TRANS-EFFNET: A HYBRID MODEL FOR BRAIN TUMOR DETECTION USING EFFICIENTNET AND TRANSFORMER ENCODER

Abstract

ACCURATE CLASSIFICATION OF BRAIN TUMORS FROM MRI IS CRITICAL FOR EFFECTIVE DIAGNOSIS AND TREATMENT. IN THIS STUDY, WE INTRODUCE TRANS-EFFNET, A HYBRID MODEL COMBINING PRE-TRAINED EFFICIENTNET ARCHITECTURES WITH A TRANSFORMER ENCODER TO ENHANCE BRAIN TUMOR CLASSIFICATION ACCURACY. BY LEVERAGING EFFICIENTNET'S DEEP CNN CAPABILITIES FOR LOCALIZED FEATURE EXTRACTION AND THE TRANSFORMER ENCODER FOR CAPTURING GLOBAL CONTEXTUAL RELATIONSHIPS, OUR MODEL IMPROVES THE IDENTIFICATION OF INTRICATE TUMOR CHARACTERISTICS. FINE-TUNED WITH IMAGENET-DERIVED WEIGHTS AND UTILIZING EXTENSIVE DATA AUGMENTATION, TRANS-EFFNET WAS VALIDATED ON BOTH MULTI-CLASS AND BINARY DATASETS. TRANS-EFFNETB1 ACHIEVED 99.49 % ACCURACY ON THE MULTI-CLASS DATASET, WHILE TRANS-EFFNETB2 RECORDED 99.83 % ACCURACY ON THE BINARY DATASET, WITH PERFECT PRECISION, RECALL, AND F1-SCORE. THESE RESULTS UNDERSCORE TRANS-EFFNET'S ROBUSTNESS AND POTENTIAL AS A SIGNIFICANT ADVANCEMENT IN BRAIN TUMOR DETECTION AND CLASSIFICATION.