CNN-LCA: A Lightweight Attention-Based Network with Confidence-Guided Grad-CAM for Explainable Brain Tumor Classification

Published in ICE-3210 Capstone Project, Pabna University of Science and Technology, 2026

Undergraduate Capstone Thesis / Research Project

This work represents my undergraduate capstone research project conducted at Pabna University of Science and Technology , under the Department of Information and Communication Engineering.

The project introduces CNN-LCA, a lightweight attention-based deep learning architecture designed for explainable brain tumor classification from Magnetic Resonance Imaging (MRI) scans.

The proposed framework integrates a Lite Convolutional Attention (LCA) mechanism to improve feature representation while maintaining a lightweight architecture suitable for resource-constrained environments.

For model interpretability, the framework incorporates confidence-guided Grad-CAM, enabling visualization of the image regions that contribute most strongly to the model's predictions.

Key Contributions:

  • Designed a lightweight CNN-based architecture with an integrated Lite Convolutional Attention mechanism.
  • Developed an explainability framework using confidence-guided Grad-CAM.
  • Conducted brain tumor classification across four categories: Glioma, Meningioma, Pituitary Tumor, and No Tumor.
  • Performed experimental evaluation using a dataset containing 7,200 brain MRI images.
  • Evaluated classification performance and model efficiency for practical deployment in resource-constrained clinical environments.
  • Conducted external validation to investigate the generalization capability of the proposed framework.

The experimental results demonstrated strong classification performance, with the proposed CNN-LCA framework achieving a reported accuracy of 98.40%.

The project also emphasizes explainable Artificial Intelligence (XAI), providing visual explanations of model predictions through Grad-CAM visualizations.


Project Details:
Course Code: ICE-3210
Project Type: Undergraduate Capstone Project
Research Area: Medical Artificial Intelligence / Deep Learning
Application: Brain Tumor Classification
Institution: Pabna University of Science and Technology
Department: Information and Communication Engineering
Supervisor: Prof. Dr. Md. Sarwar Hosain

Research Focus:
Lightweight Deep Learning · Medical Image Analysis · Brain Tumor Classification · Attention Mechanism · Explainable AI · Grad-CAM