A Multi-Model Deep Learning Approach for Classification of Different Types of Brain Tumour

Abstract

This paper examines recent advancements in Deep learning techniques with Classification of brain tumors. Brain tumours pose significant diagnostic challenges, requiring precise and timely intervention for improved patient outcomes. Deep Learning(DL), particularly Convolutional Neural Networks (CNNs), has transformed medical imaging by enabling automated and accurate tumour classification. Key methods discussed include CNNs, transfer learning, & hybrid models, which have shown promising results in improving diagnostic efficiency. Additionally, widely used datasets like TCIA and When evaluating model performance, evaluation metrics like accuracy and AUC-ROC are essential.

DL for brain tumor classification still faces several obstacles despite tremendous advancements. Data scarcity, class imbalance, and model interpretability hinder widespread clinical adoption. Addressing these limitations requires advancements in explainable AI, self-supervised learning future, and multi-model approaches, which integrate diverse data sources for more comprehensive analysis. Research should focus on these areas to enhance reliability and clinical applicability of deep learning-based diagnostic systems. This paper proposes two distinct approaches were implemented and evaluated for brain tumour classification: a Convolutional Neural Network (CNN) designed from scratch, and a Transfer Learning model using EfficientNetB0 pre-trained on ImageNet and the Transfer Learning approach using EfficientNetB0 demonstrated higher accuracy, better generalization, along with instruction behaviour that is more consistent than that of the custom CNN.