Quality-aware AI segmentation for breast tumors in DCE-MRI, solving multi-center data variability for clinical deployment
Moving beyond academic benchmarks to clinical readiness with a practical and validated solution
Automated quality assessment and filtering ensures reliable performance across institutions without retraining
Integrates information across multiple DCE-MRI phases (0000-0002) for dynamic contrast enhancement
Validated across four independent datasets ensuring reproducible and scientifically sound results
Directly scalable across hospital networks for personalized treatment planning and therapy monitoring
Addressing the critical bottleneck in breast oncology imaging
Manual tumor segmentation from MRI scans creates significant bottlenecks in clinical workflow, delaying diagnosis and treatment
Differences in scanners, imaging protocols, and image quality across hospitals cause current AI models to fail
Our analysis revealed 980 out of 1,506 cases had significant quality issues like motion artifacts and reduced contrast
Selective Phase-Aware Training Framework built on robust nnU-Net architecture
Selectively training on 247 high-quality cases yields superior performance (0.72 validation Dice score) compared to using all 1,200+ available cases. We prove that naive dataset expansion can be counterproductive.
By integrating information across multiple DCE-MRI phases, our model captures dynamic contrast enhancement patterns, resulting in a 22% performance improvement over single-phase methods.
Built upon the industry-standard robust nnU-Net framework
Quality assessment and filtering system for consistent performance
Reliable across institutions without the need for model retraining
Validated performance metrics across multiple datasets
Rigorous testing ensuring reproducible and scientifically sound results
Our work has been peer-reviewed and accepted for poster presentation at MICCAI 2025, one of the most prestigious medical imaging conferences worldwide.
Rigorously tested on four independent datasets: DUKE, NACT, ISPY1, and ISPY2, ensuring comprehensive validation across diverse imaging scenarios.
Breast cancer is the most common cancer in women globally, and DCE-MRI is a standard diagnostic tool. Our solution is directly scalable across hospital networks and crucial for enabling personalized treatment planning and objective therapy response monitoring.
Join us in advancing breast cancer diagnosis and treatment with AI-powered precision
Download Research