Deep learning to estimate gestational age from fly-to cineloop videos: A novel approach to ultrasound quality control.

OBJECTIVE: Low-cost devices have made obstetric sonography possible in settings where it was previously unfeasible, but ensuring quality and consistency at scale remains a challenge. In the present study, we sought to create a tool to reduce substandard fetal biometry measurement while minimizing care disruption. METHODS: We developed a deep learning artificial intelligence (AI) model to estimate gestational age (GA) in the second and third trimester from fly-to cineloops-brief videos acquired during routine ultrasound biometry-and evaluated its performance in comparison to expert sonographer measurement. We then introduced random error into fetal biometry measurements and analyzed the ability of the AI model to flag grossly inaccurate measurements such as those that might be obtained by a novice. RESULTS: The mean absolute error (MAE) of our model (±standard error) was 3.87 ± 0.07 days, compared to 4.80 ± 0.10 days for expert biometry (difference -0.92 days; 95% CI: -1.10 to -0.76). Based on simulated novice biometry with average absolute error of 7.5%, our model reliably detected cases where novice biometry differed from expert biometry by 10 days or more, with an area under the receiver operating characteristics curve of 0.93 (95% CI: 0.92, 0.95), sensitivity of 81.0% (95% CI: 77.9, 83.8), and specificity of 89.9% (95% CI: 88.1, 91.5). These results held across a range of sensitivity analyses, including where the model was provided suboptimal truncated fly-to cineloops. CONCLUSIONS: Our AI model estimated GA more accurately than expert biometry. Because fly-to cineloop videos can be obtained without any change to sonographer workflow, the model represents a no-cost guardrail that could be incorporated into both low-cost and commercial ultrasound devices to prevent reporting of most gross GA estimation errors.

Weight management in young adults with type 1 diabetes: The advancing care for type 1 diabetes and obesity network sequential multiple assignment randomized trial pilot results.

AIMS: Co-management of weight and glycaemia is critical yet challenging in type 1 diabetes (T1D). We evaluated the effect of a hypocaloric low carbohydrate, hypocaloric moderate low fat, and Mediterranean diet without calorie restriction on weight and glycaemia in young adults with T1D and overweight or obesity. MATERIALS AND METHODS: We implemented a 9-month Sequential, Multiple Assignment, Randomized Trial pilot among adults aged 19-30 years with T1D for ≥1 year and body mass index 27-39.9 kg/m2 . Re-randomization occurred at 3 and 6 months if the assigned diet was not acceptable or not effective. We report results from the initial 3-month diet period and re-randomization statistics before shutdowns due to COVID-19 for primary [weight, haemoglobin A1c (HbA1c), percentage of time below range <70 mg/dl] and secondary outcomes [body fat percentage, percentage of time in range (70-180 mg/dl), and percentage of time below range <54 mg/dl]. Models adjusted for design, demographic and clinical covariates tested changes in outcomes and diet differences. RESULTS: Adjusted weight and HbA1c (n = 38) changed by -2.7 kg (95% CI -3.8, -1.5, P < .0001) and -0.91 percentage points (95% CI -1.5, -0.30, P = .005), respectively, while adjusted body fat percentage remained stable, on average (P = .21). Hypoglycaemia indices remained unchanged following adjustment (n = 28, P > .05). Variability in all outcomes, including weight change, was considerable (57.9% were re-randomized primarily due to loss of <2% body weight). No outcomes varied by diet. CONCLUSIONS: Three months of a diet, irrespective of macronutrient distribution or caloric restriction, resulted in weight loss while improving or maintaining HbA1c levels without increasing hypoglycaemia in adults with T1D.

Design of the Advancing Care for Type 1 Diabetes and Obesity Network energy metabolism and sequential multiple assignment randomized trial nutrition pilot studies: An integrated approach to develop weight management solutions for individuals with type 1 diabetes.

Young adults with type 1 diabetes (T1D) often have difficulty co-managing weight and glycemia. The prevalence of overweight and obesity among individuals with T1D now parallels that of the general population and contributes to dyslipidemia, insulin resistance, and risk for cardiovascular disease. There is a compelling need to develop a program of research designed to optimize two key outcomes-weight management and glycemia-and to address the underlying metabolic processes and behavioral challenges unique to people with T1D. For an intervention addressing these dual outcomes to be effective, it must be appropriate to the unique metabolic phenotype of T1D, and to biological and behavioral responses to glycemia (including hypoglycemia) that relate to weight management. The intervention must also be safe, feasible, and accepted by young adults with T1D. In 2015, we established a consortium called ACT1ON: Advancing Care for Type 1 Diabetes and Obesity Network, a transdisciplinary team of scientists at multiple institutions. The ACT1ON consortium designed a multi-phase study which, in parallel, evaluated the mechanistic aspects of the unique metabolism and energy requirements of individuals with T1D, alongside a rigorous adaptive behavioral intervention to simultaneously facilitate weight management while optimizing glycemia. This manuscript describes the design of our integrative study-comprised of an inpatient mechanistic phase and an outpatient behavioral phase-to generate metabolic, behavioral, feasibility, and acceptability data to support a future, fully powered sequential, multiple assignment, randomized trial to evaluate the best approaches to prevent and treat obesity while co-managing glycemia in people with T1D. Clinicaltrials.gov identifiers: NCT03651622 and NCT03379792. The present study references can be found here: https://clinicaltrials.gov/ct2/show/NCT03651622 https://clinicaltrials.gov/ct2/show/NCT03379792?term=NCT03379792&draw=2&rank=1 Submission Category: "Study Design, Statistical Design, Study Protocols".

AI Estimation of Gestational Age from Blind Ultrasound Sweeps in Low-Resource Settings.

BACKGROUND: Ultrasound is indispensable to gestational age estimation and thus to quality obstetrical care, yet high equipment cost and the need for trained sonographers limit its use in low-resource settings. METHODS: From September 2018 through June 2021, we recruited 4695 pregnant volunteers in North Carolina and Zambia and obtained blind ultrasound sweeps (cineloop videos) of the gravid abdomen alongside standard fetal biometry. We trained a neural network to estimate gestational age from the sweeps and, in three test data sets, assessed the performance of the artificial intelligence (AI) model and biometry against previously established gestational age. RESULTS: In our main test set, the mean absolute error (MAE) (±SE) was 3.9±0.12 days for the model versus 4.7±0.15 days for biometry (difference, -0.8 days; 95% confidence interval [CI], -1.1 to -0.5; P<0.001). The results were similar in North Carolina (difference, -0.6 days; 95% CI, -0.9 to -0.2) and Zambia (-1.0 days; 95% CI, -1.5 to -0.5). Findings were supported in the test set of women who conceived by in vitro fertilization (MAE of 2.8±0.28 vs. 3.6±0.53 days for the model vs. biometry; difference, -0.8 days; 95% CI, -1.7 to 0.2) and in the set of women from whom sweeps were collected by untrained users with low-cost, battery-powered devices (MAE of 4.9±0.29 vs. 5.4±0.28 days for the model vs. biometry; difference, -0.6; 95% CI, -1.3 to 0.1). CONCLUSIONS: When provided blindly obtained ultrasound sweeps of the gravid abdomen, our AI model estimated gestational age with accuracy similar to that of trained sonographers conducting standard fetal biometry. Model performance appears to extend to blind sweeps collected by untrained providers in Zambia using low-cost devices. (Funded by the Bill and Melinda Gates Foundation.).

Sequence to Sequence ECG Cardiac Rhythm Classification Using Convolutional Recurrent Neural Networks.

This paper proposes a novel deep learning architecture involving combinations of Convolutional Neural Networks (CNN) layers and Recurrent neural networks (RNN) layers that can be used to perform segmentation and classification of 5 cardiac rhythms based on ECG recordings. The algorithm is developed in a sequence to sequence setting where the input is a sequence of five second ECG signal sliding windows and the output is a sequence of cardiac rhythm labels. The novel architecture processes as input both the spectrograms of the ECG signal as well as the heartbeats' signal waveform. Additionally, we are able to train the model in the presence of label noise. The model's performance and generalizability is verified on an external database different from the one we used to train. Experimental result shows this approach can achieve an average F1 scores of 0.89 (averaged across 5 classes). The proposed model also achieves comparable classification performance to existing state-of-the-art approach with considerably less number of training parameters.

The Impact of Racial and Ethnic Health Disparities in Diabetes Management on Clinical Outcomes: A Reinforcement Learning Analysis of Health Inequity Among Youth and Young Adults in the SEARCH for Diabetes in Youth Study.

OBJECTIVE: To estimate difference in population-level glycemic control and the emergence of diabetes complications given a theoretical scenario in which non-White youth and young adults (YYA) with type 1 diabetes (T1D) receive and follow an equivalent distribution of diabetes treatment regimens as non-Hispanic White YYA. RESEARCH DESIGN AND METHODS: Longitudinal data from YYA diagnosed 2002-2005 in the SEARCH for Diabetes in Youth Study were analyzed. Based on self-reported race/ethnicity, YYA were classified as non-White race or Hispanic ethnicity (non-White subgroup) versus non-Hispanic White race (White subgroup). In the White versus non-White subgroups, the propensity score models estimated treatment regimens, including patterns of insulin modality, self-monitored glucose frequency, and continuous glucose monitoring use. An analysis based on policy evaluation techniques in reinforcement learning estimated the effect of each treatment regimen on mean hemoglobin A1c (HbA1c) and the prevalence of diabetes complications for non-White YYA. RESULTS: The study included 978 YYA. The sample was 47.5% female and 77.5% non-Hispanic White, with a mean age of 12.8 ± 2.4 years at diagnosis. The estimated population mean of longitudinal average HbA1c over visits was 9.2% and 8.2% for the non-White and White subgroup, respectively (difference of 0.9%). Within the non-White subgroup, mean HbA1c across visits was estimated to decrease by 0.33% (95% CI -0.45, -0.21) if these YYA received the distribution of diabetes treatment regimens of the White subgroup, explaining ∼35% of the estimated difference between the two subgroups. The non-White subgroup was also estimated to have a lower risk of developing diabetic retinopathy, diabetic kidney disease, and peripheral neuropathy with the White youth treatment regimen distribution (P < 0.05), although the low proportion of YYA who developed complications limited statistical power for risk estimations. CONCLUSIONS: Mathematically modeling an equalized distribution of T1D self-management tools and technology accounted for part of but not all disparities in glycemic control between non-White and White YYA, underscoring the complexity of race and ethnicity-based health inequity.

Limiting adverse birth outcomes in resource-limited settings (LABOR): protocol of a prospective intrapartum cohort study.

Background: Each year, nearly 300,000 women and 5 million fetuses or neonates die during childbirth or shortly thereafter, a burden concentrated disproportionately in low- and middle-income countries. Identifying women and their fetuses at risk for intrapartum-related morbidity and death could facilitate early intervention. Methods: The Limiting Adverse Birth Outcomes in Resource-Limited Settings (LABOR) Study is a multi-country, prospective, observational cohort designed to exhaustively document the course and outcomes of labor, delivery, and the immediate postpartum period in settings where adverse outcomes are frequent. The study is conducted at four hospitals across three countries in Ghana, India, and Zambia. We will enroll approximately 12,000 women at presentation to the hospital for delivery and follow them and their fetuses/newborns throughout their labor and delivery course, postpartum hospitalization, and up to 42 days thereafter. The co-primary outcomes are composites of maternal (death, hemorrhage, hypertensive disorders, infection) and fetal/neonatal adverse events (death, encephalopathy, sepsis) that may be attributed to the intrapartum period. The study collects extensive physiologic data through the use of physiologic sensors and employs medical scribes to document examination findings, diagnoses, medications, and other interventions in real time. Discussion: The goal of this research is to produce a large, sharable dataset that can be used to build statistical algorithms to prospectively stratify parturients according to their risk of adverse outcomes. We anticipate this research will inform the development of new tools to reduce peripartum morbidity and mortality in low-resource settings.