OpenMAP-HIE: Pathology Informed Score-based Diffusion Model for Unsupervised Neonatal Hypoxic-Ischemic Encephalopathy Lesion Segmentation

Pengyu Kan; Craig Jones; Kenichi Oishi

cs.CVOct 21, 2026

OpenMAP-HIE: Pathology Informed Score-based Diffusion Model for Unsupervised Neonatal Hypoxic-Ischemic Encephalopathy Lesion Segmentation

Authors: Pengyu Kan, Craig Jones, Kenichi Oishi

Organizations: Johns Hopkins University · Johns Hopkins University, University of Alberta · Johns Hopkins University School of Medicine

Abstract

Accurate localization of neonatal hypoxic--ischemic encephalopathy (HIE) lesions on apparent diffusion coefficient (ADC) images is clinically important but is limited by scarce expert annotations and substantial variation in lesion appearance. We propose OpenMAP-HIE, a pathology-informed score-based diffusion model for unsupervised HIE lesion localization without voxel-wise lesion supervision. A denoising diffusion model learns the healthy neonatal ADC distribution, while a ControlNet-based conditional branch preserves patient-specific anatomy. At inference, anomaly evidence is derived directly from differences between conditional and unconditioned noise predictions at independently evaluated diffusion noise levels, avoiding dependence on a final pseudo-healthy reconstruction or sequential reverse-diffusion chain. Based on the characteristic ADC reduction associated with acute HIE, only discrepancies consistent with the expected pathological direction are retained and aggregated across noise levels using a geometric mean. Spatial mean filtering further suppresses isolated responses. The framework was evaluated on the BONBID-HIE dataset. The ADC-only method achieved a Dice score of $0.283\pm0.244$ , average precision of $0.505$ , Hausdorff distance of $25.45\pm15.63$ mm, mean average surface distance of $9.40\pm8.49$ mm, and normalized surface distance of $0.355\pm0.236$ , providing the best overlap and surface-distance performance among the evaluated unsupervised methods and Dice performance comparable to supervised ADC-only models. Optional population-derived ADC $Z$ -map refinement further increased Dice to $0.602\pm0.269$ . These findings support pathology-informed score discrepancies as a useful signal for neonatal HIE lesion localization, while highlighting remaining challenges in detecting small and low-contrast lesions.

Figures & tables

Figure 1: Architecture of the anatomy-conditioned diffusion model. The healthy-prior U-Net is pretrained in stage 1 and frozen during stage 2. A trainable ControlNet encoder, initialized from the U-Net encoder, incorporates the spatial condition $\mathbf{c}$ alongside the noisy input $\mathbf{x}_{t}$ and timestep $t$ . Multi-scale ControlNet features are passed through zero-initialized $1\times 1$ convolutions and added to the U-Net skip features, enabling subject-specific conditioning of the noise prediction. Dashed gray lines denote U-Net skip connections, and orange arrows indicate ControlNet residual connections.

Panel of Figure 2 — Figure 2: Illustration of inference using pathology-informed noise-prediction discrepancy aggregation for (a) a normal slice and (b) a slice containing HIE lesions. The observed ADC image $\mathbf{x}_{0}$ is independently perturbed at multiple noise levels. At each level, the patient-conditioned (green dashed line) and healthy-prior (red dashed line) branches evaluate the same noisy input, and their predicted noise difference is retained only in the direction consistent with the reduced-ADC prior. Normal tissue is expected to exhibit relatively small discrepancies, whereas lesion regions may produce larger, directionally consistent responses across noise levels. The resulting discrepancy maps are aggregated using the geometric mean, $A_{\mathrm{GM}}$ , to obtain the anomaly map. Blue points denote noisy inputs, green and orange markers denote patient-conditioned and healthy-prior predictions, respectively, and red brackets indicate directional discrepancies. The curves provide a sequential forward-diffusion trajectory. The lesion ground truth annotation in (b) is shown for reference only.

Figure 3: Qualitative comparison of unsupervised anomaly detection methods in subjects with (a) small ( $<1\%$ ), (b) medium ( $1\%-5\%$ ), and (c) large ( $\geq 5\%$ ) lesion burdens, defined by the subject-level lesion percentage. Columns show the input ADC image, expert lesion annotation, and anomaly maps from selected top-performing baselines, including DAE ( Kascenas et al., 2022 ) , ANDi ( Frotscher et al., 2026 ) , MAD-AD ( Beizaee et al., 2025b ) , where the originally proposed smoothing methods from each paper have been applied. Our proposed method before and after mean filtering (MF; $k=5$ ) is highlighted in green box. Red overlays in the ground truth (GT) column indicate annotated lesions, and cyan contours delineate their boundaries on the anomaly maps. Brighter colors indicate higher anomaly scores.

Figure 4: Qualitative comparison of reconstruction-based anomaly detection and the proposed discrepancy-based approach. (a) Input ADC image and expert lesion annotation. (b–d) Reconstructed images and corresponding anomaly maps obtained using DDPM and DDPM with ControlNet at CFG scales of 0.8 and 7.5. (e) Anomaly maps produced by the proposed method before and after mean filtering (MF, $k=5$ ). (f) The corresponding population-derived Z-map for the input and the Z-map guided post-processing lesion refinement. Cyan contours delineate the expert-annotated lesion boundaries. All methods’ results have been applied with HIE pathological informed sign direction.

Method	Dice $\uparrow$	AP $\uparrow$	HD $\downarrow$ (mm)	MASD $\downarrow$ (mm)	NSD $\uparrow$
f-AnoGAN ( Schlegl et al., 2019 )	Unstable training and unable to achieve convergence.
DAE ( Kascenas et al., 2022 )	$0.159\pm 0.268^{\dagger\ddagger}$	$\mathbf{0.665}$	$30.38\pm 15.41$	$13.65\pm 9.87$	$0.176\pm 0.210$
AnoDDPM ( Wyatt et al., 2022 )	Unstable training with simplex noise.
AutoDDPM ( Bercea et al., 2023 )	$0.085\pm 0.126^{\dagger\ddagger}$	$0.083$	$31.96\pm 15.59$	$14.53\pm 9.73$	$0.169\pm 0.194$
pDDPM ( Behrendt et al., 2024 )	$0.121\pm 0.235^{\dagger\ddagger}$	$0.084$	$39.08\pm 15.05$	$21.16\pm 11.08$	$0.050\pm 0.111$
THOR ( Bercea et al., 2024 )	$0.162\pm 0.231^{\dagger\ddagger}$	$0.281$	$31.89\pm 15.56$	$14.35\pm 9.68$	$0.196\pm 0.218$

Table 1: Quantitative comparison of unsupervised anomaly detection methods for neonatal HIE lesion localization. Values are reported as mean

\pm

standard deviation where available. The best and second-best results are shown in bold and underlined , respectively. For each model’s result, the originally proposed smoothing methods have been applied. For Dice, † and ‡ indicate significantly lower scores than Ours and Ours with mean filtering (

k=5

), respectively (under two-sided paired Wilcoxon signed-rank tests, all

p<0.001

n=44

Lesion burden	Post-processing	Dice $\uparrow$	AP $\uparrow$	HD $\downarrow$ (mm)	MASD $\downarrow$ (mm)	NSD $\uparrow$
Small ( $<1\%$ ), $n=22$	$--$	$0.080\pm 0.086$	$0.030$	$40.27\pm 13.90$	$17.25\pm 9.58$	$0.155\pm 0.111$
Small ( $<1\%$ ), $n=22$	MF5	$\mathbf{0.141\pm 0.154}$	$\mathbf{0.135}$	$\mathbf{33.60\pm 15.19}$	$\mathbf{10.49\pm 5.64}$	$\mathbf{0.262\pm 0.208}$
Medium ( $1\%$ – $5\%$ ), $n=12$	$--$	$0.287\pm 0.148$	$0.186$	$26.39\pm 8.77$	$7.80\pm 3.53$	$0.388\pm 0.150$
Medium ( $1\%$ – $5\%$ ), $n=12$	MF5	$\mathbf{0.377\pm 0.208}$	$\mathbf{0.356}$	$\mathbf{18.43\pm 6.92}$	$\mathbf{5.47\pm 3.54}$	$\mathbf{0.487\pm 0.232}$
Large ( $\geq 5\%$ ), $n=10$	$--$	$\mathbf{0.603\pm 0.147}$	0.541	$10.23\pm 5.03$	$\mathbf{2.68\pm 1.59}$	$\mathbf{0.640\pm 0.160}$
Large ( $\geq 5\%$ ), $n=10$	MF5	$0.600\pm 0.167$	$\mathbf{0.562}$	$\mathbf{10.13\pm 4.75}$	$2.76\pm 0.93$	$0.582\pm 0.137$

Table 2: Performance of our proposed model stratified by subject-level lesion burden. Subjects were divided into small (

<1\%

), medium (

1\%

–

5\%

), and large (

\geq 5\%

) lesion-burden groups according to the percentage of voxels labeled as lesions in each subject. MF5 denotes mean filtering with a kernel size of

5

(rows shaded in gray). Results are reported as mean

\pm

standard deviation where available, and the best result within each lesion-burden group is highlighted in bold.

Model	Input or refinement	Dice $\uparrow$	AP $\uparrow$	HD $\downarrow$ (mm)	MASD $\downarrow$ (mm)	NSD $\uparrow$
Supervised segmentation methods
Swin-UNETR ( Almansour et al., 2025 )	ADC only	$0.311\pm 0.251^{\dagger}$	$0.645$	$22.56\pm 13.39$	$8.31\pm 8.38$	$0.406\pm 0.247$
	ADC + $Z$ -map	$0.553\pm 0.260^{\dagger}$	$0.918$	$\mathbf{16.07\pm 12.00}$	$4.10\pm 6.31$	$0.674\pm 0.256$
	ADC + $Z$ -map + $Z\leq-2.0$	$\mathbf{0.657\pm 0.258}$	$\mathbf{0.927}$	$18.32\pm 15.57$	$\mathbf{3.89\pm 5.46}$	$\mathbf{0.724\pm 0.242}$
UNet++ ( Rao et al., 2025 )	ADC only	$0.237\pm 0.260^{\dagger}$	$0.803$	$23.58\pm 12.80$	$9.68\pm 8.37$	$0.298\pm 0.235$
	ADC + $Z$ -map	$0.578\pm 0.246$	$\mathbf{0.915}$	$\mathbf{15.88\pm 12.77}$	$3.84\pm 5.55$	$0.694\pm 0.249$

Table 3: Effect of incorporating population-derived ADC

Z

-map information. ADC-only models serve as the reference configurations (rows shaded in gray). For the supervised models, the

Z

-map and a thresholded binary

Z

-mask (

Z\leq-2.0

) are provided as additional input channels ( Almansour et al., 2025 ; Bao et al., 2025 ; Rao et al., 2025 ) . For the population-derived statistical baseline,

-Z

is used as the continuous anomaly score for AP, whereas

Z\leq-2.0

is used to obtain the binary mask for Dice and surface-distance metrics. Our unsupervised method uses the

Z

-map only during post processing. AP is not reported for this post-processing result because it directly produces a binary lesion mask. The best result within each model is shown in bold . For Dice, † indicates a significantly lower score than our proposed method with

Z

-map post-processing (under two-sided paired Wilcoxon signed-rank test at significance level

\alpha=0.05

n=44

Configuration	Dice $\uparrow$	AP $\uparrow$	HD $\downarrow$ (mm)	MASD $\downarrow$ (mm)	NSD $\uparrow$
DDPM	$0.167\pm 0.236$	$0.235$	$32.03\pm 16.62$	$14.58\pm 10.03$	$0.212\pm 0.242$
$+$ ControlNet (healthy-restoration, CFG=0.8)	$0.121\pm 0.190$	$0.414$	$39.44\pm 16.03$	$17.47\pm 10.19$	$0.169\pm 0.193$
$+$ ControlNet (lesion-emphasizing, CFG=7.5)	$0.138\pm 0.242$	$0.332$	$31.30\pm 16.29$	$15.31\pm 10.37$	$0.160\pm 0.222$
$+$ Multi noise-level discrepancy accumulation	$0.232\pm 0.239$	$0.374$	$29.29\pm 16.43$	$12.62\pm 10.11$	$0.284\pm 0.256$
$+$ Mean filtering ( $k=5$ )	$\mathbf{0.283\pm 0.244}$	$\mathbf{0.505}$	$\mathbf{25.45\pm 15.63}$	$\mathbf{9.40\pm 8.49}$	$\mathbf{0.355\pm 0.236}$

Table 4: Ablation study of the proposed framework. All compared models applied the HIE pathology informed directional information. Results are reported as mean

\pm

standard deviation where applicable. The best and second-best results are shown in bold and underlined , respectively.

Kernel size $k$	Dice $\uparrow$	AP $\uparrow$	HD $\downarrow$ (mm)	MASD $\downarrow$ (mm)	NSD $\uparrow$
0	$0.232\pm 0.239$	$0.374$	$29.29\pm 16.43$	$12.62\pm 10.11$	$0.284\pm 0.256$
3	$0.262\pm 0.245$	$0.481$	$27.24\pm 16.37$	$11.01\pm 9.74$	$0.324\pm 0.250$
5	$0.283\pm 0.244$	$\mathbf{0.505}$	$\mathbf{25.45\pm 15.63}$	$\mathbf{9.40\pm 8.49}$	$\mathbf{0.355\pm 0.236}$
7	$\mathbf{0.287\pm 0.252}$	$0.503$	$26.20\pm 15.47$	$9.58\pm 8.43$	$0.348\pm 0.229$
8	$0.282\pm 0.255$	$0.497$	$26.51\pm 15.14$	$9.74\pm 8.38$	$0.331\pm 0.224$
10	$0.270\pm 0.259$	$0.485$	$28.16\pm 16.76$	$10.97\pm 9.52$	$0.304\pm 0.222$

Table 5: Ablation study of the mean-filter kernel size applied to the direction-aware anomaly map. Kernel size

k=0

indicates that no mean filtering was applied. The gray row denotes the kernel size used in the main experiments. Results are reported as mean

\pm

standard deviation where applicable, and the best value for each metric is highlighted in bold .

Score representation	Evaluation	Dice $\uparrow$	AP $\uparrow$	HD $\downarrow$ (mm)	MASD $\downarrow$ (mm)	NSD $\uparrow$
$\epsilon_{t}^{\text{cond}}-\epsilon_{t}^{\text{uncond}}$	Independent	$0.232\pm 0.239$	$0.374$	$29.29\pm 16.43$	$12.62\pm 10.11$	$0.284\pm 0.256$
$\epsilon_{t}^{\text{cond}}-\epsilon_{t}^{\text{uncond}}$ $+$ MF5	Independent	$\mathbf{0.283\pm 0.244}$	$\mathbf{0.505}$	$\mathbf{25.45\pm 15.63}$	$\mathbf{9.40\pm 8.49}$	$\mathbf{0.355\pm 0.236}$
$\epsilon_{t}^{\text{cond}}-\epsilon_{t}^{\text{uncond}}$	Reverse chain	$0.070\pm 0.119$	$0.031$	$32.05\pm 16.01$	$15.85\pm 10.36$	$0.170\pm 0.238$
$\epsilon_{t}^{\text{cond}}-\epsilon_{t}^{\text{uncond}}$ $+$ MF5	Reverse chain	$0.070\pm 0.108$	$0.036$	$49.59\pm 25.26$	$23.36\pm 15.44$	$0.114\pm 0.154$
$\epsilon_{t}^{\text{cond}}-\epsilon_{t}$ using forward noise	Independent	$0.045\pm 0.211$	$0.017$	$75.30\pm 9.29$	$52.69\pm 7.76$	$0.000\pm 0.000$

Table 6: Ablation of the anomaly-score representation compared with sequential reverse propagation and discrepancy to forward process noise. “Independent” indicates that each noise level was sampled directly from the original patient ADC image without propagating a reverse-diffusion chain, and “Reverse chain” evaluates successive states along a sequential reverse-diffusion trajectory (Eq. ( 43 )). The final row replaces the learned unconditional noise prediction with the Gaussian noise

\epsilon_{t}

sampled during forward corruption (Eq. ( 44 )). The biological directional constraint was enabled for all configurations. Rows shaded in gray use a mean filtering with kernel size of

5

(MF5). Values are reported as mean

\pm

standard deviation where available, and the best result for each metric is highlighted in bold .

Statistic	Lesion Mean (95% CI)	Background Mean (95% CI)
Positive discrepancy probability, $\Pr(d_{t}>0)$	$0.886$ $(0.884,\,0.888)$	$0.530$ $(0.526,\,0.533)$
Mean positive discrepancy, $\mathbb{E}[d_{t}\mid d_{t}>0]$	$0.266$ $(0.260,\,0.272)$	$0.155$ $(0.151,\,0.159)$
Mean negative discrepancy, $\mathbb{E}[d_{t}\mid d_{t}<0]$	$-0.107$ $(-0.110,\,-0.104)$	$-0.176$ $(-0.178,\,-0.174)$
Mean discrepancy magnitude, $\mathbb{E}[\|d_{t}\|]$	$0.248$ $(0.242,\,0.254)$	$0.165$ $(0.162,\,0.168)$

Table 7: Directional characteristics of the discrepancy across the selected noise levels. The discrepancy is defined as

d_{t}(\mathbf{v})=\epsilon_{t}^{\mathrm{cond}}(\mathbf{v})-\epsilon_{t}^{\mathrm{uncond}}(\mathbf{v})

. A positive value indicates that the healthy-prior response predicts a higher ADC value than the patient-conditioned response, consistent with the expected direction of reduced ADC in acute HIE injury. Values are reported as the timestep-wise mean and 95% confidence interval across the selected noise levels (

n=|\mathcal{T}|=125

HIE Pathology prior directional sign $\pm$	Post-processing	Dice $\uparrow$	AP $\uparrow$	HD $\downarrow$ (mm)	MASD $\downarrow$ (mm)	NSD $\uparrow$
No	$--$	$0.115\pm 0.148$	$0.222$	$33.99\pm 16.24$	$15.18\pm 10.04$	$0.189\pm 0.184$
No	MF5	$0.209\pm 0.221$	$0.490$	$30.16\pm 14.83$	$11.48\pm 8.82$	$0.266\pm 0.207$
Yes	$--$	$0.232\pm 0.239$	$0.374$	$29.29\pm 16.43$	$12.62\pm 10.11$	$0.284\pm 0.256$
Yes	MF5	$\mathbf{0.283\pm 0.244}$	$\mathbf{0.505}$	$\mathbf{25.45\pm 15.63}$	$\mathbf{9.40\pm 8.49}$	$\mathbf{0.355\pm 0.236}$

Table 8: Ablation of the pathology-informed directional sign constraint and mean-filter post-processing. MF5 denotes mean filtering with a kernel size of 5. Rows using MF5 are shaded, and the best result for each metric is shown in bold .

Model	Dice $\uparrow$	AP $\uparrow$	HD $\downarrow$ (mm)	MASD $\downarrow$ (mm)	NSD $\uparrow$
DAE ( Kascenas et al., 2022 )	$0.172\pm 0.272^{\dagger\ddagger}$	$\mathbf{0.680}$	$\mathbf{25.13\pm 13.11}$	$13.47\pm 9.56$	$0.188\pm 0.219$
AutoDDPM ( Bercea et al., 2023 )	$0.104\pm 0.135^{\dagger\ddagger}$	$0.119$	$30.78\pm 15.99$	$13.12\pm 9.59$	$0.209\pm 0.206$
pDDPM ( Behrendt et al., 2024 )	$0.121\pm 0.235^{\dagger\ddagger}$	$0.084$	$39.07\pm 15.07$	$21.17\pm 11.08$	$0.050\pm 0.111$
MAD-AD ( Beizaee et al., 2025b )	$\mathbf{0.173\pm 0.198}^{\dagger\ddagger}$	$0.491$	$27.44\pm 13.67$	$\mathbf{11.16\pm 7.99}$	$\mathbf{0.226\pm 0.192}$

Table 9: Quantitative comparison of unsupervised anomaly-detection baselines with introducing HIE pathology directional prior information at final reconstruction. Values are reported as mean

\pm

standard deviation where available. The best result for each metric is highlighted in bold . For each model’s result, the originally proposed smoothing methods have been applied. For Dice, † and ‡ indicate significantly lower scores than Ours and Ours with mean filtering (

k=5

), respectively (under two-sided paired Wilcoxon signed-rank tests, all

p<0.001

n=44

Medical Image Segmentation

Medical image segmentation delineates organs, lesions, and other structures in scans to support measurement, diagnosis, treatment planning, and research.

See all 786 papers

Oct 8, 2026

Expert-level detection of low-voltage intervals in EEG as a robust biomarker of neonatal encephalopathy

Background: Discontinuity in neonatal EEG is a critical marker of brain dysfunction but can be difficult to quantify visually. Inter-burst intervals (IBIs) are a key characteristic of discontinuity, yet their definition does not easily translate to AI detection. Objectives: To introduce low-voltage intervals (LVIs), defined exclusively by amplitude and duration, as a context-independent alternative to IBIs, develop an AI detection method for LVIs, and assess their clinical utility as a marker of brain injury. Methods: We developed a ConvNeXt model to detect LVIs and compared its performance with existing methods. We then tested the clinical relevance of detected LVIs across four applications: 1) grading background EEG abnormalities on an open benchmark dataset, 2) classifying clinical hypoxic-ischaemic encephalopathy (HIE) grades and comparing with a normal EEG dataset, 3) seizure prediction, and 4) association with seizure burden. Results: For training and validation, 654 hours of per-channel EEG from 123 neonates were reviewed and 51,162 LVIs were annotated. On held-out validation, ConvNeXt had an AUC of 0.996 and a Matthews correlation coefficient of 0.926. Model-reader agreement exceeded inter-reader agreement (

Δκ= 0.025

, 95% CI: 0.007 to 0.045,

p<0.001

), attaining expert-level equivalence. On a combined dataset of 444 neonates, percentage LVI showed a strong association with EEG grades (Spearman's

ρ=0.801

), a moderate association with HIE grades (

ρ=0.384

), a weaker association with seizure burden (

ρ=0.244

), and a strong seizure prediction capacity (AUC

=0.872

). Conclusion: AI detection of LVIs achieves expert-level equivalence, whereas quantitative features do not. Automated LVI quantification provides a robust clinical biomarker of neonatal encephalopathy associated with background EEG activity, HIE severity, and seizure risk.

John M. O'Toole, Sean R. Mathieson, Aurel Luca +4

CergenX Ltd, Cork, Ireland · INFANT Research Centre, University College Cork, Ireland · Department of Paediatrics and Child Health, University College Cork, Ireland

Electroencephalography

Ask PDF

May 25, 2026

HRVConformer: Neonatal Hypoxic-Ischemic Encephalopathy Classification from the Heart Rate signals

This paper presents the HRVConformer, a novel deep learning architecture for the classification of hypoxic-ischemic encephalopathy (HIE) using the instantaneous heart rate (HR) signal. Unlike conventional approaches that rely on handcrafted features, HRVConformer directly processes raw HR signals in an end-to-end manner, capturing both local and long-range dependencies through a hybrid Convolution-Transformer framework. By integrating convolutional layers for local feature extraction and Transformer-based attention mechanisms for global context modelling, the architecture effectively enhances signal representation and classification performance. The model was trained using supervised learning on a large HR dataset consisting of 1,573 one-hour epochs, including 259 one-hour expert-annotated epochs and a substantial set of weakly labelled data. A 314-hour validation set provided a robust performance estimation, while an independent 215-hour dataset with expert annotations was reserved for final testing. HR signals were extracted from electrocardiogram (ECG) recordings using an improved Pan-Tompkins algorithm, which significantly enhanced both signal quality and data availability. Experimental results demonstrate that the HRVConformer achieves an AUC of 83.23% and accuracy of 74.56% on the test set. These results surpass the performance of the Transformer, ResNet50 and fully convolutional networks baselines, highlighting the advantages of integrating convolutional and Transformer-based components for HR-based HIE classification. The proposed method provides a promising step toward a more accurate and automated assessment of HIE using HR signals. The code is available at: https://github.com/syu-kylin/HRVConformer.

Shuwen Yu, William P Marnane, Geraldine B. Boylan +1

Department of Electrical & Electronic Engineering, School of Engineering and Architecture, University College Cork, Cork, Ireland · INFANT Research Centre, University College Cork, Cork, Ireland · Pediatrics and Child Health, University College Cork, Cork, Ireland

Hybrid CNN-Transformer Architectures

Ask PDF

Aug 26, 2026

Deep Learning Segmentation of Diffusion-Weighted MRI Acute Ischaemic Stroke: A Pragmatic Evaluation Across Three Datasets

Objective: Diffusion-weighted MRI (DWI-MRI) is the gold standard for visualizing and quantifying acute ischaemic stroke (AIS). Although deep learning methods can accurately segment AIS lesions, the optimal image inputs and model architecture remain uncertain. We evaluated whether accurate AIS lesion segmentation can be achieved using a pragmatic deep learning approach with minimal preprocessing and clinically feasible inference times. Materials and Methods: Self-configured nnU-Net models were trained on 1,744 DWI cases from local, national, and open-access datasets and tested on 436 cases. Four experimental conditions were evaluated using five-fold cross-validation: with or without brain extraction and using either DWI alone or DWI plus apparent diffusion coefficient (ADC) images as inputs. Two architectures were compared: the baseline nnU-Net (base) and a residual encoder nnU-Net (ResEnc). Performance was benchmarked against the DeepISLES ensemble model from the 2022 ISLES challenge. Results: In the test set (n=436), the base model achieved a median (IQR) Dice similarity coefficient (DSC) of 0.84 (0.19). For the base model, only two of six pairwise comparisons between input configurations showed significant differences. ResEnc produced small but significant improvements in DSC compared with the base model for DWI, DWI+brain extraction, and DWI+ADC inputs (all p<0.02), but not for DWI+ADC+brain extraction (p>0.50). The base model significantly outperformed DeepISLES, particularly in patients with smaller infarct volumes (signed-rank test, p<0.01). Conclusions: A baseline nnU-Net trained on DWI alone, without preprocessing, enabled fast and accurate AIS lesion segmentation. This streamlined approach may facilitate clinical research and support acute stroke imaging workflows

Atle Bjørnerud, Till Schellhorn, Thor H. Skattør +4

Computational Radiology & Artificial Intelligence Unit, Department of Physics and Computational Radiology, Oslo University Hospital, Oslo, Norway · Center for Lifespan Changes in Brain and Cognition, Department of Psychology, University of Oslo, Oslo, Norway · Division of Radiology and Nuclear Medicine, Oslo University Hospital, Oslo, Norway +5

Medical Image Segmentation U-Net Diffusion MRI

Ask PDF

Abstract

Medical Image Segmentation

Explore similar work

Expert-level detection of low-voltage intervals in EEG as a robust biomarker of neonatal encephalopathy

HRVConformer: Neonatal Hypoxic-Ischemic Encephalopathy Classification from the Heart Rate signals

Deep Learning Segmentation of Diffusion-Weighted MRI Acute Ischaemic Stroke: A Pragmatic Evaluation Across Three Datasets