Daniel Chamberlain

Application of semi-supervised deep learning to lung sound analysis

The analysis of lung sounds, collected through auscultation, is a fundamental component of pulmonary disease diagnostics for primary care and general patient monitoring for telemedicine. Despite advances in computation and algorithms, the goal of automated lung sound identification and classification has remained elusive. Over the past 40 years, published work in this field has demonstrated only limited success in identifying lung sounds, with most published studies using only a small numbers of patients (typically N<;20) and usually limited to a single type of lung sound. Larger research studies have also been impeded by the challenge of labeling large volumes of data, which is extremely labor-intensive. In this paper, we present the development of a semi-supervised deep learning algorithm for automatically classify lung sounds from a relatively large number of patients (N=284). Focusing on the two most common lung sounds, wheeze and crackle, we present results from 11,627 sound files recorded from 11 different auscultation locations on these 284 patients with pulmonary disease. 890 of these sound files were labeled to evaluate the model, which is significantly larger than previously published studies. Data was collected with a custom mobile phone application and a low-cost (US

Mobile stethoscope and signal processing algorithms for pulmonary screening and diagnostics

30) electronic stethoscope. On this data set, our algorithm achieves ROC curves with AUCs of 0.86 for wheeze and 0.74 for crackle. Most importantly, this study demonstrates how semi-supervised deep learning can be used with larger data sets without requiring extensive labeling of data.

Making Sense of a Negative Clinical Trial Result: A Bayesian Analysis of a Clinical Trial of Lorazepam and Diazepam for Pediatric Status Epilepticus.

Pulmonary diseases represent a large disease burden in terms of morbidity and mortality worldwide. For many reasons, including household air pollution and a shortage of trained doctors, this burden is concentrated in the developing world. The standard diagnostic pathway for pulmonary diseases is prohibitively expensive in developing countries, so these diseases are often misdiagnosed or underdiagnosed. To assist doctors and health workers, there is a need to create tools that can automatically recognize specific lung sounds and provide diagnostic guidance. As a first step towards this long-term goal, we have created a low-cost stethoscope and smartphone application to record lung sounds. We discuss problems we encountered with the initial design and demonstrate an improved design that is currently being used in the field. We also demonstrate an algorithm capable of automatic detection of wheeze sounds. The automatic wheeze detection algorithm uses time-frequency analysis and the Short Time Fourier Transform to identify sections of wheezing in recorded lung sound files. Unlike most published sound classification studies, we trained and tested our algorithms using sound data collected from 38 actual patients at a pulmonary clinic in Pune, India. Despite variability in the quality of the data, our algorithm demonstrated an accuracy of 86% for successfully detecting the presence of wheeze in a sound file. This mobile platform and detection algorithm demonstrates an important step in creating an automated platform for the diagnosis of pulmonary diseases in a real-world setting.

A mobile platform for automated screening of asthma and chronic obstructive pulmonary disease

Study objective: We demonstrate the application of a Bayesian approach to a recent negative clinical trial result. A Bayesian analysis of such a trial can provide a more useful interpretation of results and can incorporate previous evidence. Methods: This was a secondary analysis of the efficacy and safety results of the Pediatric Seizure Study, a randomized clinical trial of lorazepam versus diazepam for pediatric status epilepticus. We included the published results from the only prospective pediatric study of status in a Bayesian hierarchic model, and we performed sensitivity analyses on the amount of pooling between studies. We evaluated 3 summary analyses for the results: superiority, noninferiority (margin <–10%), and practical equivalence (within ±10%). Results: Consistent with the original studys classic analysis of study results, we did not demonstrate superiority of lorazepam over diazepam. There is a 95% probability that the true efficacy of lorazepam is in the range of 66% to 80%. For both the efficacy and safety outcomes, there was greater than 95% probability that lorazepam is noninferior to diazepam, and there was greater than 90% probability that the 2 medications are practically equivalent. The results were largely driven by the current study because of the sample sizes of our study (n=273) and the previous pediatric study (n=61). Conclusion: Because Bayesian analysis estimates the probability of one or more hypotheses, such an approach can provide more useful information about the meaning of the results of a negative trial outcome. In the case of pediatric status epilepticus, it is highly likely that lorazepam is noninferior and practically equivalent to diazepam.

Applying Augmented Reality to Enable Automated and Low-Cost Data Capture from Medical Devices

Chronic Obstructive Pulmonary Disease (COPD) and asthma each represent a large proportion of the global disease burden; COPD is the third leading cause of death worldwide and asthma is one of the most prevalent chronic diseases, afflicting over 300 million people. Much of this burden is concentrated in the developing world, where patients lack access to physicians trained in the diagnosis of pulmonary disease. As a result, these patients experience high rates of underdiagnosis and misdiagnosis. To address this need, we present a mobile platform capable of screening for Asthma and COPD. Our solution is based on a mobile smart phone and consists of an electronic stethoscope, a peak flow meter application, and a patient questionnaire. This data is combined with a machine learning algorithm to identify patients with asthma and COPD. To test and validate the design, we collected data from 119 healthy and sick participants using our custom mobile application and ran the analysis on a PC computer. For comparison, all subjects were examined by an experienced pulmonologist using a full pulmonary testing laboratory. Employing a two-stage logistic regression model, our algorithms were first able to identify patients with either asthma or COPD from the general population, yielding an ROC curve with an AUC of 0.95. Then, after identifying these patients, our algorithm was able to distinguish between patients with asthma and patients with COPD, yielding an ROC curve with AUC of 0.97. This work represents an important milestone towards creating a self-contained mobile phone-based platform that can be used for screening and diagnosis of pulmonary disease in many parts of the world.

Implementation of smart phone video plethysmography and dependence on lighting parameters

As an alternative to building custom electronic devices that connect to mobile phones (via Bluetooth or USB), we present a new approach using Augmented Reality (AR) and machine vision to digitally recognize a biomedical device and capture readings automatically. In the context of developing countries, this approach enables easy integration with low-cost devices, without the need for designing any electronics or obtaining new FDA regulatory approval. As an example, we illustrate the use of AR with a peak flow meter, a device used in the diagnosis and treatment of respiratory disease. In our mobile application, the AR graphic overlay is used to provide feedback to patients and doctors by displaying personalized reference values. Comparing the automated readings from this device to manual readings, our mobile application had a mean error of 5.8 L/min and a correlation of 0.99. A small user study was also conducted in an India field clinic with three health staff (two nurses and a doctor). Following one minute of instruction, the automated readings from the participants had a mean error of 5.5 L/min and a correlation of 0.99 compared to manual readings, with a median task duration of 17.5 seconds. This small case study illustrates how AR can be used to capture medical device data on a mobile phone and help automate the data recording tasks performed by health workers in developing countries. This technology can also be used in developed countries, enabling patients to automatically record readings from similar devices at home using their smart phones.

Smart phone-based auscultation platform

The remote measurement of heart rate (HR) and heart rate variability (HRV) via a digital camera (video plethysmography) has emerged as an area of great interest for biomedical and health applications. While a few implementations of video plethysmography have been demonstrated on smart phones under controlled lighting conditions, it has been challenging to create a general scalable solution due to the large variability in smart phone hardware performance, software architecture, and the variable response to lighting parameters. In this context, we present a selfcontained smart phone implementation of video plethysmography for Android OS, which employs both stochastic and deterministic algorithms, and we use this to study the effect of lighting parameters (illuminance, color spectrum) on the accuracy of the remote HR measurement. Using two different phone models, we present the median HR error for five different video plethysmography algorithms under three different types of lighting (natural sunlight, compact fluorescent, and halogen incandescent) and variations in brightness. For most algorithms, we found the optimum light brightness to be in the range 1000-4000 lux and the optimum lighting types to be compact fluorescent and natural light. Moderate errors were found for most algorithms with some devices under conditions of low-brightness (<;500 lux) and highbrightness (>4000 lux). Our analysis also identified camera frame rate jitter as a major source of variability and error across different phone models, but this can be largely corrected through non-linear resampling. Based on testing with six human subjects, our real-time Android implementation successfully predicted the measured HR with a median error of -0.31 bpm, and an inter-quartile range of 2.1bpm.