Enhanced MRI-based brain tumor diagnosis using neuro-symbolic integration

Brain tumor diagnosis from MRI scans is an essential area of contemporary medical imaging. Although deep learning approaches like Convolutional Neural Networks (CNNs) have greatly improved tumor classification, their lack of explainability hinders clinical confidence. Neuro-Symbolic Learning (NSL) rectifies this by integrating perceptual learning from CNNs with symbolic reason, creating models that are interpretable and accurate. We introduce in this paper an NSL-based system for brain tumor detection, tested on the BRATS dataset. Experimental results confirm that the proposed system reaches accuracy levels of 92 96% and a Dice score of 0.90, comparable with state-of-the-art CNN models, but with much better interpretability (5/5) through symbolic rule-based reasoning. Additionally, NSL minimizes diagnostic variability through the inclusion of clinical domain rules, leading to more precise confidence intervals (±1.2%) and stable performance across tumor subtypes. These results identify that NSL not only preserves high predictive performance but also increases clinical transparency, closing the gap between machine learning accuracy and human-interpretable reasoning in healthcare.