Abstract :

Accurate And Timely Identification Of Brain Tumours Is A Decisive Factor In Effective Clinical Treatment And Improved Patient Survival. Magnetic Resonance Imaging (MRI) Serves As A Crucial Non-invasive Modality For Visualising Tumour Morphology; However, Manual Interpretation Of MRI Scans Is Both Time-consuming And Subject To Inter-observer Variability. To Address These Limitations, This Study Investigates The Effectiveness Of Deep Learning-based Automated Classification Using Three Distinct Variants Of The DenseNet201 Architecture For Multi-class Brain Tumour Recognition. The Models Are Designed To Categorise MRI Images Into Four Classes: Glioma, Meningioma, Pituitary Tumour, And Normal (non-tumour). The First Model, Referred To As B-DenseNet201-CA, Is Trained Entirely From Random Initialisation To Establish A Baseline Performance. The Second Model, PT-DenseNet201-FB, Adopts A Transfer Learning Strategy By Utilising A DenseNet201 Network Pretrained On ImageNet, Where The Convolutional Backbone Is Frozen And Only The Classifier Layers Are Trained. The Third And Most Advanced Approach, FT-DenseNet201-TL, Performs Selective Finetuning By Retraining The Final 50 Layers Of The Pretrained Network To Better Adapt The Learned Representations To The Medical Imaging Domain. All Models Are Trained And Evaluated On A Publicly Available Brain MRI Dataset Using Standard Performance Metrics, Including Accuracy, Precision, Recall, F1-score, And Confusion Matrices. Experimental Results Reveal That The Baseline Model Achieves An Accuracy Of 74.11%, Highlighting The Difficulty Of Training Deep Networks From Scratch On Limited Medical Datasets. The Frozen Transfer Learning Model Significantly Improves Classification Performance, Achieving An Accuracy Of 90.81%, Thereby Demonstrating The Effectiveness Of Pretrained Feature Representations. The Fine-tuned Transfer Learning Model Delivers The Highest Performance, Attaining An Accuracy Of 94.18% And Exhibiting Improved Class-level Discrimination, Particularly For Visually Similar Tumour Categories. The Findings Confirm That Partial Fine-tuning Of A Pretrained DenseNet201 Architecture Provides A Computationally Efficient And Highly Accurate Solution For Brain Tumour Classification. This Work Underscores The Critical Role Of Transfer Learning In Medical Image Analysis And Suggests Strong Potential For Integration Into Clinical Decision-support Systems To Enhance Diagnostic Reliability And Consistency. Keywords: Brain Tumour Classification; Magnetic Resonance Imaging (MRI); DenseNet201; Transfer Learning; Deep Fine-Tuning; Convolutional Neural Networks; Medical Image Analysis; Multi-Class Classification; Radiomics; Tumour Subtype Identification

Published:

06-11-2023

Issue:

Vol. 23 No. 11 (2023)

Page Nos:

200-216

Section:

Articles

License:

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

How to Cite