Researchers evaluated the cross-domain generalization capabilities of deep learning models for kidney lesion segmentation on MRI scans. They trained a model on adult kidney MRI data and tested its performance on both adult lesions and pediatric Wilms tumors. The adult-trained model demonstrated promising results on pediatric Wilms tumors, achieving a median Dice Similarity Coefficient (DSC) of 0.85, only slightly lower than a pediatric-specific model which achieved 0.90. This suggests that adult datasets can provide valuable pre-training for pediatric applications, though dedicated pediatric data still offers superior performance. AI
IMPACT Demonstrates potential for leveraging larger adult medical datasets to improve pediatric diagnostic models, potentially accelerating research and clinical application.
RANK_REASON The item is a research paper detailing a study on cross-domain generalization of deep learning models for medical image segmentation. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- deep learning
- Dice Similarity Coefficient
- Hausdorff distance
- Hugging Face
- kidney cancer
- magnetic resonance imaging
- nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation
- pediatric oncology
- Renal lesions: characterization with diffusion-weighted imaging versus contrast-enhanced MR imaging
- Wilms tumors
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
