LungLiteUS: AI-Powered Portable Ultrasound for Lung Diagnostics in Resource-Limited Settings

Access to advanced medical diagnostics is severely limited in many rural and underserved regions of Africa, Asia, and North America. This paper presents LungLiteUS, a lightweight, AI-powered system for analyzing lung ultrasound images to detect normal conditions, COVID-19, and pneumonia. The proposed system is designed to be integrated with wireless ultrasound devices, making advanced diagnostics accessible in areas where specialized medical professionals are scarce. We developed a custom convolutional neural network optimized for deployment on resource-constrained devices. Our model achieves 97.92% accuracy in distinguishing between normal lung patterns, COVID-19-specific patterns, and pneumonia manifestations in our test dataset. The system is deployed using TensorFlow Serving and FastAPI, with a responsive web interface designed for healthcare workers. Experimental results show that our lightweight model (0.7 MB) achieves comparable performance to more complex models while requiring significantly fewer computational resources. LungLiteUS represents a technical implementation ready for clinical evaluation that could potentially extend specialized medical diagnostics to underserved populations.