Personalized deep learning auto-segmentation models for adaptive fractionated magnetic resonance-guided radiation therapy of the abdomen

Kawula M, Marschner S, Wei C, Ribeiro MF, Corradini S, Belka C, Landry G, Kurz C (2025)

Publication Type: Journal article

Publication year: 2025

Journal

Medical Physics Wiley

Book Volume: 52

Pages Range: 2295-2304

Journal Issue: 4

DOI: 10.1002/mp.17580

Abstract

Background: Manual contour corrections during fractionated magnetic resonance (MR)-guided radiotherapy (MRgRT) are time-consuming. Conventional population models for deep learning auto-segmentation might be suboptimal for MRgRT at MR-Linacs since they do not incorporate manual segmentation from treatment planning and previous fractions. Purpose: In this work, we investigate patient-specific (PS) auto-segmentation methods leveraging expert-segmented planning and prior fraction MR images (MRIs) to improve auto-segmentation on consecutive treatment days. Materials and Methods: Data from 151 abdominal cancer patients treated at a 0.35 T MR-Linac (151 planning and 215 fraction MRIs) were included. Population baseline models (BMs) were trained on 107 planning MRIs for one-class segmentation of the aorta, bowel, duodenum, kidneys, liver, spinal canal, and stomach. PS models were obtained by fine-tuning the BMs using the planning MRI ((Formula presented.)). Maximal improvement by continuously updating the PS models was investigated by adding the first four out of five fraction MRIs ((Formula presented.)). Similarly, PS models without BM were trained ((Formula presented.) and (Formula presented.)). All hyperparameters were optimized using 23 patients, and the methods were tested on the remaining 21 patients. Evaluation involved Dice similarity coefficient (DSC), average ((Formula presented.)) and the 95th percentile (HD95) Hausdorff distance. A qualitative contour assessment by a radiation oncologist was performed for BM, (Formula presented.), and (Formula presented.). Results: (Formula presented.) and (Formula presented.) networks had the best geometric performance. (Formula presented.) and BMs showed similar DSC and HDs values, however (Formula presented.) models outperformed BMs. (Formula presented.) predictions scored the best in the qualitative evaluation, followed by the BMs and (Formula presented.) models. Conclusion: Personalized auto-segmentation models outperformed the population BMs. In most cases, (Formula presented.) delineations were judged to be directly usable for treatment adaptation without further corrections, suggesting a potential time saving during fractionated treatment.

Involved external institutions

Klinikum der Universität München (LMU Klinikum) DE Germany (DE)

How to cite

APA:

Kawula, M., Marschner, S., Wei, C., Ribeiro, M.F., Corradini, S., Belka, C.,... Kurz, C. (2025). Personalized deep learning auto-segmentation models for adaptive fractionated magnetic resonance-guided radiation therapy of the abdomen. Medical Physics, 52(4), 2295-2304. https://doi.org/10.1002/mp.17580

MLA:

Kawula, Maria, et al. "Personalized deep learning auto-segmentation models for adaptive fractionated magnetic resonance-guided radiation therapy of the abdomen." Medical Physics 52.4 (2025): 2295-2304.

BibTeX: Download