Research Article
Attention-Based Deep Learning Model for Robust Pneumonia Classification and Categorization using Image Processing
@INPROCEEDINGS{10.4108/eai.28-4-2025.2357765, author={Deepika Lakshmi Kanumuri and Madhuri Kamma and Somitha Anna and Pavan Kumar Kolluru}, title={Attention-Based Deep Learning Model for Robust Pneumonia Classification and Categorization using Image Processing}, proceedings={Proceedings of the 4th International Conference on Information Technology, Civil Innovation, Science, and Management, ICITSM 2025, 28-29 April 2025, Tiruchengode, Tamil Nadu, India, Part I}, publisher={EAI}, proceedings_a={ICITSM PART I}, year={2025}, month={10}, keywords={transfer learning attention mechanism resnetv2152 cbam attention gaussian blur contour extraction lung opacity}, doi={10.4108/eai.28-4-2025.2357765} }- Deepika Lakshmi Kanumuri
Madhuri Kamma
Somitha Anna
Pavan Kumar Kolluru
Year: 2025
Attention-Based Deep Learning Model for Robust Pneumonia Classification and Categorization using Image Processing
ICITSM PART I
EAI
DOI: 10.4108/eai.28-4-2025.2357765
Abstract
A serious respiratory illness that causes inflammation in the lungs, pneumonia poses serious health concerns. Recent advancements in deep learning have revolutionized pneumonia detection by enabling automated, accurate, and efficient diagnosis. This project makes use of the ResNet152V2 pre-trained model with Convolutional Block Attention Module (CBAM) to further feature enhancement and classifies using neural network. The novelty of this paper highlights categorization of pneumonia into stages into Mild, Moderate and Severe pneumonia based of lung opacity values. The benchmark Kermany dataset from Kaggle is utilized and the model achieved 93.10% accuracy. To improve accessibility, a web application has developed that allows easy and efficient use of the model for real-time pneumonia detection. By integrating AI-driven diagnostics with a user-friendly interface, this project enhances early detection, reduces diagnostic errors, and contributes to improved health- care accessibility in resource-limited settings. This approach not only ensures high diagnostic precision but also facilitates timely medical intervention. The integration of CBAM with ResNet152V2 enhances feature discrimination, while stage-wise severity classification supports tailored treatment strategies. The user-centric web interface bridges technological advancements with clinical usability, promoting equitable healthcare delivery in underserved areas.
