Advanced Residual Network Architecture for Automated Brain Tumor Diagnosis: The MesResNet50 Approach

Abstract

Identifying tumors in the brain is very crucial for early diagnosis that supports in developing successful therapeutic treatments. Traditional methods were based on manual feature extraction, and classic machine learning techniques lack the ability to explain the intricate variations found in the morphology and positioning of brain tumors. In this paper, we propose MesResNet50, a modified residual network architecture for classifying different brain tumor types from magnetic resonance imaging (MRI) images. MesResNet50 utilizes transfer learning like partial layer freezing and possesses a strong classification head in an effort to improve generalization and performance. The dataset consists of 7023 MRI scans. These scans depicted four types of tumors: no tumor, glioma, meningioma, and pituitary tumor. MesResNet50 performed better than other models like CNN, AlexNet, VGG16, and ResNet50 on major evaluation metrics. With a test accuracy of 96.80%, an F1-score of 96.69%, and a ROC-AUC of 0.9949, MesResNet50 clearly demonstrates the capabilities to discriminate between classes and has a well-balanced performance on sensitivity and specificity metrics. Such effectiveness is due to residual architecture enabling greater depth and, hence, improved feature extraction and generalization. The improved residual architecture has much greater computational cost, which can be a burden in low-resource environments. The findings demonstrate the potential of utilizing MesResNet50 as a reliable and good approach for automated diagnosis of brain tumors for clinical decision support. Future work will focus on improving efficiency for rapid inference, evaluating simpler architectures to reduce resource usage, and testing the model on multicenter datasets to enhance clinical utility.



Received: 31 August 2025 | Revised: 4 February 2026 | Accepted: 3 April 2026



Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.



Data Availability Statement

The Brain Tumor MRI datasets that support the findings of this study are openly available in Kaggle at https://www.kaggle.com/datasets/masoudnickparvar/brain-tumor-mri-dataset. The data that support the findings of this study are openly available in GitHub at https://github.com/vasanthbhagawat-source/Ensembled.git.



Author Contribution Statement

Vasanth C. Bhagawat: Conceptualization, Methodology, Software, Formal analysis, Investigation, Data curation, Writing – original draft, Writing – review & editing, Visualization. Pravinth Raja Suthkar: Validation, Resources, Supervision, Project administration.