Performance Analysis of a Hybrid Deep Learning Framework Integrating CNN, RNN, LSTM, and ResNet50 for Lung Disease Recurrence Prediction Using Chest X-Ray Images and Post-Recovery Clinical Data

Abstract

Post-recovery lung disease recurrence remains a significant clinical challenge, as it can increase morbidity and delay medical treatment. Early detection of recurrence requires an analytical approach capable of comprehensively integrating visual and clinical information. This study aimed to analyze the performance of a hybrid deep learning framework using Convolutional Neural Network (CNN), ResNet50, Recurrent Neural Network (RNN), and Long Short-Term Memory (LSTM) models to predict lung disease recurrence using post-recovery X-ray images and longitudinal clinical data. The dataset consists of post-recovery chest X-ray images of patients with pneumonia, pleural effusion, and tuberculosis, enriched through augmentation techniques, and longitudinal clinical data representing the dynamics of patient conditions over time. CNN and ResNet50 models are used to extract spatial features from medical images, while clinical-data RNN and LSTM models are utilized to model temporal patterns based on the patient's clinical history. The experimental results show that both CNN and ResNet50 achieved the highest accuracy of 97.78% with an F1-score of 0.978, demonstrating excellent performance in visual classification of lung images. On clinical data, RNN achieved 86% accuracy with an F1-score of 0.85, while LSTM demonstrated more stable performance with 90% accuracy and an F1-score of 0.89 in capturing longitudinal patterns of relapse. These findings confirm that the integration of spatial and temporal modeling within a hybrid deep learning framework can improve the effectiveness of early detection of lung disease relapse. The proposed approach has significant potential as a basis for developing medical decision support systems for more accurate and continuous post-recovery patient monitoring.

Access to Document

Other files and links

• Link to publication in Scopus
• Open Access Version Available

Fingerprint