Multimodal AI Framework for Pulmonary Risk Prediction via Imaging, Patient Metadata, and Acoustic Analysis

There aren't enough multimodal datasets, different ways to diagnose, or a balanced number of clinical case classifications to make it easy to find pulmonary diseases early on. This study talks about a tri-modal artificial intelligence system that uses chest X-ray images, patient metadata, and coughing audio signals to make pulmonary risk assessment more thorough and accurate. The system uses convolutional neural networks for imaging, multilayer perceptrons for clinical features, and one-dimensional CNN networks for audio signals. It has an adaptive multimodal fusion layer. Use advanced augmentation and synthetic oversampling methods to deal with a lack of data and uneven class distributions. This makes sure that all groups of patients are equally represented. This work improves the accuracy of predictions and makes them easier to understand by using explainable AI methods like Grad-CAM and SHAP.