Trustworthy Deep Learning Models for Calcification Assessment in Radiotherapy CT using NLST Data

Principal Investigator

Name
Ivana Isgum

Degrees
PhD

Institution
Amsterdam University Medical Center

Position Title
Full Professor

Email
i.isgum@amsterdamumc.nl

About this CDAS Project

Study
NLST (Learn more about this study)

Project ID
NLST-1362

Initial CDAS Request Approval
Dec 4, 2024

Title
Trustworthy Deep Learning Models for Calcification Assessment in Radiotherapy CT using NLST Data

Summary
It has been shown that the amount of coronary artery calcifications seen on radiotherapy planning CT scans in breast cancer patients is strongly associated with CVD risk. While multiple deep learning methods for automatic calcium scoring have been developed, their performance in radiotherapy planning CT scans of breast cancer patients is limited by image quality and the low prevalence of calcifications in this data. We propose to leverage the NLST CT scans to develop accurate deep learning method(s) and assess their generalization from NLST data to diverse radiotherapy planning CT scans without contrast enhancement for the purposes of the ARTILLERY project (https://artillery-project.eu/). The NLST cohort, with its high prevalence of atherosclerotic calcifications in high-risk participants, provides an excellent opportunity to train robust models that could be fine-tuned for application in radiotherapy planning CT scans.

Aims

Aims:
(1) To develop and train deep learning models for calcium scoring using the NLST dataset, leveraging its high prevalence of calcifications for accurate detection in radiotherapy planning CT scans;
(2) To fine-tune these pre-trained models for application in radiotherapy planning CT scans, incorporating methods to handle differences in image acquisition parameters and quality;
(3) To validate the adapted models' performance in predicting cardiovascular risk in breast cancer survivors;
(4) To evaluate the performance of the developed deep learning model(s) in clinical settings in multiple hospitals by sharing a trained model.

Collaborators

Ivana Išgum, Full Professor, Amsterdam University Medical Center, Amsterdam, The Netherlands, i.isgum@amsterdamumc.nl

Sanyog Vyawahare, PhD student, Amsterdam University Medical Center, Amsterdam, The Netherlands, s.vyawahare@amsterdamumc.nl

Dimitrios Karkalousos, Postdoc, Amsterdam University Medical Center, Amsterdam, The Netherlands, d.karkalousos@amsterdamumc.nl