Background Cardiac MRI is limited by long acquisition times, yet faster acquisition of smaller-matrix images reduces spatial detail. Deep learning (DL) might enable both faster acquisition and higher spatial detail via super-resolution. Purpose To explore the feasibility of using DL to enhance spatial detail from small-matrix MRI acquisitions and evaluate its performance against that of conventional image upscaling methods. Materials and Methods Short-axis cine cardiac MRI examinations performed between January 2012 and December 2018 at one institution were retrospectively collected for algorithm development and testing. Convolutional neural networks (CNNs), a form of DL, were trained to perform super resolution in image space by using synthetically generated low-resolution data. There were 70%, 20%, and 10% of examinations allocated to training, validation, and test sets, respectively. CNNs were compared against bicubic interpolation and Fourier-based zero padding by calculating the structural similarity index (SSIM) between high-resolution ground truth and each upscaling method. Means and standard deviations of the SSIM were reported, and statistical significance was determined by using the Wilcoxon signed-rank test. For evaluation of clinical performance, left ventricular volumes were measured, and statistical significance was determined by using the paired Student t test. Results For CNN training and retrospective analysis, 400 MRI scans from 367 patients (mean age, 48 years ± 18; 214 men) were included. All CNNs outperformed zero padding and bicubic interpolation at upsampling factors from two to 64 (P < .001). CNNs outperformed zero padding on more than 99.2% of slices (9828 of 9907). In addition, 10 patients (mean age, 51 years ± 22; seven men) were prospectively recruited for super-resolution MRI. Super-resolved low-resolution images yielded left ventricular volumes comparable to those from full-resolution images (P > .05), and super-resolved full-resolution images appeared to further enhance anatomic detail. Conclusion Deep learning outperformed conventional upscaling methods and recovered high-frequency spatial information. Although training was performed only on short-axis cardiac MRI examinations, the proposed strategy appeared to improve quality in other imaging planes. © RSNA, 2020 Online supplemental material is available for this article.

Purpose

Delayed enhancement imaging is an essential component of cardiac MRI, which is used widely for the evaluation of myocardial scar and viability. The selection of an optimal inversion time (TI) or null point (TI_NP ) to suppress the background myocardial signal is required. The purpose of this study was to assess the feasibility of automated selection of TI_NP using a convolutional neural network (CNN). We hypothesized that a CNN may use spatial and temporal imaging characteristics from an inversion-recovery scout to select TI_NP , without the aid of a human observer.

Methods

We retrospectively collected 425 clinically acquired cardiac MRI exams performed at 1.5 T that included inversion-recovery scout acquisitions. We developed a VGG19 classifier ensembled with long short-term memory to identify the TI_NP . We compared the performance of the ensemble CNN in predicting TI_NP against ground truth, using linear regression analysis. Ground truth was defined as the expert physician annotation of the optimal TI. In a backtrack approach, saliency maps were generated to interpret the classification outcome and to increase the model's transparency.

Results

Prediction of TI_NP from our ensemble VGG19 long short-term memory closely matched with expert annotation (ρ = 0.88). Ninety-four percent of the predicted TI_NP were within ±36 ms, and 83% were at or after expert TI selection.

Conclusion

In this study, we show that a CNN is capable of automated prediction of myocardial TI from an inversion-recovery experiment. Merging the spatial and temporal characteristics of the VGG-19 and long short-term-memory CNN structures appears to be sufficient to predict myocardial TI from TI scout.

Background

Fall-related injuries are a well-described cause of morbidity and mortality in the community-dwelling elderly population, but have not been well described in patients with cancer. Cancer treatment with chemotherapy can result in many unwanted side effects, including peripheral neuropathy if the drugs are potentially neurotoxic. Peripheral neuropathy and other side effects of chemotherapy may lead to an increased risk of fall-related injuries.

Methods

We conducted a retrospective cohort analysis using the records of 65,311 patients with breast, colon, lung, or prostate cancer treated with chemotherapy in the SEER-Medicare database from 1994 to 2007. The primary outcome was any fall-related injury defined as a traumatic fracture, dislocation, or head injury within 12 months of the first dose of chemotherapy. The sample population was divided into 3 cohorts based on whether they most frequently received a neurotoxic doublet, single agent, or a non-neurotoxic chemotherapy. Cox proportional-hazards analyses were adjusted for baseline characteristics to determine the risk of fall-related injuries among the 3 cohorts.

Results

The rate of fall-related injuries for patients receiving a doublet of neurotoxic chemotherapy (9.15 per 1000 person-months) was significantly higher than for those receiving a single neurotoxic agent (7.76 per 1000 person-months) or a non-neurotoxic agent (5.19 per 1000 person-months). Based on the Cox proportional-hazards model risk of fall-related injuries was highest for the cohort receiving a neurotoxic doublet after the model was adjusted for baseline characteristics.

Conclusions

Among elderly patients with cancer, use of neurotoxic chemotherapy is associated with an increased risk of fall-related injuries.

Background

The Health Utilities Index Mark 3 (HUI3) is a tool composed of 41 questions, covering 8 attributes: vision, hearing, speech, ambulation, dexterity, emotion, cognition, and pain. Responses to these questions can define more than 972,000 health situations. This tool allows respondents to answer "Don't Know," for which there is no scoring instruction, to any given question. This situation creates a break in the scoring algorithm and leads to considerable amounts of missing data. The goal of this study is to develop strategies to deal with HUI3 scores for participants who have missing data.

Methods

The authors used data from 248 individuals enrolled in the Cataract Management Trial, focusing on the HUI3 vision and ambulation attributes, which had 19% and 10% of attribute levels missing, respectively. Inspection and deduction were used to fill in values independent of the value of the missing data, then alternative analytic techniques were compared, including mean substitution, model scoring, hot deck, multiple imputation, and regression imputation.

Results

Inspection and logical deduction reduced the percentage of missing information in the HUI3 by 49% to 87%. A comparison of analytic techniques used for the remaining HUI3 vision data missing demonstrated the value of building models based on internal response patterns and that simple analytic techniques fare as well as more complicated ones when the number of missing cases is small.

Conclusion

Analyzing the pattern of responses in cases where the attribute level score is missing reduces the amount of missing data and can simplify the analytic process for the remaining missing data.

Purpose

To develop and evaluate a system to prescribe imaging planes for cardiac MRI based on deep learning (DL)-based localization of key anatomic landmarks.

Materials and methods

Annotated landmarks on 892 long-axis (LAX) and 493 short-axis (SAX) cine steady-state free precession series from cardiac MR images were retrospectively collected between February 2012 and June 2017. U-Net-based heatmap regression was used for localization of cardiac landmarks, which were used to compute cardiac MRI planes. Performance was evaluated by comparing localization distances and plane angle differences between DL predictions and ground truth. The plane angulations from DL were compared with those prescribed by the technologist at the original time of acquisition. Data were split into 80% for training and 20% for testing, and results confirmed with fivefold cross-validation.

Results

On LAX images, DL localized the apex within mean 12.56 mm ± 19.11 (standard deviation) and the mitral valve (MV) within 7.68 mm ± 6.91. On SAX images, DL localized the aortic valve within 5.78 mm ± 5.68, MV within 5.90 mm ± 5.24, pulmonary valve within 6.55 mm ± 6.39, and tricuspid valve within 6.39 mm ± 5.89. On the basis of these localizations, average angle bias and mean error of DL-predicted imaging planes relative to ground truth annotations were as follows: SAX, -1.27° ± 6.81 and 4.93° ± 4.86; four chambers, 0.38° ± 6.45 and 5.16° ± 3.80; three chambers, 0.13° ± 12.70 and 9.02° ± 8.83; and two chamber, 0.25° ± 9.08 and 6.53° ± 6.28, respectively.

Conclusion

DL-based anatomic localization is a feasible strategy for planning cardiac MRI planes. This approach can produce imaging planes comparable to those defined by ground truth landmarks.© RSNA, 2019 Supplemental material is available for this article.

Risk prediction models are crucial for assessing the pretest probability of cancer and are applied to stratify patient management strategies. These models are frequently based on multivariate regression analysis, requiring that all risk factors be specified, and do not convey the confidence in their predictions. We present a framework for uncertainty analysis that accounts for variability in input values. Uncertain or missing values are replaced with a range of plausible values. These ranges are used to compute individualized risk confidence intervals. We demonstrate our approach using the Gail model to evaluate the impact of uncertainty on management decisions. Up to 13% of cases (uncertain) had a risk interval that falls within the decision threshold (e.g., 1.67% 5-year absolute risk). A small number of cases changed from low- to high-risk when missing values were present. Our analysis underscores the need for better communication of input assumptions that influence the resulting predictions.

Wearable devices and fitness trackers have gained popularity in healthcare and telemedicine as tools to reduce hospitalization costs, improve personalized health management, and monitor patients in remote areas. Smartwatches, particularly, offer continuous monitoring capabilities through step counting, heart rate tracking, and activity monitoring. However, despite being recognized as an emerging technology, the adoption of smartwatches in patient monitoring systems is still at an early stage, with limited studies delving beyond their feasibility. Developing healthcare applications for smartwatches faces challenges such as short battery life, wearable comfort, patient compliance, termination of non-native applications, user interaction difficulties, small touch screens, personalized sensor configuration, and connectivity with other devices. This paper presents a case study on designing an Android smartwatch application for remote monitoring of geriatric patients. It highlights obstacles encountered during app development and offers insights into design decisions and implementation details. The aim is to assist programmers in developing more efficient healthcare applications for wearable systems.

Background

Febrile neutropenia is one of the most common oncologic emergencies and is associated with significant, preventable morbidity and mortality. Most patients who experience a febrile neutropenia episode are hospitalized, resulting in significant economic cost.

Objective

This exploratory study implemented a remote monitoring system comprising a digital infrared thermometer and a pulse oximeter with the capability to notify providers in real time of abnormalities in vital signs that could suggest early clinical deterioration and thereby improve clinical outcomes.

Methods

The remote monitoring system was implemented and compared to standard-of-care vital signs monitoring in hospitalized patients with underlying hematologic malignancies complicated by a febrile neutropenia episode in order to assess the feasibility and validity of the system. Statistical analysis was performed using the intraclass correlation coefficient (ICC) to assess the consistency between the measurements taken using traditional methods and those taken with the remote monitoring system for each of the vital sign parameters (temperature, heart rate, and oxygen saturation). A linear mixed-effects model with a random subject effect was used to estimate the variance components. Bland-Altman plots were created for the parameters to further delineate the direction of any occurring bias.

Results

A total of 23 patients were enrolled in the study (mean age 56, SD 23-75 years; male patients: n=11, 47.8%). ICC analysis confirmed the high repeatability and accuracy of the heart rate assessment (ICC=0.856), acting as a supplement to remote temperature assessment. While the sensitivity and specificity for capturing tachycardia above a rate of 100 bpm were excellent (88% and 97%, respectively), the sensitivity of the remote monitoring system in capturing temperatures >37.8 °C and oxygen saturation <92% was 45% and 50%, respectively.

Conclusions

Overall, this novel approach using temperature, heart rate, and oxygen saturation assessments successfully provided real-time, clinically valuable feedback to providers. While temperature and oxygen saturation assessments lagged in terms of sensitivity compared to a standard in-hospital system, the heart rate assessment provided highly accurate complementary data. As a whole, the system provided additional information that can be applied to a clinically vulnerable population. By transitioning its application to high-risk patients in the outpatient setting, this system can help prevent additional use of health care services through early provider intervention and potentially improve outcomes.

Problem

With the growth in risk-based and accountable care organization contracts, creating value by redesigning care to reduce costs and improve outcomes and the patient experience has become an urgent priority for health care systems.

Approach

In 2016, UCLA (University of California, Los Angeles) Health implemented a system-wide population health approach to identify patient populations with high expenses and promote proactive, value-based care. The authors created the Patient Health Value framework to guide value creation: (1) identify patient populations with high expenses and reasons for spending, (2) create design teams to understand the patient story, (3) create custom analytics and spending-based risk stratification, and (4) develop care pathways based on spending risk tiers. Primary care patients with three chronic conditions-dementia, chronic kidney disease (CKD), and cancer-were identified as high-cost subpopulations.

Outcomes

For each patient subpopulation, a multispecialty, multidisciplinary design team identified reasons for spending and created care pathways to meet patient needs according to spending risk. Larger, lower-risk cohorts received necessary but less intensive interventions, while smaller, higher-risk cohorts received more intensive interventions. Preliminary analyses showed a 1% monthly decrease in inpatient bed day utilization among dementia patients (incident rate ratio [IRR] 0.99, P < .03) and a 2% monthly decrease in hospitalizations (IRR 0.98, P < .001) among CKD patients.

Next steps

Use of the Patient Health Value framework is expanding across other high-cost subpopulations with chronic conditions. UCLA Health is using the framework to organize care across specialties, build capacity, and grow a culture for value.