In low and middle-income countries (LMICs) pneumonia is by far the leading cause of death among children < 5 years of age. Despite progress in reducing global pneumonia deaths, pneumonia still kills over 1 million children a year. A key factor is the challenge of pneumonia diagnosis. Chest X-Ray is the gold standard for pneumonia diagnoses but exposes children to ionizing radiation and is mainly restricted to hospital settings. Outside of hospitals, pneumonia is diagnosed based on clinical grounds using the WHO?s Integrated Management of Childhood Illness algorithms. However, that approach lacks specificity, resulting in significant overdiagnosis of pneumonia. That wastes precious resources, exposes children to antibiotics unnecessarily, and delays identification of the true cause of a child?s illness. Recent years have seen an explosion of interest in bedside ultrasound as a radiation-sparing alternative to X- Ray. The accuracy of bedside US for diagnosing pneumonia is comparable to X-Ray. However, traditional bedside US suffers the same limitations as X-Rays in terms of portability, and still requires interpretation of images by a trained radiologist. Recent innovations in ultrasound technology and artificial intelligence applications suggest a possible pathway forward. Bluetooth enabled US transponders that connect to a smart phone or tablet, effectively create a truly portable US suite that can fit into one?s pocket. Similarly, advances in artificial intelligence render possible the automated interpretation of mobile bedside US (mBSUS) images on a smart phone, obviating the need for a radiologist. The twin goals of this project are: 1) to compare the accuracy of the mobile, Bluetooth US system, linked to a cell phone, against the gold standard of Chest X-Ray for diagnosis of pneumonia among children aged 1-59 months; and 2) to apply machine learning algorithms to assist in the identification of accurate classification features that reliably identify lobar pneumonia. If successful, this would be a pivotal first step towards the goal of a truly portable yet still highly accurate approach to the diagnosis of pediatric pneumonia that could function autonomously without the need for external evaluation by a skilled radiologist. Such a tool would revolutionize pneumonia case management in resource limited parts of the world and could save the lives of many.
Pneumonia remains the leading cause of death for children <5 years around the world. A key barrier is the lack of affordable and portable imaging systems. In this project, we will 1) test the accuracy of a mobile ultrasound system that connects to a smart phone via blue tooth against the gold standard of chest X-Ray, and 2) use machine learning algorithms to automate the diagnosis of pneumonia, thereby obviating the need for external review by a radiologist.
