Improving the Generalizability and Performance of an Ultrasound Deep Learning Model Using Limited Multicenter Data for Lung Sliding Artifact Identification.

Deep learning (DL) models for medical image classification frequently struggle to generalize to data from outside institutions. Additional clinical data are also rarely collected to comprehensively assess and understand model performance amongst subgroups. Following the development of a single-center model to identify the lung sliding artifact on lung ultrasound (LUS), we pursued a validation strategy using external LUS data. As annotated LUS data are relatively scarce-compared to other medical imaging data-we adopted a novel technique to optimize the use of limited external data to improve model generalizability. Externally acquired LUS data from three tertiary care centers, totaling 641 clips from 238 patients, were used to assess the baseline generalizability of our lung sliding model. We then employed our novel Threshold-Aware Accumulative Fine-Tuning (TAAFT) method to fine-tune the baseline model and determine the minimum amount of data required to achieve predefined performance goals. A subgroup analysis was also performed and Grad-CAM++ explanations were examined. The final model was fine-tuned on one-third of the external dataset to achieve 0.917 sensitivity, 0.817 specificity, and 0.920 area under the receiver operator characteristic curve (AUC) on the external validation dataset, exceeding our predefined performance goals. Subgroup analyses identified LUS characteristics that most greatly challenged the model's performance. Grad-CAM++ saliency maps highlighted clinically relevant regions on M-mode images. We report a multicenter study that exploits limited available external data to improve the generalizability and performance of our lung sliding model while identifying poorly performing subgroups to inform future iterative improvements. This approach may contribute to efficiencies for DL researchers working with smaller quantities of external validation data.

Ten Influential Point-of-Care Ultrasound Papers: 2023 in Review.

In an effort to help keep busy clinicians up to date with the latest ultrasound research, our group of experts has selected 10 influential papers from the past 12 months and provided a short summary of each. We hope to provide emergency physicians, intensivists, and other acute care providers with a succinct update concerning some key areas of ultrasound interest.

Phase-3 trial of recombinant human alkaline phosphatase for patients with sepsis-associated acute kidney injury (REVIVAL).

PURPOSE: Ilofotase alfa is a human recombinant alkaline phosphatase with reno-protective effects that showed improved survival and reduced Major Adverse Kidney Events by 90 days (MAKE90) in sepsis-associated acute kidney injury (SA-AKI) patients. REVIVAL, was a phase-3 trial conducted to confirm its efficacy and safety. METHODS: In this international double-blinded randomized-controlled trial, SA-AKI patients were enrolled < 72 h on vasopressor and < 24 h of AKI. The primary endpoint was 28-day all-cause mortality. The main secondary endpoint was MAKE90, other secondary endpoints were (i) days alive and free of organ support through day 28, (ii) days alive and out of the intensive care unit (ICU) through day 28, and (iii) time to death through day 90. Prior to unblinding, the statistical analysis plan was amended, including an updated MAKE90 definition. RESULTS: Six hundred fifty patients were treated and analyzed for safety; and 649 for efficacy data (ilofotase alfa n = 330; placebo n = 319). The observed mortality rates in the ilofotase alfa and placebo groups were 27.9% and 27.9% at 28 days, and 33.9% and 34.8% at 90 days. The trial was stopped for futility on the primary endpoint. The observed proportion of patients with MAKE90A and MAKE90B were 56.7% and 37.4% in the ilofotase alfa group vs. 64.6% and 42.8% in the placebo group. Median [interquartile range (IQR)] days alive and free of organ support were 17 [0-24] and 14 [0-24], number of days alive and discharged from the ICU through day 28 were 15 [0-22] and 10 [0-22] in the ilofotase alfa and placebo groups, respectively. Adverse events were reported in 67.9% and 75% patients in the ilofotase and placebo group. CONCLUSION: Among critically ill patients with SA-AKI, ilofotase alfa did not improve day 28 survival. There may, however, be reduced MAKE90 events. No safety concerns were identified.

Multiorgan Point-of-Care Ultrasound Assessment in Critically Ill Adults.

Traditional point-of-care ultrasound (POCUS) training highlights discrete techniques, single-organ assessment, and focused protocols. More recent developments argue for a whole-body approach, where the experienced clinician-ultrasonographer crafts a personalized POCUS protocol depending on specific clinical circumstances. This article describes this problem-based approach, focusing on common acute care scenarios while highlighting practical considerations and performance characteristics.

High and intermediate risk pulmonary embolism in the ICU.

Pulmonary embolism (PE) is a common and important medical emergency, encountered by clinicians across all acute care specialties. PE is a relatively uncommon cause of direct admission to the intensive care unit (ICU), but these patients are at high risk of death. More commonly, patients admitted to ICU develop PE as a complication of an unrelated acute illness. This paper reviews the epidemiology, diagnosis, risk stratification, and particularly the management of PE from a critical care perspective. Issues around prevention, anticoagulation, fibrinolysis, catheter-based techniques, surgical embolectomy, and extracorporeal support are discussed.

Association of age with death and withdrawal of life-sustaining therapy after severe traumatic brain injury.

BACKGROUND: Compared to younger age, older age (≥ 65 yr) is associated with worse outcomes after severe traumatic brain injury (TBI). We sought to describe the association of older age with in-hospital death and aggressiveness of intervention. METHODS: We conducted a retrospective cohort study of adult (age ≥ 16 yr) patients with severe TBI admitted to a single academic tertiary care neurotrauma centre between January 2014 and December 2015. We collected data through chart review as well as from our institutional administrative database. We provided descriptive statistics and used multivariable logistic regression to evaluate the independent association of age with the primary outcome, in-hospital death. The secondary outcome was early withdrawal of life-sustaining therapy. RESULTS: There were 126 adult patients (median age 67 yr [Q1-Q3, 33-80 yr]) with severe TBI during the study period who met our eligibility criteria. The most common mechanism was high-velocity blunt injury (55 patients [43.6%]). The median Marshall score was 4 (Q1-Q3, 2-6), and the median Injury Severity Score was 26 (Q1-Q3, 25-35). After controlling for confounders including clinical frailty, pre-existing comorbidity, injury severity, Marshall score and neurologic examination at admission, we observed that older patients were more likely than younger patients to die in hospital (odds ratio 5.10, 95% confidence interval 1.65-15.78). Older patients were also more likely to experience early withdrawal of life-sustaining therapy and less likely to receive invasive interventions. CONCLUSION: After controlling for confounding factors relevant to older patients, we observed that age was an important and independent predictor of in-hospital death and early withdrawal of life-sustaining therapy. The mechanism by which age influences clinical decision-making independent of global and neurologic injury severity, clinical frailty and comorbidities remains unclear.

Ten Influential Point-of-Care Ultrasound Papers: 2022 in Review.

Keeping up with the latest developments in the point-of-care ultrasound (POCUS) literature is challenging, as with any area of medicine. Our group of POCUS experts has selected 10 influential papers from the past 12 months and provided a short summary of each. We hope to provide emergency physicians, intensivists, and other acute care providers with a succinct update concerning some key areas of ultrasound interest.

Agitated Saline Contrast Injection in Patients with Severe Hypoxemia.

The use of agitated saline contrast (ASC) during echocardiographic examinations is a well-established practice, most commonly performed to identify atrial septal abnormalities in the context of stroke. In the intensive care unit, this technique may be employed to identify anatomic right-to-left shunts (either intracardiac or transpulmonary) that may be contributing to hypoxemic respiratory failure. This narrative review will describe the technique of ASC injection, summarize clinical scenarios where it may be useful, and review the strengths and limitations of the tool.

Ten Influential Point-of-Care Ultrasound Papers: 2021 in Review.

The ongoing rapid expansion of point-of-care ultrasound (POCUS) and its corresponding supporting literature leaves the frontline clinician in a difficult position when trying to keep abreast of the latest developments. Our group of POCUS experts has selected ten influential POCUS-related papers from the past twelve months and provided a short summary of each. Our aim is to give to emergency physicians, intensivists, and other acute care providers key information, helping them to keep up to date on rapidly evolving POCUS literature.

Characterizing the biomechanical differences between novice and expert point-of-care ultrasound practitioners using a low-cost gyroscope and accelerometer integrated sensor: A pilot study.

Introduction: Point-of-care ultrasound (POCUS) has become an important diagnostic tool in acute care medicine; however, little is known about the biomechanical differences between novice and expert practitioners. Methods: A low-cost ($50 CAD) gyroscope and accelerometer integrated sensor was assembled and affixed to an ultrasound probe. Seventeen participants, nine novices and eight experts, were recruited to perform three abdominal and four cardiac scans on a standardized patient. Participant demographics, time per scan, average acceleration, average angular velocity, decay in acceleration and angular velocity over time, and frequency of probe movements were analyzed. Video capture with blinded video review was scored. Results: On video review, experts had higher image optimization and acquisition scores for both abdominal and cardiac scans. Experts had shorter scan times for abdominal (7 s vs. 26 s, p = 0.003) and cardiac (11 s vs. 26 s, p < 0.001) scans. There was no difference in average acceleration (g) between novices and experts performing abdominal (1.02 vs. 1.01, p = 0.50) and cardiac (1.01 vs. 1.01, p = 0.45) scans. Experts had lower angular velocity (°/s) for abdominal scans (10.00 vs. 18.73, p < 0.001) and cardiac scans (15.61 vs. 20.33, p = 0.02) There was a greater decay in acceleration over time for experts performing cardiac scans compared to novices (-0.194 vs. -0.050, p = 0.03) but not for abdominal scans or when measuring angular velocity. The frequency of movements (Hz) was higher for novices compared to experts for abdominal (16.68 vs. 13.79, p < 0.001) and cardiac (17.60 vs. 13.63, p = 0.002) scans. Discussion: This study supports the feasibility of a low-cost gyroscope and accelerometer integrated sensor to quantify the biomechanics of POCUS. It may also support the concept of "window shopping" as a method by which experts obtain abdominal and cardiac views, where sliding is used to find an acoustic window, then smaller rocking and tilting probe movements are used to refine the image.

Setting and Titrating Positive End-Expiratory Pressure.

Although maintaining some amount of positive end-expiratory pressure (PEEP) seems essential, selecting and titrating a specific level for patients with ARDS remains challenging despite extensive research on the subject. Although an "open lung" approach to ventilation is popular and has some degree of biological plausibility, it is not without risk. Furthermore, there is no clear evidence-based guidance regarding initial PEEP settings or how to titrate them early in the course of the illness. Many busy clinicians use a "one-size-fits-all" approach based on local medical culture, but an individualized approach has the potential to offer significant benefit. Here we present a pragmatic approach based on simple measurements available on all ventilators, focused on achieving balance between the potential risks and benefits of PEEP. Acknowledging "best PEEP" as an impossible goal, we aim for a straightforward method to achieve "better PEEP."

Stroke Volume Determination by Echocardiography.

Basic critical care echocardiography emphasizes two-dimensional (2D) findings, such as ventricular function, inferior vena cava size, and pericardial assessment, while generally excluding quantitative findings and Doppler-based techniques. Although this approach offers advantages, including efficiency and expedited training, it complicates attempts to understand the hemodynamic importance of any 2D abnormalities detected. Stroke volume (SV), as the summative event of the cardiac cycle, is the most pragmatic available indicator through which a clinician can rapidly determine, no matter the 2D findings, whether aberrant cardiac physiology is contributing to the state of shock. An estimate of SV allows 2D findings to be placed into better context in terms of both hemodynamic significance and acuity. This article describes the technique of SV determination, reviews common confounding factors and pitfalls, and suggests a systematic approach for using SV measurements to help integrate important 2D findings into the clinical context.

Automation of Lung Ultrasound Interpretation via Deep Learning for the Classification of Normal versus Abnormal Lung Parenchyma: A Multicenter Study.

Lung ultrasound (LUS) is an accurate thoracic imaging technique distinguished by its handheld size, low-cost, and lack of radiation. User dependence and poor access to training have limited the impact and dissemination of LUS outside of acute care hospital environments. Automated interpretation of LUS using deep learning can overcome these barriers by increasing accuracy while allowing point-of-care use by non-experts. In this multicenter study, we seek to automate the clinically vital distinction between A line (normal parenchyma) and B line (abnormal parenchyma) on LUS by training a customized neural network using 272,891 labelled LUS images. After external validation on 23,393 frames, pragmatic clinical application at the clip level was performed on 1162 videos. The trained classifier demonstrated an area under the receiver operating curve (AUC) of 0.96 (±0.02) through 10-fold cross-validation on local frames and an AUC of 0.93 on the external validation dataset. Clip-level inference yielded sensitivities and specificities of 90% and 92% (local) and 83% and 82% (external), respectively, for detecting the B line pattern. This study demonstrates accurate deep-learning-enabled LUS interpretation between normal and abnormal lung parenchyma on ultrasound frames while rendering diagnostically important sensitivity and specificity at the video clip level.

Risks and Benefits of Fluid Administration as Assessed by Ultrasound.

For patients in shock, decisions regarding administering or withholding IV fluids are both difficult and important. Although a strategy of relatively liberal fluid administration has traditionally been popular, recent trial results suggest that moving to a more fluid-restrictive approach may be prudent. The goal of this article was to outline how whole-body point-of-care ultrasound can help clarify both the possible benefits and the potential risks of fluid administration, aiding in the risk/benefit calculations that should always accompany fluid-related decisions.

Ultrasound Assessment of the Inferior Vena Cava for Fluid Responsiveness: Making the Case for Skepticism.

Determining whether a patient in shock is in a state of fluid responsiveness (FR) has long been the Holy Grail for clinicians who care for acutely ill patients. While various tools have been put forth as solutions to this important problem, ultrasound assessment of the inferior vena cava has received particular attention of late. Dozens of studies have examined its ability to determine whether a patient should receive volume expansion, and general enthusiasm has been strengthened by the fact that it is easy to perform and non-invasive, unlike many competing FR tests. A deeper examination of the technique, however, reveals important concerns regarding inaccuracies in measurement and a high prevalence of confounding factors. Furthermore, a detailed review of the evidence (small individual studies, multiple meta-analyses, and a single large trial) reveals that the tool performs poorly in general and is unlikely to be helpful at the bedside in circumstances where genuine clinical uncertainty exists.

Completeness of reporting for systematic reviews of point-of-care ultrasound: a meta-research study.

OBJECTIVES: Systematic reviews are often considered among the highest quality of evidence. Completely reported systematic reviews, however, are required so readers can assess for generalisability of the research to practice and risk of bias. The objective of this study was to assess the completeness of reporting for systematic reviews assessing the diagnostic accuracy of point-of-care ultrasound (POCUS) using the Preferred Reporting Items for Systematic Reviews and Meta-analyses for Diagnostic Test Accuracy (PRISMA-DTA) checklist that was published in 2018. DESIGN AND SETTING: In this meta-research study, MEDLINE, EMBASE and Cochrane Library databases were searched, with no date restriction, on March 1st, 2020 for systematic reviews assessing the diagnostic accuracy of POCUS. Adherence to PRISMA-DTA for the main text and abstract was scored independently and in duplicate using a modified checklist. Prespecified subgroup analyses were performed. MAIN OUTCOME MEASURES: The primary outcome was the mean PRISMA-DTA checklist adherence for the full-text and abstract. RESULTS: A total of 71 studies published from 2008 to 2020 met the inclusion criteria. The overall adherence for the full-text was moderate: 19.8 out of 26.0 items (76%) and for the abstract was 7.0 out of 11.0 items (64%). Although many items in the PRISMA-DTA checklist were frequently reported, several were r infrequently reported (<33% of studies), including item 5 (protocol registration), item D2 (minimally acceptable test accuracy) and item 14 (variability in target condition, index test and reference standards). Subgroup analyses showed a higher PRISMA-DTA mean adherence (SD) for high impact journals (20.9 (2.52) vs 18.9 (1.95); p<0.001), studies including supplemental materials (20.6 (2.48) vs 18.9 (2.28); p=0.004), studies citing adherence to PRISMA reporting guidelines (20.4 (1.95) vs 19.0 (3.00); p=0.038) and studies published in journals endorsing PRISMA guidelines (20.2 (2.47) vs 18.6 (2.37); p=0.025). There was variable adherence based on journal of publication (p=0.006), but not for study population (adult vs paediatric vs mixed) (p=0.62), year of publication (p=0.94), body region (p=0.78) or country (p=0.40). There was no variability in abstract adherence based on whether the abstract was structured with subheadings or not (p=0.25). A Spearman's correlation found moderate correlation between higher word counts and abstractand full-text adherence (R=0.45, p<0.001 and R=0.38, p=0.001), respectively. CONCLUSIONS: Overall, the reporting of POCUS diagnostic accuracy systematic reviews and meta-analyses was moderate. We identified deficits in several key areas including the preregistration of systematic reviews in an online repository, handling of multiple definitions of target conditions, index tests and reference standards and specifying minimally acceptable test accuracy. Prospective registration of reviews and detailed reporting as per PRISMA-DTA during the research process could improve reporting completeness. At an editorial level, word count and supplemental material limitations may impede reporting completeness, whereas endorsement of reporting guidelines on journal websites could improve reporting.