PGR-Net: Prior-Guided ROI Reasoning Network for Brain Tumor MRI Segmentation

Jiacheng Lu; Hui Ding; Shiyu Zhang; Guoping Huo

cs.CVMar 23, 2026

PGR-Net: Prior-Guided ROI Reasoning Network for Brain Tumor MRI Segmentation

Authors: Jiacheng Lu, Hui Ding, Shiyu Zhang, Guoping Huo

Organizations: College of Information Engineering, Capital Normal University, Beijing, China · School of Artificial Intelligence, China University of Mining and Technology-Beijing, Beijing, China

Abstract

Brain tumor MRI segmentation is essential for clinical diagnosis and treatment planning, enabling accurate lesion detection and radiotherapy target delineation. However, tumor lesions occupy only a small fraction of the volumetric space, resulting in severe spatial sparsity, while existing segmentation networks often overlook clinically observed spatial priors of tumor occurrence, leading to redundant feature computation over extensive background regions. To address this issue, we propose PGR-Net (Prior-Guided ROI Reasoning Network) - an explicit ROI-aware framework that incorporates a data-driven spatial prior set to capture the distribution and scale characteristics of tumor lesions, providing global guidance for more stable segmentation. Leveraging these priors, PGR-Net introduces a hierarchical Top-K ROI decision mechanism that progressively selects the most confident lesion candidate regions across encoder layers to improve localization precision. We further develop the WinGS-ROI (Windowed Gaussian-Spatial Decay ROI) module, which uses multi-window Gaussian templates with a spatial decay function to produce center-enhanced guidance maps, thus directing feature learning throughout the network. With these ROI features, a windowed RetNet backbone is adopted to enhance localization reliability. Experiments on BraTS-2019/2023 and MSD Task01 show that PGR-Net consistently outperforms existing approaches while using only 8.64M Params, achieving Dice scores of 89.02%, 91.82%, and 89.67% on the Whole Tumor region. Code is available at https://github.com/CNU-MedAI-Lab/PGR-Net.

Figures & tables

Figure 1: Model complexity vs. BraTS 2023 performance (average Dice over the three tumor regions).

Figure 2: An illustrative example of lesion distribution in the BraTS 2023 dataset. From left to right: a sample MRI image, lesion center distribution map across the entire dataset, and lesion ROI distribution heatmap across the entire dataset.

Figure 3: Example templates from BraTS 2023 training set: sample FLAIR image, extracted ROI, and ROI overlaid on the image.

Figure 4: Lesion size distribution of the BraTS 2023 training set, with lesion size on the horizontal axis and frequency on the vertical axis. The lower plot shows a magnified view of the 10–90 range with annotated 10 peaks.

Figure 5: Overall architecture of PGR-Net (Prior-Guided ROI Reasoning Network), which consists of (a) Hierarchical Top- $K$ ROI Decision, (b) ROI Win-RetNet Backbone, (c) ROI-Only Up-sampling, and (d) ROI-Aware Skip-connections. The WinGS-ROI module is shown in Fig. 6

Figure 6: Illustration of the computation process of the WinGS-ROI mechanism. For each ROI region, a Gaussian template is applied within the ROI and a spatial decay is applied along the outer boundary, and the fused guidance map is then applied to the input image.

Figure 7: Example results after WinGS-ROI guidance. From left to right are the input image example, the WinGS-ROI guidance map of a single ROI, the WinGS-ROI guided image of a single ROI, the WinGS-ROI guidance map of dual ROIs, and the WinGS-ROI guided image of dual ROIs.

Dataset	Slices	Modalities
BraTS-2019	51925	T1/T1ce/T2/FLAIR
BraTS-2023	193905	T1/T1ce/T2/FLAIR
MSD-Task01	75020	T1/T1ce/T2/FLAIR

Table 1: Modalities and the number of 2D slices for each dataset.

Model	Backbone		WinGS-ROI			Dice_score (%)			Hausdorff95
	RWR	HTK	WR	SC	UP	WT↑	TC↑	ET↑	WT↓	TC↓	ET↓
A (Baseline)	-	-	-	-	-	87.82 / 91.06	88.91 / 92.97	91.05 / 93.13	1.3264 / 1.1868	0.8409 / 0.7085	0.6645 / 0.6622
B	+	-	-	-	-	87.85 / 91.10	88.89 / 93.02	91.15 / 93.08	1.3205 / 1.1840	0.8351 / 0.7055	0.6650 / 0.6605
C	+	+	-	-	-	88.55 / 91.66	89.64 / 93.42	91.99 / 93.35	1.2911 / 1.1551	0.8275 / 0.6803	0.6439 / 0.6419
D	+	+	+	-	-	88.63 / 91.76	90.33 / 93.75	92.72 / 93.57	1.2887 / 1.1457	0.8125 / 0.6754	0.6406 / 0.6327
E	+	+	+	+	-	88.85 / 91.80	90.32 / 93.79	92.88 / 93.74	1.2692 / 1.1380	0.8129 / 0.6703	0.6400 / 0.6153

Table 2: Ablation study results of PGR-Net on the BraTS-2019/2023 dataset. RWR: ROI Win-RetNet; HTK: Hierarchical Top-

K

ROI Decision; WinGS-ROI includes WR: WinGS-ROI in Win-RetNet; SC: WinGS-ROI in Skip-connection; UP: WinGS-ROI in Up-sampling. Dice (%)

\uparrow

and Hausdorff95 (HD95)

\downarrow

are reported for WT, TC, and ET regions.

Model	Year	Dice_score (%)			Hausdorff95
Model	Year	WT↑	TC↑	ET↑	WT↓	TC↓	ET↓
UNet	2015	87.36 / 90.71 / 88.15	88.59 / 93.05 / 88.56	90.69 / 93.36 / 90.33	1.3582 / 1.1863 / 1.3800	0.9076 / 0.7329 / 0.9356	0.6897 / 0.6730 / 0.7861
Cascaded UNet	2019	87.81 / 90.32 / 89.06	89.40 / 92.85 / 88.29	90.92 / 92.47 / 90.24	1.3349 / 1.2091 / 1.3510	0.9002 / 0.7488 / 0.9268	0.6719 / 0.7591 / 0.7648
TransUNet	2021	84.50 / 90.71 / 87.26	86.72 / 92.52 / 88.58	88.39 / 92.92 / 89.52	1.3911 / 1.1810 / 1.3936	0.9300 / 0.7276 / 0.9349	0.7396 / 0.6869 / 0.8073
nnUNet	2021	87.81 / 90.34 / 89.33	90.23 / 92.74 / 88.79	90.96 / 92.37 / 90.41	1.2970 / 1.2100 / 1.3420	0.8311 / 0.7358 / 0.9250	0.6628 / 0.6722 / 0.7849
UNETR	2022	85.29 / 88.35 / 85.92	87.16 / 89.16 / 85.82	89.54 / 91.43 / 88.52	1.3831 / 1.2427 / 1.4480	0.9504 / 0.8926 / 1.0447	0.7042 / 0.7211 / 0.8310

Table 3: Comparison with The SOTA methods on BraTS-2019 / BraTS-2023 / MSD Task01 Datasets.

Figure 8: Examples of segmentation results from multiple methods. From left to right: Flair modality input image, Ground Truth (GT), and segmentation results from various comparison algorithms. Green boxes highlight regions with obvious segmentation errors.

Model	Year	Params(M)	FLOPs(G)	Inf Time(min)
UNet	2015	39.40	321.19	12:32
Cascaded UNet	2019	85.59	568.25	25:24
TransUNet	2021	105.21	237.83	11:02
nnUNet	2021	–	–	86:52
UNETR	2022	87.67	150.71	18:31
Swin UNETR	2022	25.11	106.80	21:33

Table 4: Comparison of model parameters and computational complexity on the BraTS 2023 test set, computed using a batch size of 16.

Explore similar work

Sep 24, 2026

Lightweight Vision Transformer-Based U-Net for Brain Tumor Segmentation from MRI

Accurate brain tumor segmentation from Magnetic Resonance Imaging is essential for diagnosis, treatment planning, and surgical guidance. Although Convolutional Neural Networks, particularly UNet, have achieved significant success in medical image segmentation, they often struggle to capture the long-range spatial dependencies required to model tumors with irregular shapes and complex boundaries. This paper proposes a lightweight Vision Transformer UNet that combines the hierarchical feature extraction capability of UNet with the global context modeling of Vision Transformers. The proposed architecture incorporates a compact ViT bottleneck within a U-Net encoder-decoder framework, enabling effective learning of both local and global features while maintaining computational efficiency with only 2.6 million trainable parameters. The model was evaluated on the TCGA LGG MRI Segmentation dataset, achieving a mean Intersection over Union of 0.8100 and a Dice score of 0.8446, outperforming the baseline UNet by 3.75% and 3.15%, respectively. Extensive quantitative and qualitative analyses, including confusion matrix evaluation, precision recall curves, per-image performance distribution, and tumor size dependency analysis, demonstrate the effectiveness and robustness of the proposed method for brain tumor segmentation.

Sheekar Banerjee, Md. Srabon Chowdhury, Md. Mahbub Hasan Akash +1

Department of CSE, IUBAT, Dhaka, Bangladesh

Hybrid CNN-Transformer Architectures Vision Transformer U-Net Medical Image Segmentation Brain Tumor Segmentation

Ask PDF

May 21, 2026

SegGuidedNet: Sub-Region-Aware Attention Supervision for Interpretable Brain Tumor Segmentation

Accurate segmentation of brain tumour sub-regions from multi-parametric MRI is critical for treatment planning yet remains challenging due to morphological variability, class imbalance, and overlapping appearances of tumour regions across imaging sequences. We propose SegGuidedNet, a three-dimensional residual encoder--decoder network introducing a novel SegAttentionGate module that explicitly supervises the decoder to produce spatially discriminative attention maps for each tumour sub-region necrotic core, peritumoral oedema, and enhancing tumour via a lightweight auxiliary loss, adding less than 0.2% parameter overhead. This sub-region supervision maintains decoder discriminability between visually ambiguous classes while providing free-of-cost spatial interpretability at inference without any post-hoc explanation method. Evaluated independently on BraTS2021 and BraTS2023 GLI across 251 held-out subjects each, SegGuidedNet achieves mean Dice of 0.905 (ET= 0.873, TC=0.906, WT=0.935) and 0.897 (ET=0.859, TC=0.902, WT=0.931) respectively, surpassing ensemble-based nnU-Net and HNF-Netv2 as a single model and approaching Swin UNETR a 10-model ensemble within 2--4 Dice points at a fraction of the inference cost. The consistency of results across two benchmark editions further confirms the generalisability of the proposed approach, offering competitive accuracy with built-in interpretability in a lightweight, clinically practical framework.

Hasaan Maqsood, Saif Ur Rehman Khan, Sebastian Vollmer +2

National University of Sciences and Technology (NUST), Islamabad, Pakistan. · Department of Computer Science, Rhineland-Palatinate Technical University of Kaiserslautern-Landau, Kaiserslautern, 67663, Germany. · German Research Center for Artificial Intelligence, Kaiserslautern, 67663, Germany. +2

Image Segmentation BraTS Medical Image Segmentation 3D Medical Image Segmentation Explainable Medical Image Analysis Brain MRI+1

Ask PDF

May 28, 2026

A Novel Global Context-aware Deep Neural Network for Enhanced Brain Tumor Segmentation using Magnetic Resonance Images

Brain cancer's severity necessitates precise brain tumor segmentation, which is crucial for effective brain tumor diagnosis. Manual identification, burdened by high costs, labor, and error risks, highlights the need for automated methods. In this study, we introduce the Global Context-aware Squeeze and Excite Residual UNet (GCSER-UNet), which facilitates a fusion of spatial and channel-wise attention and thus enhances the model's capacity to capture intricate spatial dependencies and contextual information. GCSER-UNet efficiently extracts tumor segments from multimodal MRI slices, delivering exceptional performance. Evaluations on benchmark databases exhibit its superiority, achieving a notable 94 percent dice score on the TCGA LGG dataset, surpassing the state-of-the-art dice score of 91.8 percent. In the BraTS 2020 dataset, the proposed GCSER-UNet ensemble approach yielded dice scores of 95 percent, 92 percent, and 90 percent for the tumor regions - Whole Tumor (W), Tumor Core (T), and Enhancing Tumor (E), respectively. The current state-of-the-art dice scores were 94 percent, 93 percent, and 88 percent. These compelling outcomes highlight the efficacy of GCSER-UNet in precise brain tumor segmentation and thus can aid neurologists in effective brain cancer management and treatment planning.

Sourjya Mukherjee, Ananya Bhattacharjee, R. Murugan

Biomedical Imaging Laboratory, Department of Electronics and Communication Engineering, National Institute of Technology Silchar, Silchar, Assam, India, 788010

U-Net Medical Image Segmentation Attention Mechanisms Brain Tumor Segmentation

Ask PDF