Write a PREreview

Lung disease is a major global health challenge causing millions of deaths annually. Early diagnosis and treatment of lung disease is crucial for effective treatment, preventing mortality and reducing long-term morbidity. While most existing diagnostic research primarily utilizes unimodal medical image data, this approach often provides limited information. To incorporate additional clinical information in the diagnosis, multimodal strategies are increasingly being explored. Medical image and clinical data are the key medical information utilized by physicians to diagnose lung disease in addition to physical examination. In this work, we propose a comprehensive multimodal machine learning framework for lung disease detection that integrates structured clinical data with medical imaging modalities specifically, chest X-rays and computed tomography scans. The methodology includes robust data preprocessing, feature extraction using VGG16 for images and multiple techniques (mutual information, principal component analysis, and random forest) for clinical data, followed by fusion and classification using both classical machine learning and deep learning models. We introduce and evaluate a newly collected lung disease dataset comprising over 27,635 records combining imaging and clinical data from Ethiopian hospitals. Experiments conducted show that unimodal chest X-Ray image based detection achieves 95.28% accuracy while multimodal chest X-ray and clinical data based detection achieves accuracy 98.88%. Similar results are obtained for computed tomography scan based experiments with 97.62% for unimodal and 98.91% for multimodal detection. This study demonstrated the critical importance of multimodal data fusion in developing more accurate and clinically viable diagnostic system for lung diseases.