Skip to Main Content

An official website of the United States government

Generalization of deep learning based processing pipelines for lung examinations

Principal Investigator
Name
Pedro Rodrigues
Degrees
Ph.D.
Institution
Philips Medical Systems Technologies, Ltd
Position Title
Clinical Scientist
Email
About this CDAS Project
Study
NLST (Learn more about this study)
Project ID
NLST-828
Initial CDAS Request Approval
Aug 6, 2021
Title
Generalization of deep learning based processing pipelines for lung examinations
Summary
Deep learning algorithms to assist in the reading of chest CT scans have been developed. In this study, we propose to study the generalization of previously developed deep learning algorithms designed for lung examinations (that exclude NLST datasets) to a large-scale, multi-year database. As part of this study, and in case of performance issues, we will investigate what will be the best strategy (cohort selection) for algorithm re-training. Up to three work packages are planned.
Aims

1) Evaluation and re-training of automatic quantification tools of lung diseases, including multi-time point follow-up scans.
2) Evaluation and re-training of automatic quantification tools for emphysema and fibrosis.
3) Evaluation and re-training of an automated nodule scoring risk method. Using a commercial release tool for lung nodule assessment, segment 3D nodules. 3D segmentations will be fed into a research feature extraction pipeline, combined with non-imaging patient data, and compared with the PanCan risk calculator versus survival data.
4) Evaluation and re-training of automated quantification of other pulmonary structures (e.g., airways).

Collaborators

Pedro Rodrigues, Philips Medical Systems Technologies, Ltd
Mark Rabotnikov, Philips Medical Systems Technologies, Ltd
Tobias Klinder, Philips Research Hamburg
Heike Carolus, Philips Research Hamburg
Rafael Wiemker, Philips Research Hamburg
Alexander Schmidt-Richberg, Philips Research Hamburg
Olivier Nempont, Philips Research France
Pascal Cathier, Philips Research France