A neutrosophic set-based deep transfer learning framework for brain tumor classification

Abstract Brain tumors remain a major medical challenge and accurate classification of MRI scans is essential for early diagnosis. Conventional deep learning methods often struggle with noise, low contrast and uncertainty in medical images. This study presents a framework that combines Neutrosophic Set (NS) theory with Deep Transfer Learning (DTL) to improve the reliability of brain tumor classification. MRI images are transformed into the neutrosophic domain to generate Truth (T), Indeterminacy (I) and Falsity (F) components, which are used independently to train DenseNet-121, VGG-19 and ResNet-50 models. The neutrosophic transformation enhances structural clarity and uncertainty modeling, allowing the network to learn more discriminative features. Comparative experiments show that the neutrosophic components provide more stable and informative representations than standard preprocessing approaches and each component contributes uniquely to the classification process. This framework offers improved robustness, interpretability and clinical relevance for automated brain tumor diagnosis.