DuTu Pulmo: Towards Intelligent Telehealth with YOLOv11-Driven Lung Disease Detection from Chest X-Rays

Tóm tắt

Lung cancer and chronic respiratory diseases remain leading causes of morbidity and mortality in Vietnam, with over 24,000 new lung cancer cases annually, approximately 75% of which are diagnosed at advanced stages. Diagnosis based on chest X-rays depends heavily on radiologists, who face increasing workload pressure alongside a low physician-to-population ratio, leading to delays and missed abnormalities. This paper presents DuTu Pulmo, an AI-assisted telehealth platform that integrates deep-learning-based chest X-ray analysis with a full clinical workflow system. The core component is a YOLOv11 object detection model fine-tuned on the VinDrCXR dataset of over 18,000 annotated images, enabling simultaneous detection and localization of 14 pulmonary abnormalities. Detected findings are categorized into four risk levels and visualized with color-coded bounding boxes to facilitate rapid clinical interpretation. The system adopts a microservice architecture, including physician and patient interfaces, a backend for electronic health records and real-time services, and a dedicated AI inference module. A retrieval-augmented generation (RAG) chatbot further supports patient guidance and symptom triage. Experimental results achieve 91.9% precision, 67.5% recall, and 82.1% mAP@50, with inference time under 5 seconds per image, demonstrating a scalable solution for AI-assisted respiratory care in resource-limited settings