Bootstrapped Physically-Primed Neural Networks for Robust T2 Distribution Estimation in Low-SNR Pancreatic MRI
Authors: Hadas Ben Atya, Nicole Abramenkov, Noa Mashiah, Luise Brock, Daphna Link Sourani, Ram Weiss, Moti Freiman
Organizations: Faculty of Biomedical Engineering, Technion – Israel Institute of Technology, Haifa, Israel · Pattern Recognition Lab, Friedrich-Alexander-Universität, Erlangen, Germany. · The May-Blum-Dahl MRI Research Center, Faculty of Biomedical Engineering, Technion- IIT, Haifa, Israel · Department of Pediatrics A, Rambam Healthcare Campus, Haifa, Israel
Estimating multi-component T2 relaxation distributions from Multi-Echo Spin Echo (MESE) MRI is a severely ill-posed inverse problem, traditionally solved using regularized non-negative least squares (NNLS). In abdominal imaging, particularly the pancreas, low SNR and residual uncorrelated noise challenge classical solvers and deterministic deep learning models. We introduce a bootstrap-based inference framework for robust distributional T2 estimation that performs stochastic resampling of the echo train and aggregates predictions across multiple subsets. This treats the acquisition as a distribution rather than a fixed input, yielding variance-reduced, physically consistent estimates and converting deterministic relaxometry networks into probabilistic ensemble predictors. Applied to the P2T2 architecture, our method uses inference-time bootstrapping to smooth noise artifacts and enhance fidelity to the underlying relaxation distribution. Noninvasive pancreatic evaluation is limited by location and biopsy risks, highlighting the need for biomarkers capable of capturing early pathophysiological changes. In type 1 diabetes (T1DM), progressive beta-cell destruction begins years before overt hyperglycemia, yet current imaging cannot assess early islet decline. We evaluate clinical utility via a test-retest reproducibility study (N=7) and a T1DM versus healthy differentiation task (N=8). Our approach achieves the lowest Wasserstein distances across repeated scans and superior sensitivity to physiology-driven shifts in the relaxation-time distribution, outperforming NNLS and deterministic deep learning baselines. These results establish inference-time bootstrapping as an effective enhancement for quantitative T2 relaxometry in low-SNR abdominal imaging.
Figures & tables
Figure 1: Representative T2 distribution estimation workflow (Control Subject). Left: Axial MRI slice (TE=77.4 ms) with the pancreatic ROI centroid indicated. Center: Signal decay analysis: The model accurately fits the physical decay curve ( R2=0.986 ), matching the predicted signal (line) to the observed echoes (dots). Right: Regional T2 distribution analysis. Faint colored curves show the final bootstrapped T2 distributions for individual pixels within the ROI, illustrating local spatial heterogeneity. The solid black curve denotes the spatially averaged distribution across the ROI. Dashed lines depict the Gaussian decomposition of this average into Short- T2 (red) and Long- T2 (blue) components.
Figure 2: Comparison of the evaluated model architectures. The baseline MIML directly maps signals to T2 distributions; P2T2 incorporates the explicit TE encoding; and the proposed Bootstrapped-P2T2 adds inference-time resampling and ensemble averaging to improve robustness and stability.
Figure 3: Test–retest stability across anatomical regions. Boxplots show the Wasserstein distance ( W1 ) between repeated scans for the pancreatic body and tail. The proposed Bootstrapped-P2T2 model (red) exhibits consistently lower median error and reduced variance compared with the deterministic P2T2 (green) and baseline MIML (blue), demonstrating improved robustness and repeatability across both regions.
Model / Metric
AUC ↑
Hellinger ↑
KS p-value ↓
MIML
0.823
0.287
2.36×10−3
MIML Ensemble
0.794
0.246
8.30×10−3
P2T2
0.693
0.237
1.14×10−1
Bootstrapped-P2T2
0.800
0.427
4.79×10−4
Scalar T2 (Least Squares)
0.629
0.244
9.95×10−2
Scalar M0 (Least Squares)
0.632
0.124
1.37×10−1
Table 1: Evaluation of separation capability. Bootstrapped-P2T2 achieves the strongest statistical significance (lowest p-value) and separation (Hellinger), outperforming all baselines.
Figure 4: Sensitivity analysis across models (Pancreas Body & Tail). (a) The combined Wasserstein distance plot compares all distribution-based models. The Bootstrapped-P2T2 model (far right in a) shows the distinct separation between T1DM (orange) and Controls (blue), reducing the overlap seen in other models. (b–c) Classical scalar metrics ( T2 and M0 ) exhibit larger overlaps between groups, highlighting the improved sensitivity gained from the full-distribution modeling in (a).
Figure 5: Representative response to glucose challenge in the pancreatic tail, using the Bootstrapped-P2T2. Left: Healthy Control subject. Right: T1DM subject. Top Row: Axial MRI (10th echo) with the pancreatic tail segmentation overlaid in red. Middle Row: Average T2 distributions ( pˉ(T2) ) within the ROI before (blue) and after (orange) glucose intake. Bottom Row: Cumulative Distribution Functions (CDFs) highlighting the gap between timepoints. Note the significantly larger distributional shift in the T1DM subject compared to the stable profile of the control, indicating an altered microstructural response to metabolic stress.
Figure 6: Sensitivity of Test-Retest Reliability to Bootstrap Subset Size ( M ). Boxplots showing the Wasserstein distance ( W1 ) between repeated scans for varying subset sizes M∈{14,16,20,24} . The performance remains stable across all configurations, demonstrating that the method’s robustness is not sensitive to small variations in the resampling fraction.
Appendix figures & tables2 assets
Supplementary material from the paper’s appendix.
Appendix
Parameter
Value
Scanner Field Strength
3.0 T
Sequence Type
Radial Multi-Echo Spin Echo (MESE)
Number of Echoes
32
Echo Spacing ( Δ TE)
7.9 ms (Test-retest) / 7.74 ms (Diabetes analysis)
Field of View (FOV)
360×360 mm
Acquisition Matrix
208×208
Appendix
Table 2: Detailed acquisition parameters derived from the image metadata and protocol.
Component Label
Biophysical Proxy
T2 Mean Range (ms)
T2 Std Range (ms)
Myelin
Macromolecular / Fibrosis
[15,30]
[0.1,5]
Intra-Cellular (IS)
Parenchyma (Short)
[50,120]
[0.1,12]
Extra-Cellular (ES)
Parenchyma (Long)
[50,120]
[0.1,12]
Gray Matter (GM)
Edema / Inflammation
[60,300]
[0.1,12]
Pathology
Severe Edema / Lesions
[300,1000]
[0.1,5]
CSF
Fluid / Cysts / Ducts
[1000,2000]
[0.1,5]
Appendix
Table 3: Parameter ranges used for EPG-based synthetic training data generation. The network was trained on random mixtures of these components (using Dirichlet distributions) to learn the inverse mapping for any continuous T2 distribution within the 1−2000 ms range.
Automatic pancreas segmentation is fundamental to abdominal MRI analysis, yet deep learning models trained on one MRI sequence often fail catastrophically when applied to another-a challenge that has received little systematic investigation. We introduce CrossPan, a multi-institutional benchmark comprising 1,386 3D scans across three routinely acquired sequences (T1-weighted, T2-weighted, and Out-of-Phase) from eight centers. Our experiments reveal three key findings. First, cross-sequence domain shifts are far more severe than cross-center variability: models achieving Dice scores above 0.85 in-domain collapse to near-zero (<0.02) when transferred across sequences. Second, state-of-the-art domain generalization methods provide negligible benefit under these physics-driven contrast inversions, whereas foundation models like MedSAM2 maintain moderate zero-shot performance through contrast-invariant shape priors. Third, semi-supervised learning offers gains only under stable intensity distributions and becomes unstable on sequences with high intra-organ variability. These results establish cross-sequence generalization-not model architecture or center diversity-as the primary barrier to clinically deployable pancreas MRI segmentation. Dataset and code are available at https://crosspan.netlify.app/.
Linkai Peng, Cuiling Sun, Zheyuan Zhang +10
Department of Radiology, Northwestern University, Chicago, USA. · Department of Electrical and Computer Engineering, Northwestern University, Evanston, USA. · Department of Computer Science, Northwestern University, Evanston, USA. +4
Magnetic resonance imaging involves an inherent trade-off among spatial resolution, acquisition time, and noise. This trade-off contributes to long scan times and high cost. Deep learning has improved image denoising, but cardiac MRI remains difficult because high-resolution, rapid acquisitions generally lack corresponding low-noise ground truth. Self-supervised denoising offers a potential solution by learning from noisy image pairs or even single noisy acquisitions. However, we show that Noise2Void-style blind-spot denoising, which uses a mean squared error (MSE) loss and assumes zero-mean, independent and identically distributed (i.i.d.) noise, is poorly suited to MR magnitude images. When applied to short-axis T2-weighted and T1ρ-weighted cardiac MRI with synthetic Rician noise, it produces biased denoised images and biased parametric maps of T2 and T1ρ. To address this limitation, we formulate self-supervised denoising as maximum likelihood estimation under a known Rician noise model. This yields unbiased denoisers that are competitive with supervised baselines.
Slice-to-volume reconstruction (SVR) is the standard method for obtaining high-resolution (HR) 3D fetal brain volumes from motion-corrupted 2D MRI slice stacks acquired in multiple orientations. Existing SVR methods are optimized and validated only for clinical-range echo times (TEs), limiting their use at non-clinical TEs and making them incompatible with quantitative T2 mapping, a protocol- and center-independent biomarker of fetal brain maturation requiring HR reconstructions across multiple TEs. We present PRIME-SVR, the first implicit neural representation (INR) framework for joint HR reconstruction from multi-echo MRI. A single fully connected network models a continuous function from spatial coordinates to signal intensities across TEs, while a second network estimates slice-specific acquisition degradations. Cross-TE coherence is enforced via a Bloch equation-derived regularization penalizing deviations from expected T2 decay, with adaptive weighting that strengthens coupling for degraded stacks. The method is fully self-supervised. We validate PRIME-SVR on 39 in vivo fetal acquisitions (13 subjects x 3 TEs) from two centers, two vendors, and two field strengths (1.5 T and 0.55 T). Compared to state-of-the-art SVR, PRIME-SVR improves reconstruction sharpness by 47%, anatomical accuracy by 30%, and cross-TE structural consistency by 14%. It enables reconstruction at late TEs previously inaccessible to SVR, yielding the first 0.8 mm isotropic T2 maps at 0.55 T and the first T2 maps derived from INR-based SVR. PRIME-SVR also accelerates quantitative imaging by reducing the data needed for multi-TE reconstruction, cutting acquisition from 15 to 10 minutes while keeping T2 accuracy within 1.7% in white and deep gray matter, or to 5 minutes with a mean T2 error of 2.3% for high-quality acquisitions.
Busra Bulut, Maik Dannecker, Thomas Sanchez +14
Department of Medical Radiology, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland · CIBM Center for Biomedical Imaging, Lausanne, Switzerland · Chair for AI in Healthcare and Medicine, Technical University of Munich (TUM) and TUM University Hospital, Munich, Germany +6