Multi-scale Cascaded Foundation Model for Whole-body Organs-at-risk Segmentation

The introduction of the Multi-scale Cascaded Fusion Network (MCFNet) marks a significant advancement in the segmentation of organs-at-risk (OARs), crucial for improving the safety and precision of radiotherapy and surgery. By utilizing a Sharp Extraction Backbone and a Flexible Connection Backbone, MCFNet effectively aggregates features across multiple scales, enhancing boundary localization and preserving fine structures. This innovative approach has been validated through experiments involving 36,131 image-mask pairs from 671 patients across 10 datasets, showcasing consistent robustness and strong cross-dataset generalization. The model outperforms existing methods, providing reliable image-guided support for computer-aided diagnosis and personalized treatment. The adaptive loss-aggregation strategy further stabilizes optimization, yielding gains in accuracy and training efficiency. With its code available on GitHub, MCFNet not only contributes to the field of medical imaging but als…