Deep Neural Network for Cardiac Magnetic Resonance Image Segmentation.

The analysis and interpretation of cardiac magnetic resonance (CMR) images are often time-consuming. The automated segmentation of cardiac structures can reduce the time required for image analysis. Spatial similarities between different CMR image types were leveraged to jointly segment multiple sequences using a segmentation model termed a multi-image type UNet (MI-UNet). This model was developed from 72 exams (46% female, mean age 63 ± 11 years) performed on patients with hypertrophic cardiomyopathy. The MI-UNet for steady-state free precession (SSFP) images achieved a superior Dice similarity coefficient (DSC) of 0.92 ± 0.06 compared to 0.87 ± 0.08 for a single-image type UNet (p < 0.001). The MI-UNet for late gadolinium enhancement (LGE) images also had a superior DSC of 0.86 ± 0.11 compared to 0.78 ± 0.11 for a single-image type UNet (p = 0.001). The difference across image types was most evident for the left ventricular myocardium in SSFP images and for both the left ventricular cavity and the left ventricular myocardium in LGE images. For the right ventricle, there were no differences in DCS when comparing the MI-UNet with single-image type UNets. The joint segmentation of multiple image types increases segmentation accuracy for CMR images of the left ventricle compared to single-image models. In clinical practice, the MI-UNet model may expedite the analysis and interpretation of CMR images of multiple types.

Incorporating RTLS-Based Spatiotemporal Information in Studying Physical Activities of Clinical Staff.

Clinicians and staff who work in intense hospital settings such as the emergency department (ED) are under an extended amount of mental and physical pressure every day. They may spend hours in active physical pressure to serve patients with severe injuries or stay in front of a computer to review patients' clinical history and update the patients' electronic health records (EHR). Nurses on the other hand may stay for multiple consecutive days of 9-12 working hours. The amount of pressure is so much that they usually end up taking days off to recover the lost energy. Both of these extreme cases of low and high physical activities are shown to affect the physical and mental health of clinicians and may even lead to fatigue and burnout.In this study Real-Time location systems (RTLS) are used for the first time, to study the amount of physical activity exerted by clinicians. RTLS systems have traditionally been used in hospital settings for locating staff and equipment, whereas our proposed method combines both time and location information together to estimate the duration, length, and speed of movements within hospital wards such as the ED. It is also our first step towards utilizing non-wearable devices to measure sedentary behavior inside the ED. This information helps to assess the workload on the care team and identify means to reduce the risk of performance compromise, fatigue, and burnout.We used one year worth of raw RFID data that covers movement records of 38 physicians, 13 residents, 163 nurses, 33 staff in the ED. We defined a walking path as the continuous sequences of movements and stops and identified separate walking paths for each individual on each day. Walking duration, distance, and speed, along with the number of steps and the duration of sedentary behavior, are then estimated for each walking path. We compared our results to the values reported in the literature and showed despite the low spatial resolution of RTLS, our non-invasive estimations are closely comparable to the ones measured by Fitbit or other wearable pedometers.Clinical Relevance- Adequate assessment of workload in a dynamic care delivery space plays an important role in ensuring safe and optimal care delivery [7]. Systems capable of measuring physical activities on a continuous basis during daily work can provide precious information for a variety of purposes including automated assessment of sedentary behaviors and early detection of work pressure. Such systems could help facilitate targeted changes in the number of staff, duration of their working shifts leading to a safer and healthier environment for both clinicians and patients.

NATURAL LANGUAGE PROCESSING BASED MACHINE LEARNING MODEL USING CARDIAC MRI REPORTS TO IDENTIFY HYPERTROPHIC CARDIOMYOPATHY PATIENTS.

Hypertrophic Cardiomyopathy (HCM) is the most common genetic heart disease in the US and is known to cause sudden death (SCD) in young adults. While significant advancements have been made in HCM diagnosis and management, there is a need to identify HCM cases from electronic health record (EHR) data to develop automated tools based on natural language processing guided machine learning (ML) models for accurate HCM case identification to improve management and reduce adverse outcomes of HCM patients. Cardiac Magnetic Resonance (CMR) Imaging, plays a significant role in HCM diagnosis and risk stratification. CMR reports, generated by clinician annotation, offer rich data in the form of cardiac measurements as well as narratives describing interpretation and phenotypic description. The purpose of this study is to develop an NLP-based interpretable model utilizing impressions extracted from CMR reports to automatically identify HCM patients. CMR reports of patients with suspected HCM diagnosis between the years 1995 to 2019 were used in this study. Patients were classified into three categories of yes HCM, no HCM and, possible HCM. A random forest (RF) model was developed to predict the performance of both CMR measurements and impression features to identify HCM patients. The RF model yielded an accuracy of 86% (608 features) and 85% (30 features). These results offer promise for accurate identification of HCM patients using CMR reports from EHR for efficient clinical management transforming health care delivery for these patients.

Trauma Activation Responsiveness: An RFID-enabled trauma flowsheet.

Manually documented trauma flow sheets contain critical information regarding trauma resuscitations in the emergency department (ED). The American College of Surgeons (ACS) has enforced certain thresholds on trauma surgeons' arrival time to the trauma bay. Due to the complex and fast-paced ED environment, this information can be easily overlooked or erroneously recorded, affecting compliance with ACS standards. This paper is a retrospective study conducted at a Level I trauma center equipped with an RFID system to investigate an automated solution to evaluate and improve the accuracy of measuring trauma surgeons' response time to the highest level (red) trauma activations.Clinical Relevance- Demonstration of timely response to trauma activations is required for ACS verification. As real-time location systems become more prevalent, they may improve a hospital's ability to report accurate response times for trauma team activations.

Explanatory Analysis of a Machine Learning Model to Identify Hypertrophic Cardiomyopathy Patients from EHR Using Diagnostic Codes.

Hypertrophic cardiomyopathy (HCM) is a genetic heart disease that is the leading cause of sudden cardiac death (SCD) in young adults. Despite the well-known risk factors and existing clinical practice guidelines, HCM patients are underdiagnosed and sub-optimally managed. Developing machine learning models on electronic health record (EHR) data can help in better diagnosis of HCM and thus improve hundreds of patient lives. Automated phenotyping using HCM billing codes has received limited attention in the literature with a small number of prior publications. In this paper, we propose a novel predictive model that helps physicians in making diagnostic decisions, by means of information learned from historical data of similar patients. We assembled a cohort of 11,562 patients with known or suspected HCM who have visited Mayo Clinic between the years 1995 to 2019. All existing billing codes of these patients were extracted from the EHR data warehouse. Target ground truth labeling for training the machine learning model was provided by confirmed HCM diagnosis using the gold standard imaging tests for HCM diagnosis echocardiography (echo), or cardiac magnetic resonance (CMR) imaging. As the result, patients were labeled into three categories of "yes definite HCM", "no HCM phenotype", and "possible HCM" after a manual review of medical records and imaging tests. In this study, a random forest was adopted to investigate the predictive performance of billing codes for the identification of HCM patients due to its practical application and expected accuracy in a wide range of use cases. Our model performed well in finding patients with "yes definite", "possible" and "no" HCM with an accuracy of 71%, weighted recall of 70%, the precision of 75%, and weighted F1 score of 72%. Furthermore, we provided visualizations based on multidimensional scaling and the principal component analysis to provide insights for clinicians' interpretation. This model can be used for the identification of HCM patients using their EHR data, and help clinicians in their diagnosis decision making.

Sleep Posture Classification Using Bed Sensor Data and Neural Networks.

Sleep posture has been shown to be important in monitoring health conditions such as congestive heart failure (CHF), sleep apnea, pressure ulcers, and even blood pressure abnormalities. In this paper, we investigate the use of four hydraulic bed transducers placed underneath the mattress to classify different sleep postures. For classification, we employed a simple neural network. Different combinations of parameters were studied to determine the best configuration. Data were collected on four major postures from 58 subjects. We report the results of classification for different combinations of these four postures. Both 10-Fold and Leave-One-Subject-Out (LOSO) Cross-validations (CV) were used to evaluate the accuracy of our predictions. Our results show that there are multiple configuration settings that make classification accuracy as high as 100% using k-Fold CV for all postures. Maximum classification accuracy after applying LOSO is 93% for a two-class classification of separating Left vs. Right lateral positions. The second-best classification accuracy with LOSO is 92% for the classification of lateral versus non-lateral.