Brain tumor segmentation using particle swarm optimized histogram equalization and a VGG19 based U-NetDownload PDF Download PDF ArticleOpen accessPublished: 07 October 2026Shoffan Saifullah ORCID: orcid.org/0000-0001-6799-38341,2 &Rafał Dreżewski1 Scientific Reports (2026) Cite this articleSave articleView saved research We’re sharing this article early to provide faster access to peer-reviewed, accepted research. It is citable and carries a permanent DOI. This version is subject to further edits and will be replaced automatically by the final Version of Record. All legal disclaimers apply.AbstractAccurate brain tumor segmentation in MRI scans is critical for diagnosis and treatment planning. Conventional methods are often time-consuming, error-prone, and subject to operator bias. Although recent deep-learning approaches show promise for automating medical image segmentation, challenges such as interpretability hinder their clinical adoption. To address these issues, we propose SPOHE-VGGNet, a novel segmentation model that integrates a modified U-Net architecture with VGG-19 and Particle Swarm Optimized Histogram Equalization (PSO-HE). This approach enhances MRI contrast and improves segmentation performance relative to the evaluated preprocessing baselines. PSO-HE outperforms traditional and PSO-based image enhancement techniques, delivering superior image quality and facilitating precise feature extraction by VGG-19. We evaluated SPOHE-VGGNet on the Figshare T1-CE MRI dataset (3,064 images encompassing meningioma, glioma, and pituitary tumor classes) and the BraTS 2021 dataset comprising 1,251 cases, with four MRI modalities processed independently for two-dimensional Whole Tumor segmentation. On the Figshare dataset, the model achieved Dice Similarity Coefficients (DSC) of 0.9522 for meningioma, 0.9428 for glioma, and 0.9518 for pituitary tumors. For whole tumor segmentation on the BraTS dataset, it attained a DSC of 0.9463. Performance metrics for Figshare included accuracy, precision, recall, and F1-score of 0.9989, 0.9680, 0.9647, and 0.9662, respectively. On the BraTS dataset, these metrics were 0.9977, 0.9544, 0.9738, and 0.9640, demonstrating its applicability to independently evaluated two-dimensional slices from different MRI modalities. By combining segmentation performance with interpretability analyses, SPOHE-VGGNet provides a research-oriented framework for two-dimensional brain tumor MRI segmentation. Further external, volumetric, and clinician-centered validation is required before its use in clinical decision-making can be considered.AcknowledgementsThis paper was partially supported by the Polish Ministry of Science and Higher Education funds assigned to AGH University of Krakow. We also gratefully acknowledge Polish high-performance computing infrastructure PLGrid (HPC Center: ACK Cyfronet AGH) for providing computer facilities and support within computational grant no. PLG/2023/016757, PLG/2024/017503, and PLG/2025/018784.FundingResearch funding was provided by AGH University of Krakow (Program “Excellence initiative – research university”), ACK Cyfronet AGH–PLGrid (Grant Nos. PLG/2023/016757, PLG/2024/017503, and PLG/2025/018784), and Polish Ministry of Science and Higher Education funds assigned to AGH University of Krakow.Author informationAuthors and AffiliationsFaculty of Computer Science, AGH University of Krakow, Krakow, 30-059, PolandShoffan Saifullah & Rafał DreżewskiDepartment of Informatics, Universitas Pembangunan Nasional Veteran Yogyakarta, Yogyakarta, 55281, IndonesiaShoffan SaifullahAuthorsShoffan SaifullahView author publicationsSearch author on:PubMed Google ScholarRafał DreżewskiView author publicationsSearch author on:PubMed Google ScholarCorresponding authorCorrespondence to Shoffan Saifullah.Ethics declarationsCompeting interestsThe authors declare no competing interests.Ethics approval and consent to participateNot applicable.Consent for publicationNot applicable.Additional informationPublisher's noteSpringer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.AppendicesAppendix A performance metrics of the SPOHE-VGGNet modelThis appendix presents a comprehensive visualization of the SPOHE-VGGNet model’s performance metrics, highlighting the impact of various preprocessing methods and the effectiveness of the optimal configuration across different datasets. Figure 13 displays the results for the Figshare dataset, while Fig. 14 illustrates the BRATS 2021 dataset performance using the best configuration from the Figshare experiments.Fig. 13Full size imagePerformance Heatmaps for Figshare Dataset showing (a) Dice Similarity Coefficient (DSC) and (b) Jaccard Index (JI) metrics across different preprocessing methods.From Fig. 13:1.The heatmaps display the Dice Similarity Coefficient (DSC) and Jaccard Index (JI) for the Figshare dataset, categorized by preprocessing methods (No Preprocessing, HE, CLAHE, CLAHE-HE, and PSO-HE) for the Training and Validation sets.2.Results indicate that PSO-HE consistently yields the highest DSC and JI values, particularly for the Meningioma and Pituitary classes. The glioma class exhibits greater variability, suggesting segmentation challenges.3.Figure A.13(a) presents the DSC heatmaps, while Figure A.13(b) shows the JI heatmaps, highlighting the influence of preprocessing on model performance.Fig. 14Full size imagePerformance Heatmaps for BRATS 2021 Dataset Using the Best Results of SPOHE-VGGNet.From Fig. 14:1.Figure 14 depicts the SPOHE-VGGNet model’s performance on the BRATS 2021 dataset, optimized with the best preprocessing method (PSO-HE) from the Figshare experiments.2.The heatmaps show high DSC and JI values across different MRI modalities (FLAIR, T1, T1-CE, and T2), for both Training and Validation sets.3.Despite some variability between Training and Validation results, the overall segmentation performance remains strong, indicating the robustness of the optimized model.Appendix B detailed algorithms for PSO-based image enhancement techniquesAlgorithm 1: PSO-based image enhancement (PSO_IE)The Algorithm 1 details the PSO-based image enhancement method, which utilizes Particle Swarm Optimization (PSO) to optimize the gamma value for contrast enhancement. This process includes Power-Law Transformation and evaluates the fitness based on image standard deviation.Algorithm 1Full size imagePSO-Based Image Enhancement (PSO_IE).Algorithm 2: PSO-based histogram equalization (PSO_HE)The proposed Algorithm 2 optimizes the gamma value to improve image contrast, followed by thresholding and histogram equalization. The fitness function in this case is based on entropy, aiming to enhance image details.Algorithm 2Full size imagePSO-Based Histogram Equalization (PSO_HE).Rights and permissionsOpen Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.Reprints and permissionsAbout this article