Field strength-dependent performance variability in deep learning-based analysis of magnetic resonance imaging
🤖 AI Summary
Magnetic resonance imaging (MRI) field strength constitutes a critical confounding factor in deep learning–based medical image segmentation, yet its systematic impact on model performance and generalizability remains poorly quantified.
Method: We systematically evaluated the influence of 1.5T versus 3.0T MRI field strength on deep learning segmentation using three public datasets—breast tumor, pancreas, and cervical spinal cord. We trained nnU-Net models under field-strength–specific (1.5T-only, 3.0T-only) and mixed-field conditions, and employed UMAP dimensionality reduction alongside a 23-feature radiomics analysis to characterize field-strength dependence.
Contribution/Results: We provide the first quantitative evidence that field strength is a key confounder for soft-tissue segmentation. Counterintuitively, 3.0T-only models significantly outperformed mixed-field models across both 1.5T and 3.0T test sets (p < 0.0001), with Dice similarity coefficient (DSC) improvements of 0.494–0.840—challenging the prevailing assumption that data mixing inherently enhances generalizability. Cervical spinal cord segmentation exhibited high specificity but low transfer degradation (DSC > 0.92), revealing organ-specific heterogeneity in field-strength effects.
Application Category
📝 Abstract
This study quantitatively evaluates the impact of MRI scanner magnetic field strength on the performance and generalizability of deep learning-based segmentation algorithms. Three publicly available MRI datasets (breast tumor, pancreas, and cervical spine) were stratified by scanner field strength (1.5T vs. 3.0T). For each segmentation task, three nnU-Net-based models were developed: A model trained on 1.5T data only (m-1.5T), a model trained on 3.0T data only (m-3.0T), and a model trained on pooled 1.5T and 3.0T data (m-combined). Each model was evaluated on both 1.5T and 3.0T validation sets. Field-strength-dependent performance differences were investigated via Uniform Manifold Approximation and Projection (UMAP)-based clustering and radiomic analysis, including 23 first-order and texture features. For breast tumor segmentation, m-3.0T (DSC: 0.494 [1.5T] and 0.433 [3.0T]) significantly outperformed m-1.5T (DSC: 0.411 [1.5T] and 0.289 [3.0T]) and m-combined (DSC: 0.373 [1.5T] and 0.268[3.0T]) on both validation sets (p<0.0001). Pancreas segmentation showed similar trends: m-3.0T achieved the highest DSC (0.774 [1.5T], 0.840 [3.0T]), while m-1.5T underperformed significantly (p<0.0001). For cervical spine, models performed optimally on same-field validation sets with minimal cross-field performance degradation (DSC>0.92 for all comparisons). Radiomic analysis revealed moderate field-strength-dependent clustering in soft tissues (silhouette scores 0.23-0.29) but minimal separation in osseous structures (0.12). These results indicate that magnetic field strength in the training data substantially influences the performance of deep learning-based segmentation models, particularly for soft-tissue structures (e.g., small lesions). This warrants consideration of magnetic field strength as a confounding factor in studies evaluating AI performance on MRI.
Problem
Research questions and friction points this paper is trying to address.
Evaluates how MRI scanner field strength affects deep learning segmentation performance.
Investigates generalizability of models trained on different field strengths across datasets.
Assesses field strength as a confounding factor in AI-based MRI analysis.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Used nnU-Net models trained on different MRI field strengths
Applied UMAP clustering and radiomic analysis for evaluation
Assessed field strength impact on deep learning segmentation performance
🔎 Similar Papers
No similar papers found.
