• Modality-AGnostic image Cascad...

Treffer: Modality-AGnostic image Cascade (MAGIC) for multi-modality cardiac substructure segmentation.

Title:
Modality-AGnostic image Cascade (MAGIC) for multi-modality cardiac substructure segmentation.
Authors:
Summerfield N; Department of Human Oncology, University of Wisconsin-Madison, Madison, WI, USA; Department of Medical Physics, University of Wisconsin-Madison, Madison, WI, USA. Electronic address: nsummerfield@wisc.edu., He Q; Department of Computer Science, Wayne State University, Detroit, MI, USA. Electronic address: Qisheng.He@wayne.edu., Kuo A; Department of Human Oncology, University of Wisconsin-Madison, Madison, WI, USA; Department of Medical Physics, University of Wisconsin-Madison, Madison, WI, USA. Electronic address: askuo2@wisc.edu., Ruff C; Department of Human Oncology, University of Wisconsin-Madison, Madison, WI, USA; Department of Medical Physics, University of Wisconsin-Madison, Madison, WI, USA. Electronic address: cruff3@wisc.edu., Pan J; Department of Human Oncology, University of Wisconsin-Madison, Madison, WI, USA. Electronic address: joshuapan@berkeley.edu., Ghanem AI; Department of Radiation Oncology, Henry Ford Cancer Institute, Detroit, MI, USA; Alexandria Department of Clinical Oncology, Faculty of Medicine, Alexandria University, Alexandria, Egypt. Electronic address: AGHANEM1@hfhs.org., Zhu S; Department of Radiation Oncology, The Ohio State University, Columbus, OH, USA. Electronic address: Simeng.Zhu@osumc.edu., Kumar A; Department of Human Oncology, University of Wisconsin-Madison, Madison, WI, USA. Electronic address: kumar256@wisc.edu., Nagpal P; Department of Radiology, University of Wisconsin-Madison, Madison, WI, USA. Electronic address: pnagpal@wisc.edu., Zhao J; Department of Statistics, University of Wisconsin-Madison, Madison, WI, USA; Department of Biostatistics and Medical Informatics, University of Wisconsin-Madison, Madison, WI, USA. Electronic address: jiwei.zhao@wisc.edu., Dong M; Department of Computer Science, Wayne State University, Detroit, MI, USA. Electronic address: mdong@wayne.edu., Glide-Hurst C; Department of Human Oncology, University of Wisconsin-Madison, Madison, WI, USA; Department of Medical Physics, University of Wisconsin-Madison, Madison, WI, USA. Electronic address: glidehurst@humonc.wisc.edu.
Source:
Radiotherapy and oncology : journal of the European Society for Therapeutic Radiology and Oncology [Radiother Oncol] 2026 Jan; Vol. 214, pp. 111296. Date of Electronic Publication: 2025 Nov 19.
Publication Type:
Journal Article
Language:
English
Journal Info:
Publisher: Elsevier Scientific Publishers Country of Publication: Ireland NLM ID: 8407192 Publication Model: Print-Electronic Cited Medium: Internet ISSN: 1879-0887 (Electronic) Linking ISSN: 01678140 NLM ISO Abbreviation: Radiother Oncol Subsets: MEDLINE
Imprint Name(s):
Publication: Limerick : Elsevier Scientific Publishers
Original Publication: Amsterdam : Elsevier Science Publishers, c1983-
MeSH Terms:
Heart*/diagnostic imaging , Heart*/radiation effects , Deep Learning* , Radiotherapy Planning, Computer-Assisted*/methods, Humans ; Computed Tomography Angiography/methods ; Magnetic Resonance Imaging ; Tomography, X-Ray Computed ; Image Processing, Computer-Assisted/methods
Comments:
Update of: ArXiv. 2025 Jun 12:arXiv:2506.10797v1.. (PMID: 40980770)
Contributed Indexing:
Keywords: Auto-Segmentation; Computed Tomography Angiography; Convolutional Neural Networks; Deep Learning; Heart; Magnetic Resonance Imaging; Radiotherapy; Tomography, X-ray Computed
Entry Date(s):
Date Created: 20251121 Date Completed: 20251219 Latest Revision: 20251222
Update Code:
20251223
DOI:
10.1016/j.radonc.2025.111296
PMID:
41271169
Database:
MEDLINE

Weitere Informationen

Purpose: Cardiac substructure delineation is emerging in treatment planning to minimize the risk of radiation-induced heart disease. Deep learning offers efficient methods to reduce contouring burden but currently lacks generalizability across different modalities and overlapping structures. This work introduces and validates a Modality-AGnostic Image Cascade (MAGIC) deep-learning pipeline for comprehensive and multi-modal cardiac substructure segmentation.
Materials and Methods: MAGIC is implemented through replicated encoding and decoding branches of an nnU-Net backbone to handle multi-modality inputs and overlapping labels. First benchmarked on the multi-modality whole-heart segmentation (MMWHS) dataset including cardiac CT-angiography (CCTA) and MR modalities, twenty cardiac substructures (heart, chambers, great vessels (GVs), valves, coronary arteries (CAs), and conduction nodes) from clinical simulation CT (Sim-CT), low-field MR-Linac, and cardiac CT-angiography (CCTA) modalities were delineated to train semi-supervised (n = 151), validate (n = 15), and test (n = 30) MAGIC. For comparison, fourteen single-modality comparison models (two MMWHS modalities and four subgroups across three clinical modalities) were trained. Methods were evaluated for efficiency and against reference contours through the Dice similarity coefficient (DSC) and two-tailed Wilcoxon Signed-Rank test (p < 0.05).
Results: Average MMWHS DSC scores across CCTA and MR inputs were 0.88 ± 0.08 and 0.87 ± 0.04 respectively with significant improvement (p < 0.05) over unimodal baselines. Average 20-structure DSC scores were 0.75 ± 0.16 (heart, 0.96 ± 0.01; chambers, 0.89 ± 0.05; GVs, 0.81 ± 0.09; CAs, 0.60 ± 0.13; valves, 0.70 ± 0.18; nodes, 0.66 ± 0.12) for Sim-CT, 0.68 ± 0.21 (heart, 0.94 ± 0.01; chambers, 0.87 ± 0.05; GVs, 0.72 ± 0.18; CAs, 0.50 ± 0.18; valves, 0.62 ± 0.16; nodes, 0.52 ± 0.16) for MR-Linac, and 0.80 ± 0.16 (heart, 0.95 ± 0.01; chambers, 0.93 ± 0.04; GVs, 0.84 ± 0.06; CAs, 0.77 ± 0.12; valves, 0.68 ± 0.23; nodes, 0.72 ± 0.11) for CCTA. Furthermore, > 80% and > 70% reductions in training time and parameters were achieved, respectively.
Conclusions: MAGIC offers an efficient, lightweight solution capable of segmenting multiple image modalities and overlapping structures in a single model without compromising segmentation accuracy.
(Copyright © 2025. Published by Elsevier B.V.)

Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.