Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PSO-UNet: Particle Swarm-Optimized U-Net Framework for Precise Multimodal Brain Tumor Segmentation

Shoffan Saifullah, Rafał Dreżewski

TL;DR

This work tackles accurate brain tumor segmentation in multimodal MRI by addressing the heavy reliance on manual hyperparameter tuning in U‑Net models. It proposes PSO‑UNet, which embeds Particle Swarm Optimization to dynamically tune critical hyperparameters (filters, kernel size, learning rate) during training, using segmentation performance to guide the search. The approach delivers state‑of‑the‑art results on BraTS 2021 and Figshare with DSC around 0.958 and 0.952 and IoU around 0.919 and 0.909, while using a lightweight architecture with about 7.8 million parameters and a run time near 906 seconds. The findings demonstrate robust generalization across MRI modalities and tumor classes, with significant efficiency gains over traditional hyperparameter tuning and several PSO‑tuned baselines, suggesting strong clinical potential and avenues for future enhancements such as attention mechanisms and hybrid optimization.

Abstract

Medical image segmentation, particularly for brain tumor analysis, demands precise and computationally efficient models due to the complexity of multimodal MRI datasets and diverse tumor morphologies. This study introduces PSO-UNet, which integrates Particle Swarm Optimization (PSO) with the U-Net architecture for dynamic hyperparameter optimization. Unlike traditional manual tuning or alternative optimization approaches, PSO effectively navigates complex hyperparameter search spaces, explicitly optimizing the number of filters, kernel size, and learning rate. PSO-UNet substantially enhances segmentation performance, achieving Dice Similarity Coefficients (DSC) of 0.9578 and 0.9523 and Intersection over Union (IoU) scores of 0.9194 and 0.9097 on the BraTS 2021 and Figshare datasets, respectively. Moreover, the method reduces computational complexity significantly, utilizing only 7.8 million parameters and executing in approximately 906 seconds, markedly faster than comparable U-Net-based frameworks. These outcomes underscore PSO-UNet's robust generalization capabilities across diverse MRI modalities and tumor classifications, emphasizing its clinical potential and clear advantages over conventional hyperparameter tuning methods. Future research will explore hybrid optimization strategies and validate the framework against other bio-inspired algorithms to enhance its robustness and scalability.

PSO-UNet: Particle Swarm-Optimized U-Net Framework for Precise Multimodal Brain Tumor Segmentation

TL;DR

This work tackles accurate brain tumor segmentation in multimodal MRI by addressing the heavy reliance on manual hyperparameter tuning in U‑Net models. It proposes PSO‑UNet, which embeds Particle Swarm Optimization to dynamically tune critical hyperparameters (filters, kernel size, learning rate) during training, using segmentation performance to guide the search. The approach delivers state‑of‑the‑art results on BraTS 2021 and Figshare with DSC around 0.958 and 0.952 and IoU around 0.919 and 0.909, while using a lightweight architecture with about 7.8 million parameters and a run time near 906 seconds. The findings demonstrate robust generalization across MRI modalities and tumor classes, with significant efficiency gains over traditional hyperparameter tuning and several PSO‑tuned baselines, suggesting strong clinical potential and avenues for future enhancements such as attention mechanisms and hybrid optimization.

Abstract

Medical image segmentation, particularly for brain tumor analysis, demands precise and computationally efficient models due to the complexity of multimodal MRI datasets and diverse tumor morphologies. This study introduces PSO-UNet, which integrates Particle Swarm Optimization (PSO) with the U-Net architecture for dynamic hyperparameter optimization. Unlike traditional manual tuning or alternative optimization approaches, PSO effectively navigates complex hyperparameter search spaces, explicitly optimizing the number of filters, kernel size, and learning rate. PSO-UNet substantially enhances segmentation performance, achieving Dice Similarity Coefficients (DSC) of 0.9578 and 0.9523 and Intersection over Union (IoU) scores of 0.9194 and 0.9097 on the BraTS 2021 and Figshare datasets, respectively. Moreover, the method reduces computational complexity significantly, utilizing only 7.8 million parameters and executing in approximately 906 seconds, markedly faster than comparable U-Net-based frameworks. These outcomes underscore PSO-UNet's robust generalization capabilities across diverse MRI modalities and tumor classifications, emphasizing its clinical potential and clear advantages over conventional hyperparameter tuning methods. Future research will explore hybrid optimization strategies and validate the framework against other bio-inspired algorithms to enhance its robustness and scalability.

Paper Structure

This paper contains 13 sections, 14 equations, 6 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Method
  4. Overview of the Proposed Framework (PSO-UNet)
  5. Datasets and Preprocessing
  6. Implementation and Evaluation Metrics
  7. Results and Discussion
  8. PSO: Critical Analysis of UNet Architecture Parameters
  9. Evaluation of the Proposed PSO-UNet Framework Across Modalities and Tumor Classes
  10. Segmentation Results of Predicted Models
  11. Evaluation of the Proposed Method Compared with State-of-the-Art Techniques
  12. Computational Efficiency and Evaluation of the Proposed Method
  13. Conclusion

Figures (6)

  • Figure 1: U-Net framework optimized by PSO with an encoder-decoder structure and skip connections to enhance spatial and semantic feature retention.
  • Figure 2: MRI images from (a) BraTS 2021 and (b) Figshare with their masks.
  • Figure 3: (a) PSO-UNet convergence: DSC (train in blue, validation in red); (b) Evolution of architecture parameters (filters, kernel size, learning rate) across 10 particles and generations; (c) Correlation heatmap of evaluation metrics.
  • Figure 4: Evaluation metrics (Accuracy, Loss, DSC, and IoU) for the PSO-UNet framework across BraTS 2019 modalities (T1, T2, T1Gd, FLAIR) and Figshare tumor classes (Meningioma, Glioma, Pituitary).
  • Figure 5: Performance Metrics for Predicted vs. Mask Overlaps
  • ...and 1 more figures