International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 12 | Dec 2025
p-ISSN: 2395-0072
www.irjet.net
Comparative Analysis of ResNet50 and ResNet152 Models for MRIBased Brain Tumor Classification Samanvaya K J, Arpitha C N, Dr Pushpa Ravikumar 1, PG Scholar, Dept. of Computer Science and Engineering, Adichunchanagiri Institute of Technology, Karnataka,
India.
2, Assistant Professor, Dept. of Computer Science and Engineering, Adichunchanagiri Institute of Technology,
Karnataka, India.
3, Professor & Head, Dept. of Computer Science and Engineering, Adichunchanagiri Institute of Technology,
Karnataka, India. -----------------------------------------------------------------------***--------------------------------------------------------------------------
Abstract - Brain tumor classification using Magnetic
Res Nets overcome the vanishing gradient limitation of traditional deep CNNs through skip-connections, enabling extremely deep architectures to extract rich hierarchical features from MRI scans [3]. Prior studies have demonstrated that residual models significantly outperform conventional CNNs in capturing tumor boundaries, intensity variations, and heterogeneous morphological patterns associated with glioma, meningioma, and pituitary tumors [4]. Mid-depth architectures such as ResNet50 have shown strong generalization capabilities and computational efficiency in medical image classification tasks, making them suitable for real-time systems [5].
Resonance Imaging (MRI) has significantly improved with the adoption of deep residual neural networks. In this study, a detailed comparative analysis of ResNet50 and ResNet152 models is performed to evaluate their effectiveness in multiclass brain tumor classification. The models were trained on a curated T1-weighted contrast-enhanced MRI dataset containing glioma, meningioma, pituitary tumor, and notumor images. Experimental results show that ResNet50 achieved a classification accuracy of 98.12%, with fast convergence and lower computational overhead, while ResNet152 attained a superior accuracy of 99.03%, demonstrating enhanced capability to extract deep hierarchical tumor features. Loss convergence, feature extraction differences, computational cost, and generalization performance were also analyzed. The findings indicate that although ResNet152 provides slightly better diagnostic performance, ResNet50 remains preferable for real-time clinical systems due to its lightweight architecture. This comparative study offers valuable insights into selecting appropriate deep residual models for medical imaging applications.
On the other hand, ultra-deep architectures such as ResNet152 have proven effective in capturing more complex tumor textures and subtle pathological features, providing improved accuracy in multi-class MRI brain tumor classification [6]. Several researchers have shown that deeper residual networks achieve superior performance for intricate medical imaging tasks due to their ability to learn multi-scale feature representations [7]. Moreover, transfer learning applied to deep residual networks has been widely adopted to boost diagnostic performance even with limited annotated medical datasets [8].
Key Words: Brain Tumour Classification, MRI, Deep Learning, ResNet50, ResNet152, Residual Networks, Medical Image Analysis, Transfer Learning, etc.
Recent advancements in hybrid residual attention models and deep feature pyramid architectures further highlight the importance of depth and residual learning mechanisms for robust tumor classification [9]. Comparative studies on residual architectures and transformer-based models also emphasize the continued relevance of Res Net variants in clinical diagnostic pipelines [10]. Despite extensive progress, a detailed comparison of moderately deep and ultra-deep residual models for brain tumor MRI classification remains essential for determining the optimal trade-off between accuracy, computational complexity, and deployment feasibility. In this work, we conduct a comprehensive comparative analysis of ResNet50 and ResNet152. Both models are trained on a curated MRI dataset and evaluated for accuracy, convergence behavior, feature extraction capability, and computational efficiency. The findings of this study aim to provide clear insights into selecting the
1. INTRODUCTION Brain tumors remain one of the most critical neurological diseases, requiring early and accurate diagnosis to reduce mortality and improve treatment outcomes. Magnetic Resonance Imaging (MRI) is the preferred diagnostic modality because of its superior soft-tissue contrast and ability to visualize structural abnormalities in multiple imaging sequences [1]. In recent years, deep learning techniques have transformed medical image analysis, particularly through Convolutional Neural Networks (CNNs) and advanced architectures such as Residual Networks (Res Nets) [2].
© 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 120