A noninvasive, image-based smartphone app for diagnosing anemia

Abstract

Smartphone-based telehealth is steadily transforming the delivery of medical care worldwide, moving diagnosis of disease from the clinic to the home to potentially anywhere in the globe. Smartphone images alone have recently been used by physicians to remotely diagnose a myriad of diseases. However, smartphone telehealth approaches have yet to non-invasively replace blood-based testing, which remains a major cornerstone of disease diagnosis in modern medicine. While the addition of specialized smartphone attachments and supplemental calibration tools may enable point-of-care diagnosis and analysis of tissue and bodily fluid samples, the additional burden of blood and/or tissue sample collections combined with the additional cost and inconvenience associated with this equipment, prevents worldwide use of these potentially disruptive approaches. Therefore, a smartphone-based system, requiring nothing other than the smartphones native technology and capable of non-invasively replacing blood-based diagnostics, would transform the very nature of telehealth and the delivery of healthcare worldwide. Towards that end, I specifically focused on anemia, a potentially life-threatening disorder characterized by low blood hemoglobin (Hgb) levels that affects approximately 2 billion people worldwide. Despite the high prevalence of anemia, all existing diagnostic approaches to measure Hgb require specialized equipment and represent tradeoffs between invasiveness, accuracy, infrastructure needs, and expense. Aside from being cost-prohibitive, the necessary invasive blood sampling to measure Hgb levels causes discomfort and trauma in younger pediatric patients. By examining clinical pallor, a common symptom of anemia, I developed a methodology that quantitatively analyzes patient-sourced photos using smartphone-based algorithms to enable a noninvasive, accurate, and accessible anemia diagnostic. Here, a patient simply takes a picture of their fingernail beds using their smartphone, and the image analysis algorithm analyzes color data and image metadata to measure the corresponding Hgb level. By quantifying clinical pallor, this system non-invasively measures Hgb levels to within a clinically significant and well accepted margin of error (±2.6 g/dL) of the gold standard Hgb measurement tool with a sensitivity and specificity of 0.90 and 0.82, respectively, of predicting anemia (defined as Hgb < 11.0g/dL) in 100 pediatric patients at Children’s Healthcare of Atlanta with anemia of any etiology mixed with healthy subjects. This algorithm has been implemented into a smartphone app that is capable of outperforming trained hematologists in physical examination-based Hgb measurement. Overall, this technology has the capability to change the treatment paradigm for anemia as patients no longer need to visit a clinic to monitor their hemoglobin. In this thesis, I discuss the development of this image analysis algorithm and the implementation of the algorithm into a smartphone app.