An Integrated AI and Sensor-Based System for Optimizing Gallbladder Ultrasound Image Acquisition

Cavero J.A.Z.; Perez M.A.F.; De Moura Mendoza J.M.; Alfaro F.A.N.; Gonzales A.A.G. · C3 2025 - IEEE Colombian Caribbean Conference (2025) · DOI: 10.1109/c366505.2025.11339995

Resumen

The limited availability of trained sonographers in low-resource and underserved regions hinders timely and accurate medical diagnoses, particularly in pediatric abdominal conditions. This paper presents an integrated AI-driven system combining a convolutional neural network (CNN), inertial sensors, and a graphical user interface (GUI) to assist non-specialist healthcare providers in acquiring high-quality gallbladder ultrasound images. The CNN, based on a modified AlexNet architecture, achieved 96.02% accuracy, 97.56% specificity, and 93.73% sensitivity in real-time gallbladder assessment. A custom-built inertial measurement unit (IMU) provided angular feedback with an average error below ± 4° compared to smartphone gyroscope standards. The GUI integrates sensor data and AI outputs to offer intuitive, real-time guidance during scanning. Designed using the Stanford Biodesign framework, this system aims to empower frontline healthcare workers, reduce misdiagnoses, and enhance ultrasound accessibility in resource-limited environments. Future work includes clinical validation and system expansion to support other organs and pathologies.

Instituciones: Pontificia Universidad Católica del Perú