التعلّم بالنقل المُحسَّن مع إعادة معايرة الانتباه لتصنيف أورام الدماغ باستخدام صور الرنين المغناطيسي (MRI)

الملخص

The accurate and timely identification of brain tumors in MRI scans remains a complex task due to tumor variability and limitations of manual analysis, which has led to the development of automated intelligent systems. This work suggests an attention-boosted deep learning model that combines transfer learning via the pretrained EfficientNetB0 and Squeeze-and-Excitation (SE) attention mechanism for enhanced feature representation. The training dataset comprises 3264 brain MRI images spanning four classes: glioma meningioma pituitary tumor, and absence of tumor. The image preprocessing steps include resizing, normalization, and label encoding. The proposed architecture fuses effective feature extraction, attention-driven recalibration, and a regularized classification head for better results. A systematic strategy for model development with early stopping and adaptive learning rate fine-tuning is implemented for reliable convergence. The results demonstrate that our proposed framework yields a validation accuracy of 97. 05% and a validation loss of 0. 0959 at the optimal epoch. In addition, the model surpasses traditional models like MobileNetV1 and DenseNet121 as well as recently developed hybrid models. Moreover, this study reveals that the use of attention mechanisms significantly enhances classification performance while keeping the computational costs low.