ExplainabilityTransparencyTrustworthinesscs.CV
Demystifying Deep Learning-based Brain Tumor Segmentation with 3D UNets and Explainable AI (XAI): A Comparative Analysis
Ming Jie Ong, Sze Yinn Ung, Sim Kuan Goh, Jimmy Y. Zhong
October 9, 2025
arXiv.org
Key Takeaways
-
Explainable AI helps doctors understand AI decisions: The study demonstrates that visualization techniques like Grad-CAM can show physicians exactly which parts of brain MRI images the AI model is focusing on when making segmentation decisions, potentially increasing trust in AI-assisted diagnosis.
-
Different AI model architectures perform differently: Among the three models tested (UNet, ResUNet, and AttUNet), ResUNet achieved the best performance metrics for brain tumor segmentation, suggesting that model selection matters significantly for medical imaging applications.
-
Transparency in medical AI is achievable: By combining deep learning models with explainable AI techniques, researchers demonstrated that it's possible to make AI systems used in clinical settings more interpretable rather than treating them as "black boxes," which is important for responsible deployment in healthcare.
-
Visualization techniques reveal model reasoning: Grad-CAM and attention-based visualization provided insights into which tumor subregions different models focused on during analysis, offering a pathway for clinicians to validate whether the AI is making decisions based on medically relevant features.
-
Responsible AI requires multiple evaluation metrics: The study evaluated models across multiple dimensions—accuracy, speed, segmentation quality, and interpretability—rather than relying on a single metric, reflecting best practices for assessing AI systems intended to support clinical decision-making.
Abstract
The current study investigated the use of Explainable Artificial Intelligence (XAI) to improve the accuracy of brain tumor segmentation in MRI images, with the goal of assisting physicians in clinical decision-making. The study focused on applying UNet models for brain tumor segmentation and using the XAI techniques of Gradient-weighted Class Activation Mapping (Grad-CAM) and attention-based visualization to enhance the understanding of these models. Three deep learning models - UNet, Residual UNet (ResUNet), and Attention UNet (AttUNet) - were evaluated to identify the best-performing model. XAI was employed with the aims of clarifying model decisions and increasing physicians' trust in these models. We compared the performance of two UNet variants (ResUNet and AttUNet) with the conventional UNet in segmenting brain tumors from the BraTS2020 public dataset and analyzed model predictions with Grad-CAM and attention-based visualization. Using the latest computer hardware, we trained and validated each model using the Adam optimizer and assessed their performance with respect to: (i) training, validation, and inference times, (ii) segmentation similarity coefficients and loss functions, and (iii) classification performance. Notably, during the final testing phase, ResUNet outperformed the other models with respect to Dice and Jaccard similarity scores, as well as accuracy, recall, and F1 scores. Grad-CAM provided visuospatial insights into the tumor subregions each UNet model focused on while attention-based visualization provided valuable insights into the working mechanisms of AttUNet's attention modules. These results demonstrated ResUNet as the best-performing model and we conclude by recommending its use for automated brain tumor segmentation in future clinical assessments. Our source code and checkpoint are available at https://github.com/ethanong98/MultiModel-XAI-Brats2020
Paper Info
arXiv ID2510.07785
PublishedOctober 9, 2025
Citations
0
Relevance72%
VenuearXiv.org
TypeJournalArticle
Categorycs.CV
Responsible AI Topics
ExplainabilityTransparencyTrustworthiness
Links
Citation data from Semantic Scholar