International Journal of Computing and Digital Systems | 2026
Authors: Khan M.S.H.; Rashid M.R.A.; Hasan M.; Haque A.; Antu A.B.; Tanha A.T.; Rahman A.
DOI: 10.12785/ijcds/1571107283
Journal: International Journal of Computing and Digital Systems
Year: 2026
Publisher: University of Bahrain
Document Type: Article
Open Access: All Open Access; Gold Open Access
Cited by: 0
Advances in artificial intelligence (AI) have opened new possibilities for improving diagnostic accuracy and clinical decision support in healthcare. This study introduces an integrated framework that combines Convolutional Neural Networks (CNNs) with a Large Language Model (LLM) to enhance the detection and management of respiratory diseases, particularly pneumonia and COVID-19. The proposed system bridges the gap between image-based diagnosis and text-based clinical reasoning by linking visual feature extraction with evidence-driven recommendation generation. The CNN component, incorporating InceptionV3, MobileNetV2, and NASNet architectures, was trained to classify chest X-ray images with high precision, achieving accuracies of 92.85%, 91.88%, and 95.92%, respectively. Each model was fine-tuned using transfer learning and extensive data augmentation to improve generalization across varied imaging conditions. The LLM component, built upon the LLaMA2-7B-chat-GGML model, interprets the diagnostic outcomes from the CNNs and formulates therapeutic suggestions grounded in a curated medical knowledge base containing clinical literature, treatment protocols, and WHO guidelines. The integrated framework achieved a 33.1% reduction in inference time—from 165.6 to 111.9 seconds—while maintaining diagnostic reliability and contextual accuracy. Quantitative results demonstrate that coupling visual pattern recognition with natural language reasoning substantially enhances the interpretability, scalability, and responsiveness of automated clinical systems. Beyond improving classification accuracy, the proposed method also enables real-time generation of patient-specific insights that can support medical professionals in decision-making and follow-up planning. Overall, the findings highlight the potential of a unified CNN–LLM approach for developing transparent, adaptable, and domain-aware medical AI systems capable of assisting clinicians in delivering faster and more personalized respiratory disease management. © 2026, University of Bahrain. All rights reserved.
Bioinformatics; Convolutional neural network; Large language models; Medical diagnosis; Natural language processing; Patient healthcare; Respiratory diseases