Deep Learning–Driven Automated Grading of Astrocytoma from Brain MRI Using an Enhanced DenseNet-169 Framework

Abstract

Background/Objectives: Astrocytoma represents one of the most widespread categories of primary brain tumors, with higher-grade variants linked to aggressive progression and poor survival outcomes. Conventional diagnostic techniques, including biopsy procedures, are invasive and may introduce clinical risks. This study aims to develop an artificial intelligence-based, non-invasive framework for automated grading of astrocytoma using brain magnetic resonance imaging (MRI). Methods: A dataset of brain MRI images compiled retrospectively was collected from the Radiology Department of Bahawal Victoria Hospital, Bahawalpur, Pakistan. Multiple pre-trained convolutional neural network architectures namely ResNet-152, VGG-19, and MobileNetV3 were implemented using transfer learning for com-parative evaluation. An optimized DenseNet-169 model was proposed, incorporating fine-tuning strategies, advanced regularization, and attention-inspired feature learning to improve classification performance. Standard preprocessing and data augmentation techniques were applied. The model's performance was evaluated by accuracy, precision, recall, F1-score, and AUC-ROC metrics under a consistent validation protocol. Results: The proposed DenseNet-169-based framework demonstrated superior classification capability, achieving an accuracy of 99.68%, and outperformed ResNet-152 (96.34%), VGG-19 (95.22%), and MobileNetV3 (97.42%). The model also exhibited strong generali-zation performance across evaluation metrics. Conclusions: The presented approach offers an ef-fective and non-invasive solution for astrocytoma grading using MRI data. Integrating deep learning techniques improves diagnosis accuracy and may aid clinical decision-making, especially in re-source-limited healthcare environments. The proposed framework emphasizes the potential of artificial intelligence to advance medical image analysis and improve patient outcomes.