Intelligent Heartsound Diagnostics on a Cellphone Using a Hands-Free Kit

In resource-constrained environments, supply chains for consumables, repairs and calibration of diagnostic equipment are generally poor. To obviate this issue, we propose the use of widely available hardware with a strong supply chain: a cellphone with a hands-free kit. In particular, we focus on the use of the audio channel to determine heart rate (HR) and heart rate variability (HRV) in order to provide a first level screening system for infection. This article presents preliminary work performed on a gold standard database and a cellphone platform. Results indicate that HR and HRV can be accurately assessed from acoustic recordings of heart sounds using only a cellphone and hands-free kit. Heart sound analysis software, which can run on a standard cellphone in real time, has been developed that detects S1 heart sounds with a sensitivity of 92.1% and a positive predictivity of 88.4%. Evaluation of data recorded from cellphones demonstrates that the low-frequency response (<100 Hz) is key to the success of heart sound analysis on cellphones. Noise rejection is also shown to be important. Introduction and Background There is a wide rural-urban divide in health care delivery, especially in developing nations. Medical specialists in these countries are scarce and are often only found in the cities. For people living in remote or resource-poor locations, travel to see these specialists can deprive them of a whole day’s income. For many rural clinics, the time it takes to send information to the nearest physician and receive a diagnosis and advice can take weeks. As a result, diagnosis and treatment are often delayed and patient follow-up is difficult when a long journey or wait time is involved, resulting in higher mortality and costs than are necessary. Although training programs exist to increase the numbers of community health workers, such programs are not scalable and sustainable, requiring constant resources, Copyright © 2009, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. the effectiveness of which is reduced as the knowledge radiates out from the centers of training. Maternal and childhood mortality is a particularly pressing issue. Each year, over half a million women die from pregnancy or childbirth (Richards 2009). Furthermore, women in least developed countries are 300 times more likely to die in childbirth. With proper prenatal care and routine screening, mothers can learn to take proper safety measures during pregnancy, including preparations for a high-risk delivery if necessary. Through simple monitoring of the mother and fetus (i.e. measuring fetal heartbeat and respiration), a healthcare worker would be able to check on the heart condition and general growth of the baby and any infections of the mother. Following childbirth, infections in the young children (such as TB) require detection. One promising method for such screening is through fetal and pediatric heart rate (HR) analysis. Changes in heart rate variability (HRV) have been shown to be linked to infection (Kovatchev et al 2003, Blad et al 2008, Frasch et al 2009). However, to-date all analyses of fetal and pediatric heart rate variability has been via ultrasound or electrocardiogram (ECG), with the exception of a prototype computer-base system in India (Mittra 2009), which uses high-end microphones to subtract ambient noise and render a heart rate. However, Mittra gives no details of the heart rate extraction or its accuracy on the small number of mothers tested. Moreover, a large amount of equipment is required to perform the screening. In the area of mobile diagnostics using cellular technologies there have been several recent developments. For example, Jin et al (2009) have developed a system to record ECGs via a cellphone. Tan and Masek (2009) have developed a system to interface with Doppler devices for fetal ultrasound assessment. Black et al (2009) have created a low cost pulse oximeter attached to a cellphone to try to distinguish pneumonia from other febrile illnesses. However, all these systems require a reliable supply chain infrastructure to deliver the diagnostic peripherals, maintenance and supplies. Furthermore, perhaps with the exception of the pulse oximeter, expert knowledge and training is usually required to use the equipment. In this article we present an analysis of heart-rate extraction from off-line acoustic data and compare them to that extracted from a clean electrocardiogram. We then present results from a preliminary cellphone-based system which performs the same operation. Finally we describe a telemedicine framework for collecting expert-labeled data, an essential requirement for training our system.