A Hybrid Deep Learning and Machine Learning Model for Multi-Class Lung Disease Detection in Medical Imaging
Unknown authors
Abstract
Lung diseases encompass a broad range of conditions that affect the respiratory system, driven by factors such as genetic predisposition, environmental pollution, infections, and smoking.Early detection is essential for effective treatment and improved patient outcomes.This paper introduces an innovative framework that combines deep learning (DL) and machine learning (ML) models to autonomously detect and classify lung diseases using medical images from volumetric datasets.The proposed model aims to classify lung cancer into subtypes, including adenocarcinoma, squamous cell carcinoma, and large cell carcinoma, while also distinguishing between pneumonia and COVID-19.Convolutional Neural Networks (CNNs) are employed for feature extraction, demonstrating strong performance when applied to biomedical image datasets, such as the COVID-19 Radiography Database from Kaggle.A 2D CNN model is used in this study to efficiently extract features while reducing computational overhead.These CNN-extracted features are then classified using various machine learning algorithms, including Random Forest, Adaboost, and Voting classifiers.The hybrid model showed robust performance across diverse testing scenarios, highlighting its potential for practical applications and establishing it as a valuable tool for the automated detection and classification of lung diseases.The datasets employed in this study are sourced exclusively from open-access datasets available on the Kaggle website contains 32,975 images of CT-scan and CXR types with dimensions of 224 224.Notably, the hybrid CNN-ML classifier achieved high accuracy in identifying pneumonia, with a recall of 0.94, precision of 0.98, and F1-score of 0.96.For COVID-19, the CNN classifier yielded a recall, precision, and F1-score of 0.94 across all metrics.The hybrid approach can accurately detect and categorize lung diseases, particularly pneumonia, and suggest its potential for broader medical use.By integrating a Random Forest classifier with a Convolutional Neural Network, the proposed model demonstrated a significant improvement in accuracy, reaching 89% compared to the baseline CNN's 85.4%.