Cloud-Driven Predictive Healthcare System using CNN, HierbaNetV1, and LSTM for Chest X-ray and Patient Data Analysis

Abstract

Early and accurate diagnostic procedures are, therefore, essential in modern medicine dealing with thoracic disorders, where late diagnosis could bear a fatal consequence. In this paper, a cloud-integrated predictive healthcare system based on deep-learning methods is proposed to classify chest X-rays with sequential patient metadata. The presented model integrates a Convolutional Neural Network (CNN) for extracting spatial features, HierbaNetV1 for learning deep representation, and Long Short-Term Memory (LSTM) to capture temporal patterns from patient health histories. Experiments on the NIH Chest X-ray 14 database containing a total of 112,120 images belonging to 14 classes of diseases were performed. The said system achieved an accuracy of 99.64%, a precision of 99.75%, a recall of 99.51%, and F1 score of 99.63%, thereby outclassing the traditional approaches. Also, an AUC-ROC rating of 0.9975 and Average Precision of 0.9978 further confirm the astounding discriminatory performance of the model. The integrated imaging and temporal data residing on a cloud platform thus allows for a scalable real-time prediction and decision support for diseases, one of the suitable solutions for the intelligent healthcare environment.