Enhancing lesion segmentation algorithm to include lung nodules
Principal Investigator
Name
Andre Neubauer
Degrees
M.Sc. Ph.D.
Institution
AGFA HealthCare NV
Position Title
Researcher and Developer
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-1444
Initial CDAS Request Approval
Jul 14, 2025
Title
Enhancing lesion segmentation algorithm to include lung nodules
Summary
Accurate lesion segmentation is a foundational component in effective lung cancer screening, diagnosis, and longitudinal management. AGFA HealthCare’s Enterprise Imaging platform supports clinicians with advanced tools for visualizing and assessing lesions in medical imaging. This project proposes to further develop the AGFA HealthCare lesion segmentation algorithm, with a particular focus on improving its performance on complex and clinically significant lesion types to include lung nodules.
Lung nodules, especially in early or subtle forms, are challenging to delineate and track over time. However, they are critical indicators of potential malignancy and require precise segmentation to support volume assessment, doubling time calculation, and treatment planning. The extension and validation of the algorithm on lung nodule lesions will significantly improve its clinical applicability. Utilizing the National Lung Screening Trial (NLST) dataset, which offers a rich and diverse collection of CT scans with associated radiological and pathological data, the project aims to train and evaluate a machine learning model for robust and reproducible segmentation across lesion subtypes. The long-term goal is to provide clinicians with reliable, efficient tools that support early detection, diagnosis, and consistent longitudinal follow-up of lung abnormalities.
Lung nodules, especially in early or subtle forms, are challenging to delineate and track over time. However, they are critical indicators of potential malignancy and require precise segmentation to support volume assessment, doubling time calculation, and treatment planning. The extension and validation of the algorithm on lung nodule lesions will significantly improve its clinical applicability. Utilizing the National Lung Screening Trial (NLST) dataset, which offers a rich and diverse collection of CT scans with associated radiological and pathological data, the project aims to train and evaluate a machine learning model for robust and reproducible segmentation across lesion subtypes. The long-term goal is to provide clinicians with reliable, efficient tools that support early detection, diagnosis, and consistent longitudinal follow-up of lung abnormalities.
Aims
1) To enhance the lesion segmentation algorithm to include lung nodules with high accuracy.
2) To evaluate the effectiveness of the enhanced lesion segmentation algorithm in segmenting various types of lung nodules using CT data on nodules within the NLST data.
Collaborators
Katja Bueler VRVis GmbH
David Major, MSc VRVis GmbH
Dimitrios Lenis, BSc VRVis GmbH
Maria Wimmer, PhD VRVis GmbH
Philip Winter, MSc VRVis GmbH