Explainable AI in MRI-Based Brain Tumor Segmentation: Advances, Challenges, Opportunities, and Future Directions
K. Arun Kumar, U. Veeresh, Yadala Sucharitha, S. Bhargavi Latha, Chatrapathy K., Shaik Asif, Iftikhar Aslam Tayubi, Pundru Chandra Shaker ReddyIntroduction:
Untreated brain tumors (BT) cause abnormal growth of brain tissue and pose serious health risks, including the possibility of death. Because of its high resolution and ability to differentiate between different types of tissues, magnetic-resonance-imaging (MRI) has replaced previous imaging modalities as the gold standard for diagnosing brain malignancies. In recent years, artificial intelligence and deep learning techniques have demonstrated significant potential in automating medical image analysis and improving diagnostic accuracy; however, the limited interpretability of conventional deep learning models has motivated the adoption of explainable artificial intelligence (XAI) for transparent and clinically interpretable brain tumor analysis.
Methods:
Various methods have been investigated for the MRI-based brain tumor segmentation and interpretable tumor characterization of brain tumors (BTs), including traditional machine-learning (ML) and deep learning (DL) strategies, and characteristics that have been hand-engineered. Recent approaches have improved segmentation accuracy, computational efficiency, transparency, and clinical interpretability while addressing the shortcomings of earlier methods. This PRISMA-guided review included 67 eligible studies selected using predefined inclusion and exclusion criteria from major scientific databases,
Results and Discussion:
Deep learning algorithms have demonstrated remarkable performance in MRI-based brain tumor segmentation across several application domains, including medical image analysis, computer vision, and clinical decision support. Despite substantial performance gains, achieving clinically meaningful interpretability remains a significant challenge. Clinicians have legitimate worries about the explainability, investigation, trust, and interpretability of DL, in addition to the complex models used for brain tumor segmentation (BTS). From traditional ML methods developed by hand to deep learning and explainable AI (XAI) algorithms, this review provides a comprehensive overview of traditional machine learning, deep learning, and explainable artificial intelligence techniques for MRI-based brain tumor segmentation. It also discusses the difficulties of DL algorithms and suggests neuro-symbolic learning (NSL) designs for BTS.
Conclusions:
This review provides a comprehensive and critical synthesis of explainable artificial intelligence techniques for MRI-based brain tumor segmentation. The study highlights recent advances in segmentation architectures, interpretable learning frameworks, and clinically relevant explainability methods, while identifying limitations in robustness, generalizability, and clinical adoption. Furthermore, future research opportunities involving neuro-symbolic learning, federated learning, and prototype-based explainability are discussed to support the development of trustworthy and clinically deployable segmentation systems.