with NSCLC Treated with PD-1 Blockade
Kathryn C Arbour1,2, Anh Tuan Luu3, Jia Luo1
1Thoracic Oncology Service, Memorial Sloan Kettering Cancer Center, New York, New York.
Cancer Discovery
|September 22, 2020
Summary
Deep learning models can now estimate cancer treatment outcomes from radiology reports. This approach enhances the analysis of real-world evidence (RWE) for large patient datasets, improving cancer research scalability and reliability.
Area of Science:
- Oncology
- Radiology
- Artificial Intelligence
- Machine Learning
Background:
- Real-world evidence (RWE) is crucial for discovery, reducing disparities, and regulatory approval in cancer research.
- Analyzing large RWE datasets is challenging due to the lack of scalable and reproducible outcome assessments.
- Machine learning can enhance the value of RWE by integrating and analyzing underutilized data.
Purpose of the Study:
- To develop and validate a deep-learning model capable of estimating objective outcomes from radiology text reports.
- To accurately assess treatment-specific outcomes, such as best overall response and progression-free survival, in non-small cell lung cancer patients.
- To enable scalable and reliable analysis of large real-world oncology clinical databases.
Main Methods:
- A deep-learning model was trained using radiology text reports from non-small cell lung cancer patients treated with PD-1 blockade.
- The model was developed and validated on a training cohort and two independent test cohorts.
- The model estimated gold-standard RECIST-defined outcomes, including best overall response and progression-free survival.
Main Results:
- The deep-learning model accurately estimated best overall response in patients with non-small cell lung cancer.
- The model also accurately estimated progression-free survival using radiology text reports.
- The developed model demonstrated the potential for accurate outcome estimation at scale.
Conclusions:
- A validated deep-learning model can reliably estimate objective response categories from radiology text reports.
- This tool facilitates large-scale analysis of real-world oncology data, overcoming current limitations in outcome assessment.
- The model offers a scalable solution for determining objective outcome metrics in clinical trial assessments and real-world oncology research.
