Abstract—Glioblastoma is the most aggressive primary brain tumor and demands reliable image-based analysis for accurate interpretation. This study analyzes how different magnetic resonance image preprocessing strategies affect visual representation learning and deep feature discrimination in deep learning-based glioblastoma detection. An EfficientNetB0 convolutional neural network with transfer learning was employed to learn hierarchical image features from brain magnetic resonance images. The dataset consisted of 4465 images spanning 15 tumor classes, including 204 glioblastoma cases. Five preprocessing scenarios were evaluated: baseline (no preprocessing), contrast enhancement (Contrast Limited Adaptive Histogram Equalization (CLAHE)), image denoising, skull stripping, and intensity normalization. The results indicate that intensity normalization yields the most stable and discriminative feature representations, achieving the highest accuracy of 89.70% with balanced glioblastoma sensitivity and precision. Conversely, aggressive preprocessing methods distort image intensity distributions and spatial cues, leading to degraded feature learning and reduced performance. Visual interpretability analysis using Gradient-weighted Class Activation Mapping (Grad-CAM) further demonstrates that intensity normalization enables consistent attention to tumor-relevant regions while preserving critical morphological structures. These findings highlight that simple intensity standardization is more effective than complex enhancement techniques for robust and visually meaningful glioblastoma image analysis.

 

Keywords—medical image representation, feature learning, image preprocessing, magnetic resonance imaging, glioblastoma detection

Cite: Nada Firdaus, Lailil Muflikhah, Rizal Setya Perdana, and Nashi Widodo, "Analyzing the Impact of Preprocessing on Deep Learning-based Brain Tumor Classification Using Glioblastoma Magnetic Resonance Image," Journal of Image and Graphics, Vol. 14, No. 4, pp. 601-613, 2026.

Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).