Emerging Applications in Point-of-Care Ultrasound — AI, CEUS, RTMUS/RTMSPUS Study Notes

AI in POCUS

  • Context: Advances in point-of-care ultrasound (POCUS) are driven by technology trends (handheld devices, post-processing for image quality, wireless and cloud-based storage) and are expanding how clinicians use POCUS at the bedside.
  • Three emerging applications highlighted to illustrate how new tech broadens POCUS capabilities (not exhaustive): AI in POCUS, Contrast-Enhanced Ultrasound (CEUS), and Remote Telementored Ultrasound (RTMUS) including self-performed variants.
  • AI in medicine (and POCUS): AI is an umbrella term covering machine learning (ML) and deep learning (DL). It enables computer systems to perform tasks that typically require human cognition. In POCUS, DL is the dominant approach in current AI applications.
  • Key definitions:
    • Artificial Intelligence (AI): umbrella term for systems performing cognitive tasks.
    • Machine Learning (ML): algorithms that learn from data and make predictions.
    • Deep Learning (DL): neural networks with multiple layers enabling robust feature representation and classification.
  • ML modalities in ultrasound:
    • Supervised learning: uses labeled datasets with expert-defined ground truths to train models. Ground truths are the gold standard against which the algorithm is evaluated.
    • Unsupervised learning: discovers structure in unlabeled data (not described in depth in the transcript).
    • Reinforcement learning: learns via trial-and-error with feedback (not detailed in the transcript).
  • Why supervised learning is prevalent in ultrasound ML:
    • Example: pleural line with B-lines: expert panels set the presence/absence of B-lines as the ground truth; the ML model tries to match this ground truth.
    • Challenges: requires expert-annotated, patient-level labeled datasets; quality depends on human input; dataset creation is labor-intensive.
  • Opportunities created by AI in POCUS:
    • Partnerships between POCUS experts and AI companies to build prospective training/validation/testing datasets.
    • AI tools for diagnostic support across various organs and conditions.
  • Demonstrated AI capabilities in POCUS (current trends):
    • Lung ultrasound: high sensitivity and specificity for B-line detection, quantification, and severity scoring. Comparisons show differences in performance between physicians and AI, underscoring the need for continual refinement.
    • Cardiac ultrasound: AI for ejection fraction (EF) classification can match cardiologist/POCUS-expert visual estimation.
    • Diastolic function assessment: AI shows accuracy in diastolic assessment using POCUS.
    • Bladder volume: AI shows promise because traditional bladder scanners can misidentify ascites, cysts, fibroids, and other pelvic structures as the bladder; early work shows high sensitivity and specificity relative to sonographer measurements, suggesting potential cost savings on cheaper handheld devices. Some devices include an "auto-sweep" feature that steers the beam across multiple planes from a single view to help calculate volumes.
  • Current status and deployment:
    • Most AI tools for POCUS are in research & development; several commercial products exist for select abdominal, cardiac, lung, musculoskeletal, and vascular applications.
    • Barriers slowing adoption include: operator dependence and variability in image quality, lack of standardized acquisition protocols, artifacts, reliance on dynamic cine/video rather than high-resolution static frames, subjectivity in interpretation, and scarcity of large labeled datasets.
    • These barriers also motivate continued AI development to address variability and standardization.
  • Core benefits of POCUS AI (categories):
    • Detection: identifying organs/structures or pathologies (e.g., ascites, pulmonary edema).
    • Classification: grading disease states (e.g., EF category).
    • Segmentation: delineating borders of structures (e.g., diaphragm, rib shadows, pleural line).
  • AI-assisted image acquisition and workflow enhancements:
    • Auto-capture: software automatically saves acquisition when quality thresholds are met.
    • Image quality meters and machine-guided probe maneuvers: real-time feedback to optimize image capture, supporting novice users and reducing inter-operator variability.
    • Auto-labeling and auto-documentation: automated annotation of images and documentation of findings to streamline charting and billing.
    • Immediate quality improvement feedback and archiving: facilitates credentialing, privileging, and continuous quality improvement (CQI).
  • Administrative and educational impact:
    • AI can provide on-the-spot feedback to individual users, potentially reducing the need for direct supervision by expert faculty.
    • Auto-documentation and standardized image sets can improve reproducibility and training.
  • Outlook:
    • With advancing processing power and DL approaches, AI is expected to become more integrated into everyday POCUS use, influencing image acquisition, interpretation, and administrative tasks.

Contrast-Enhanced Ultrasound (CEUS)

  • What CEUS is:
    • CEUS uses ultrasound contrast agents (UCAs) to augment conventional ultrasonography.
    • UCAs are microbubbles with second-generation shells (phospholipid) and inert gas (e.g., perfluorocarbon or sulfur hexafluoride). They are intravascular blood-pool agents that can persist in circulation for at least 5 minutes.
    • Microbubbles are similar in size to or smaller than red blood cells.
  • Imaging mechanism:
    • At low acoustic power (mechanical index, MI, typically between 0.2 and 0.4), microbubbles produce a nonlinear harmonic response that can be separated from tissue signals using contrast harmonic imaging to evaluate capillary-level tissue perfusion.
  • Regulatory status and scope:
    • CEUS in trauma is off-label in the United States, with decades of European research support.
    • Focused Assessment with Sonography in Trauma (FAST) can incorporate CEUS to look for direct injury evidence (lacerations, extravasation) beyond indirect free-fluid assessment.
    • UCAs are metabolized and cleared within about 15 minutes and are not nephrotoxic, enabling repeatable studies without pre-test blood work (e.g., creatinine).
  • Advantages of CEUS:
    • Repeatable and safe for multiple imaging passes; no nephrotoxicity; no requirement for pre-imaging renal testing.
    • Radiation-sparing compared to CT, which is particularly advantageous in pediatric populations.
  • Limitations and limitations relative to CT:
    • CEUS does not replace CT for all diagnostic tasks; CT remains superior for evaluating adrenal glands, renal collecting system, bowel, and diaphragm when detailed cross-sectional anatomy is required.
  • Emerging and notable CEUS applications:
    • Trauma evaluation: direct visualization of organ injury (e.g., lacerations, active extravasation) during CEUS-FAST.
    • Acute myocardial infarction (AMI): pre- and post-intervention use for sonothrombolysis, where high-power insonation causes microbubble collapse producing fluid jets that can disrupt nearby thrombus; early trials suggest potential reductions in infarct size and preservation of left ventricular systolic function.
    • Other indications: evaluation of acute abdominal aortic aneurysm rupture and endoleak, actively bleeding soft-tissue hematomas, testicular and adnexal torsion, infectious processes, and ectopic pregnancy.
  • Additional technical notes and figures:
    • CEUS can monitor perfusion at capillary level and provide dynamic information about tissue vascularity and lesion characterization.
    • Figure references in the source illustrate liver laceration visible with CEUS compared with grayscale imaging (for example, Fig. 29.6).
  • Practical considerations:
    • When using CEUS, be mindful of timing relative to contrast injection and the attenuation of signal with depth; operator experience remains important.

Remote Telementored Ultrasound (RTMUS) and Self-Performed Ultrasound (RTMSPUS)

  • RTMUS concept:
    • Remote telementored ultrasound is telemedicine where a remote expert guides a distant user to obtain clinically meaningful images that can be interpreted asynchronously or in real time.
    • Other acronyms describe related ideas: remotely-supported ultrasound, tele-ultrasound, telesonography, telementoring.
    • The paradigm aims to improve access to imaging in remote or underserved settings, including space medicine on the International Space Station (ISS).
    • Effectiveness in robust clinical trials remains to be fully established; ongoing research is needed to prove outcomes across diverse settings.
  • RTMSPUS (remote tele-mentored self-performed ultrasound):
    • A subset where users image themselves under remote guidance, or autonomously with just-in-time education or AI assistance.
    • Historical note: described in the space program context where astronauts in space were guided to perform FAST, echocardiography, and central vasculature assessments.
    • Implications: expands opportunities for physical examination during remote encounters and strengthens the physician-patient relationship beyond the hospital setting.
  • Current state and evidence:
    • Self-performed ultrasound can be either guided by a remote expert or autonomous with minimal training.
    • Case reports show ultrasound-naïve users achieving expert-quality clips after short training (e.g., around 15 minutes) for lung ultrasound; another report demonstrates fetal wellness imaging with acceptable image quality for amniotic fluid assessment, though detecting fetal heartbeat and facial profile can be more challenging.
    • These case reports suggest feasibility but highlight limitations in generalizability due to sample size, population differences, and study design.
  • Examples of self-performed ultrasound indications (from Table 29.1):
    • Remotely tele-mentored self-performed ultrasound indications:
    • Space medicine: assessing central venous pressure, cardiac function, and muscle mass.
    • COVID-19 lung examinations: diagnosis and monitoring during the pandemic.
    • Maternal wellness examination: fetal assessment during pregnancy.
    • Appendicitis: self-diagnosis at home.
    • Self-performed autonomous ultrasound indications:
    • Heart failure and lung examinations.
    • Fetal wellness.
  • Practical considerations for RTMUS/RTMSPUS:
    • Training requirements are modest (e.g., short tutorials) but access to reliable guidance and communication channels is critical.
    • Just-in-time education and potential AI augmentation may enhance performance and reduce reliance on expert presence.
    • The field envisions self-imaging and remote mentoring becoming more common as technology improves, potentially enabling home or space-based diagnostics.
  • General limitations and needs:
    • More robust evidence from larger trials is needed to establish efficacy, safety, and clinical outcomes.
    • Considerations include patient safety, accuracy, data privacy, device accessibility, and digital equity.
  • Disclosures/Conflicts of Interest:
    • Andrew W. Kirkpatrick: Consulting for Zoll Medical and Innovative Trauma Care; Grant funding by 3M/Acelity.
    • Rachel B. Liu: Consulting for Caption Health, Inc. (GE).

Key data points and illustrative details

  • EF calculation example (from Fig. 29.2):
    • End-diastolic volume: EDVA4C=79.71extmlEDV_{A4C} = 79.71 ext{ ml}
    • End-systolic volume: ESVA4C=26.39extmlESV_{A4C} = 26.39 ext{ ml}
    • Stroke volume: SVA4C=53.32extmlSV_{A4C} = 53.32 ext{ ml}
    • Ejection fraction: EFA4C=66.89extextendashext?EF_{A4C} = 66.89 ext{ extendash} ext{?}
    • Note: The figure lists these values with EF = 66.89%; EF can be calculated as
      EF = rac{EDV - ESV}{EDV} imes 100rac;
      which for the given numbers yields the EF shown.
  • Example LED/UX cues from AI tools:
    • Auto-labeling of landmarks and regions (e.g., rib shadows, pleural line) to support faster, more consistent labeling.
    • Image acquisition guidance: auto-capture, image quality meters, and machine-guided probe maneuvers for optimal views.
    • Auto-calculated measurements (e.g., EF) displayed to support rapid interpretation.
  • Technical parameter reference:
    • CEUS uses a low mechanical index (MI) around the range extMIextin[0.2,0.4]ext{MI} ext{ in } [0.2, 0.4] to elicit nonlinear microbubble responses for perfusion assessment.
  • Practical imaging notes:
    • CEUS advantages include repeatability and safety (no nephrotoxicity) and lack of ionizing radiation, which makes it attractive for serial exams and pediatric patients.
    • CEUS has limitations relative to CT for complex evaluation of deep structures or detailed anatomic delineation.
  • Broader implications:
    • AI-enabled POCUS, CEUS, and RTMUS collectively point toward more autonomous, accessible, and standardized ultrasound care across settings, including remote, space, and home environments.
    • Ethical and practice considerations include data privacy, algorithm transparency, training requirements, and ensuring equitable access to advanced imaging technology.

Connections to foundational principles and real-world relevance

  • Foundational principles:
    • Medical imaging relies on pattern recognition; ML/DL can learn to recognize patterns in ultrasound data with adequate labeled training sets.
    • Standardization of image acquisition and interpretation improves diagnostic performance and reduces inter-operator variability.
    • Telemedicine and remote guidance can expand access to expertise, particularly in remote or space environments.
  • Real-world relevance:
    • AI tools can assist with real-time decision support, potentially reducing time to diagnosis and standardizing image quality across operators with varying experience.
    • CEUS provides a radiation-free, repeatable imaging option for viable tissue perfusion assessment and certain trauma scenarios, with potential to complement or reduce dependence on CT in selected cases.
    • RTMUS/RTMSPUS holds promise for expanding imaging access in space, rural areas, disaster zones, and during home-based monitoring, while requiring careful evaluation of safety, efficacy, and training needs.

Formulas and numerical references (LaTeX)

  • Ejection fraction (definition):
    EF = rac{EDV - ESV}{EDV} imes 100\%
  • Example measurements (A4C view):
    EDV<em>A4C=79.71  mlEDV<em>{A4C} = 79.71\ \,mlESV</em>A4C=26.39  mlESV</em>{A4C} = 26.39\ \,ml
    SV<em>A4C=53.32  mlSV<em>{A4C} = 53.32\ \,mlEF</em>A4C=66.89%EF</em>{A4C} = 66.89\%
  • Mechanical index range for CEUS:extMI∈[0.2,0.4]ext{MI} \in [0.2, 0.4]
  • Time-related facts for UCAs:
    • UCAs persist in circulation for at least 5  minutes5\ \,minutes
    • UCAs are cleared from the body within approximately 15  minutes15\ \,minutes and are not nephrotoxic.
  • Training and adoption time referenced for RTMSPUS:
    • Self-imaging after a brief tutorial: approximately 15  minutes15\ \,minutes
  • View and sequence notes from figures (described in text):
    • Fig. 29.2 demonstrates auto-calculation of EF (A4C view).
    • Fig. 29.3 shows segmentation labeling (rib shadow vs pleural line) for training AI.
    • Fig. 29.4 shows auto-labeling during a right upper quadrant view.
    • Fig. 29.5 shows image guidance with auto-assistance and auto EF calculation.
    • Fig. 29.6 compares CEUS liver laceration with grayscale image.

Table 29.1 — Examples of self-performed ultrasound (summary)

  • Remotely tele-mentored self-performed ultrasound indications:
    • Space medicine: Assessing central venous pressure, cardiac function, and muscle mass.
    • COVID-19 lung examinations: Diagnosis and monitoring during the pandemic.
    • Maternal wellness examination: Fetal assessment during pregnancy.
    • Appendicitis: Self-diagnosis of appendicitis at home.
  • Self-performed autonomous ultrasound indications:
    • Heart failure: Lung examinations.
    • Fetal wellness.
  • Existing or theoretical status: summarized as functioning in real-world cases with varying degrees of formal validation; studies indicate that non-clinicians could self-image after minimal training with acceptable accuracy in some contexts.
  • Overall takeaway: These examples illustrate the potential for remote mentoring and autonomous ultrasound to expand access to imaging and empower patients, with ongoing need for robust research and thoughtful integration into care pathways.

Ethical, practical, and regulatory considerations

  • Conflicts of interest disclosed by authors (relevant for interpreting industry collaboration and potential bias).
  • Practical considerations:
    • Training requirements, data privacy, device accessibility, and the need for robust clinical trials to validate effectiveness and safety.
    • Ensuring equitable access to AI/CEUS/RTMUS technologies across patient populations and settings.