Modality-AGnostic Image Cascade (MAGIC) for Multi-Modality Cardiac Substructure Segmentation.

Area of Science:

• Medical Imaging
• Radiotherapy Planning
• Artificial Intelligence in Medicine

Background:

• Cardiac substructure segmentation is crucial for thoracic radiation therapy to prevent heart damage.
• Deep learning (DL) models offer efficiency but often lack generalizability across different imaging types and overlapping structures.

Purpose of the Study:

• To introduce and validate a Modality-AGnostic Image Cascade (MAGIC) for comprehensive cardiac substructure segmentation.
• To develop a DL model that is generalizable across multiple imaging modalities and handles overlapping structures.

Main Methods:

• MAGIC utilizes a U-shaped backbone with replicated encoding and decoding branches, based on nnU-Net.
• Trained and tested on 20 cardiac substructures from simulation CT, low-field MR-Linac, and CCTA modalities (n=111 total).
• Compared against 12 other models using Dice Similarity Coefficient (DSC) and statistical tests.

Main Results:

• MAGIC achieved average DSC scores of 0.75±0.16 for Sim-CT, 0.68±0.21 for MR-Linac, and 0.80±0.16 for CCTA.
• Outperformed comparison models in 57% of cases, showing strong performance across different cardiac substructures.
• Demonstrated effectiveness in segmenting multiple modalities and overlapping structures within a single model.

Conclusions:

• MAGIC provides an effective, accurate, and lightweight solution for multi-modal cardiac substructure segmentation.
• The model simplifies computational requirements, enhancing its clinical applicability and flexibility.
• MAGIC represents a significant advancement in streamlining segmentation processes for radiation therapy planning.