ATTENTION-ENHANCED CNN WITH GRAD-CAM FOR EXPLAINABLE BRAIN TUMOR CLASSIFICATION

Creative Commons License

Mustafa M. A., Erdem O. A., Söğüt E.

International Conference of Innovative Computer Engineering (ICE 2025), Ankara, Türkiye, 06 Kasım 2025, ss.7, (Özet Bildiri)

  • Yayın Türü: Bildiri / Özet Bildiri
  • Basıldığı Şehir: Ankara
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.7
  • Açık Arşiv Koleksiyonu: AVESİS Açık Erişim Koleksiyonu
  • Gazi Üniversitesi Adresli: Evet

Özet

Abstract

Aim: To analyze the clinical interpretability of deep learning models for classifying brain tumors on MRI.

Methods: We compared a baseline Convolutional Neural Network (CNN), an attention-enhanced CNN, and transfer-learning back-bones (MobileNetV2, InceptionV3, Xception) on a four-class MRI dataset (glioma, meningioma, pituitary, no tumor). Models were trained with Adam and evaluated using Grad-CAM visualizations, F1-score, accuracy, precision, recall, and confusion matrices.

Result: MobileNetV2 achieved the highest accuracy (95.7%), with closely aligned precision, recall, and F1-score. The attention-augmented CNN performed competitively, and Grad-CAM highlighted tumor-relevant regions, supporting model reliability. 

Conclusion: Transfer learning, especially MobileNetV2, offers strong performance for MRI tumor classification, while attention mechanisms and Grad-CAM improve focus and interpretability, facilitating clinical adoption.