Oops, you're using an old version of your browser so some of the features on this page may not be displaying properly.

MINIMAL Requirements: Google Chrome 24+Mozilla Firefox 20+Internet Explorer 11Opera 15–18Apple Safari 7SeaMonkey 2.15-2.23

Poster Display session

393P - Integrating deep learning-based dose distribution prediction with Bayesian networks for decision support in respiratory motion controlled radiotherapy

Date

27 Jun 2024

Session

Poster Display session

Presenters

DONG-YUN KIM

Citation

Annals of Oncology (2024) 35 (suppl_1): S119-S161. 10.1016/annonc/annonc1481

Authors

D. KIM1, B. Jang2, E. Kim3, E.K. Chie4

Author affiliations

  • 1 Chung-Ang University Hospital, Seoul/KR
  • 2 Seoul National University Hospital, Seoul/KR
  • 3 SMG-SNU Boramae Medical Center, Seoul/KR
  • 4 SNUH - Seoul National University Hospital, Seoul/KR

Resources

Login to get immediate access to this content.

If you do not have an ESMO account, please create one for free.

Abstract 393P

Background

Selecting the better of the two institutionally available techniques to harbor optimal motion management, either a high dose rate stereotactic linear accelerator delivery using TrueBeam (TBX) or Magnetic Resonance (MR)-guided gated delivery using MRIdian (MRG), can be time-consuming and costly as independent treatment plan generation is a prerequisite. To address this challenge, we aimed to develop a decision-supporting algorithm that predicts the optimal treatment modality based on a combination of deep learning-generated dose distributions and relevant clinical data.

Methods

Clinical data and images were collected for 65 patients diagnosed with primary liver cancer (N=40) and pancreatic cancer (N=25) to facilitate the development of a predictive model for RT dose distribution. Treatment plans were created for both TBX and MRG. A three-dimensional U-Net deep learning approach was employed to predict RT dose distribution and generate dose volume histograms (DVHs). Generated DVH profiles were included in the Bayesian network (BN) decision-supporting algorithm by incorporating clinical data to nominate superior treatment plan.

Results

The root-mean-square error (RMSE) of voxels between the ground truth and the predictive dose from developed model was 3.04 (with a range of 2.29 to 3.79). The RMSE for the MRG prediction model was lower than that of the TBX. In general, the predictive normalized DVH metrics demonstrated superior performance in the MRG prediction models. Then, a BN model was established for determining the optimal treatment plan. The BN structure incorporated factors such as age, tumor volume, tumor to hollow viscus relationship as abutting vs apart, and the normalized value of the maximum dose to the neighboring dose limiting organ for MRG. The overall area under the receiver operating characteristic curve was 0.84.

Conclusions

We demonstrated a decision-supporting algorithm that could provide guidance to clinicians to select the superior plan from two competing RT systems for upper GI malignancy. This is anticipated to save resources in terms of manpower, time, and cost for both care-provider and the patient.

Legal entity responsible for the study

The authors.

Funding

Ministry of Science and ICT, Republic of Korea.

Disclosure

All authors have declared no conflicts of interest.

This site uses cookies. Some of these cookies are essential, while others help us improve your experience by providing insights into how the site is being used.

For more detailed information on the cookies we use, please check our Privacy Policy.

Customise settings
  • Necessary cookies enable core functionality. The website cannot function properly without these cookies, and you can only disable them by changing your browser preferences.