From black box AI to XAI in neuro-oncology: a survey on MRI-based tumor detection

Brain tumor detection remains a critical focus in neuro-oncology, requiring precise and efficient diagnostic methods to support timely clinical decisions. Magnetic Resonance Imaging (MRI) is the modality of choice for this task, owing to its exceptional soft tissue contrast and spatial resolution. Recent advancements in deep transfer learning have driven transformative progress in automating tumor detection, achieving notable improvements in accuracy, scalability, and computational efficiency. This paper systematically reviews state-of-the-art transfer learning techniques, including pre-trained neural networks, domain adaptation approaches, fine-tuning methodologies, and ensemble learning frameworks. It also provides an in-depth analysis of widely used MRI datasets, highlighting persistent challenges such as data imbalance, overfitting, and variability across imaging domains. Emerging trends in model optimization, architectural innovation, and generalization strategies are discussed, offering a forward-looking perspective on the future of automated tumor detection. Through a detailed comparative evaluation, this review identifies the strengths and limitations of existing methods, while outlining potential directions for research aimed at developing robust, interpretable, and clinically deployable systems for brain tumor diagnosis.