T3SSLNet: Triple-Method Self-Supervised Learning for Enhanced Brain Tumor Classification in MRI

This study addresses the critical need for accurate brain tumor classification from MRI images for early diagnosis and effective treatment planning. Existing challenges include low image quality, sparsely labeled data, and tumor variability. The proposed T3SSLNet framework explores three self-supervised learning (SSL) approaches — SimCLR, MoCo, and BYOL — using ResNet-50 as the backbone, on a newly constructed dataset combining five public datasets. The framework comprises four key components: an Imaging Spectrum Enhancement Block for data augmentation, a Frozen Feature Extractor Block for hierarchical feature extraction, a Neural Representation Projection Learning Block for contrastive-positive pair learning, and an Unfrozen Classification Block for tumor classification.