Lung Image Classifier – COVID, Pneumonia, and Normal Detection
February 10, 2024
Overview
This project implements a lung image classification system using deep learning models to distinguish between three critical classes: COVID-19, Pneumonia, and Normal lung conditions. It leverages Convolutional Neural Networks (CNNs), including a custom model, DenseNet, and GoogLeNet, trained on grayscale X-ray images of shape (232, 232, 1).
Key Features
Multi-Model Evaluation: Trained and compared the performance of three models to identify the most effective architecture.
Live Prediction Interface: Designed a user-friendly interface where users can upload lung images and get instant diagnostic predictions.
Error Feedback Handling: Integrated error handling to improve user experience during training or inference failures.
Visual Demonstrations: Provided full GUI demo with live model prediction feedback.
Technologies Used
Python & Keras: For model architecture, training, and evaluation.
TensorFlow: As the backend deep learning framework.
OpenCV: For handling and preprocessing medical images.
Flask: To serve the classification model through a web interface.
Google Drive: Used to store and share pretrained models and datasets.
Matplotlib & Seaborn: For visualization of training metrics and results.
Dataset
The dataset consists of labeled grayscale X-ray lung images. Images were preprocessed and classified into three categories: COVID, Pneumonia, and Normal. You can access the full dataset and pretrained models via this Google Drive link.
Challenges and Learnings
Handling medical imaging data required careful preprocessing and validation to avoid misclassification. Model interpretability and balancing sensitivity between COVID and Pneumonia predictions were key focuses. Building a reliable web interface with Flask further enhanced my experience in AI-powered health applications.
Outcome
The application provides an accessible way for users to analyze lung X-ray images and receive predictions across three major diagnostic classes. The project is a step toward practical AI applications in healthcare and medical imaging.
