Lightweight CNN for accurate brain tumor detection from MRI with limited training data

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Early brain tumor detection is crucial for effective treatment.
• Limited data availability poses a significant challenge for developing accurate diagnostic models.
• Deep learning, particularly Convolutional Neural Networks (CNNs), shows promise in medical image analysis.

Purpose of the Study:

• To develop a robust and lightweight deep learning model for early brain tumor detection using MRI.
• To design a CNN-based diagnostic model capable of accurately classifying MRI scans as tumor-positive or tumor-negative.
• To address the challenge of limited data availability in brain tumor diagnostics.

Main Methods:

• A five-layer CNN architecture was implemented using TensorFlow and TFlearn.
• The model was trained on a dataset of 189 grayscale brain MRI images with balanced classes.
• Training utilized the Adam optimizer over 10 epochs and 202 iterations, with evaluation metrics including accuracy, precision, recall, F1 Score, and ROC AUC.

Main Results:

• The proposed CNN model achieved 99% accuracy in both training and validation.
• High performance was confirmed by precision (98.75%), recall (99.20%), F1-score (98.87%), and ROC-AUC (0.99).
• The model outperformed a baseline TensorFlow model trained on a larger dataset.

Conclusions:

• Accurate brain tumor detection is feasible with limited data through optimized CNNs.
• The developed model demonstrates high reliability and clinical relevance for early detection.
• Future research will focus on expanding datasets and integrating explainable AI for clinical applications.