Skip to Main Content

An official website of the United States government

Principal Investigator
Name
Veysel Kocaman
Degrees
Ph.D.
Institution
Gesund AI
Position Title
VP of Engineering
Email
About this CDAS Project
Study
NLST (Learn more about this study)
Project ID
NLST-1105
Initial CDAS Request Approval
Aug 8, 2023
Title
Automate processing multimodal Cancer imaging data for clinical validation process of AI products
Summary
In this project, Gesund.ai aims to automate the processing of multimodal Cancer imaging data for the clinical validation process of AI products. The primary objective is to leverage the National Lung Screening Trial (NLST) dataset provided by the National Institutes of Health (NIH) to curate an appropriate dataset for model validation and development. Gesund.ai's platform will utilize this high-quality, diverse clinical data to assess and validate AI models designed for lung cancer detection and screening.

The project will follow a federated validation approach, where the model owner shares their AI model with Gesund.ai for evaluation. Gesund.ai's federated validation platform resides on hospital premises or a private cloud, ensuring data privacy and security. The model will be tested against a previously unseen validation dataset, which will be curated using the NLST data and other relevant multimodal imaging sources.

Through this process, Gesund.ai will evaluate the model's accuracy metrics and performance, taking into account patient characteristics, scenario analyses, and stress testing. The platform will display the results, allowing further examination and refinement of the AI model's capabilities.

The ultimate goal of this project is to provide a standardized, unified, and diversified dataset that meets both ML needs and regulatory requirements. The curated dataset will enable rigorous validation of AI models for lung cancer detection, enhancing their reliability and efficacy in clinical settings. The insights generated from the model's evaluation will be exported into a comprehensive report, which the model owner can use to supplement their regulatory submission for approval and adoption in medical practice.
Aims

- Utilize the NLST dataset and other relevant multimodal imaging data to curate a comprehensive and diverse validation dataset for AI models in lung cancer detection.
- Develop an automated data processing pipeline that ensures the integrity and privacy of patient information while facilitating the model validation process.
- Implement stress testing scenarios and scenario analyses to evaluate the robustness and generalizability of AI models under different clinical conditions.
- Assess model accuracy metrics, including sensitivity, specificity, precision, and recall, to determine the performance of AI models in lung cancer detection.
- Provide a user-friendly interface on the Gesund.ai platform to display and analyze the model validation results, allowing model owners to gain valuable insights into their AI products' capabilities.

Collaborators

Gesund.ai: Gesund.ai is the primary organization leading this project and providing the AI model validation platform.
National Institutes of Health (NIH): NIH is the data provider, granting access to the National Lung Screening Trial (NLST) dataset for this project.
Lung Screening Study group (LSS) and American College of Radiology Imaging Network (ACRIN): These organizations conducted the NLST trial and contributed to data collection, making their collaboration essential for successful dataset curation and validation.