Rationale: 

Left ventricular diastolic dysfunction (LVDD) is widespread among patients undergoing surgery, and it is usually more common in the geriatric surgical population ^[1]. In patients undergoing noncardiac surgery, LV diastolic dysfunction is associated with a four-fold increase in pulmonary oedema/ congestive heart failure, a two-fold increase in myocardial infarction, and the composite outcome of major adverse cardiovascular events ^[2]. Furthermore, in patients undergoing cardiac surgery, it is associated with higher mortality and morbidities like major adverse cardiac events and prolonged mechanical ventilation, independent of systolic dysfunction ^[3,4]. A formal diagnosis of LVDD, according to the current American Society of Echocardiography and European Association of Cardiovascular Imaging (ASE/EACVI) update ^[5], involves acquisition and analysis of multiple parameters like mitral inflow velocity (E/A ratio), tissue Doppler imaging of septal and lateral mitral annulus (E/e’ ratio), left atrial (LA) volume and maximum tricuspid regurgitation (TR Vmax). This requires advanced training and involves time‐consuming algorithms. 

Perioperative trans thoracic echocardiography (TTE) performed by anesthesiologists is becoming increasingly popular and effective. A recent study found that perioperative-focused TTE performed by anesthesiologists resulted in step-down (eliminating extensive monitoring such as pulmonary artery catheter) or step-up (inotropic/vasopressor therapy, volume infusion, additional monitoring) in a significant number of patients ^[6]. However, measuring many parameters remains a considerable issue in accurately following the ASE/ EACVI 2016 LVDD standards. According to a recent survey ^[7], there are substantial discrepancies among anesthesiologists in how they evaluate, grade, and monitor LVDD in a busy perioperative context. If anaesthetists performing point‐of‐care ultrasound (POCUS) can screen for LVDD in high‐risk surgical patients, they can change the management. During the early stages (LVDD grade I), LV filling can be improved by increasing preload and preventing tachycardia. Patients in the advanced stage (LVDD grades II and III) have higher baseline left atrial (LA) pressure and are preload intolerant. These patients need careful preoperative fluid control and titrated doses of diuretics ^[8].

There is ongoing growth of handheld US devices in clinical practice. Most handheld devices offer limited modalities like 2-D, M mode, colour flow Doppler (CD) and pulse wave Doppler (PWD). However, only one such device, Kosmos by Echonous Inc., offers continuous wave Doppler (CWD). The device is equipped with AI trio software, which labels the echo images, grades the quality of images and guides the clinician to obtain good-quality images by prompting probe movements. The device also measures the ejection fraction automatically (AutoEF). It also calculates stroke volume, cardiac output, and heart rate.

In an earlier study, this unique ability to use CWD was validated to measure aortic valve (AV) gradients, and it could reliably detect clinically significant aortic stenosis (AS) ^[9]. This device has also been studied in various other settings. In an initial validation study of 100 patients, the AutoEF results acquired using the HUD were compared with manually traced biplane Simpson’s rule measurements on a cart-based system.  There were 38 patients with reduced LVEF (<50%).  The analysis revealed good agreement between AutoEF and reference manual EF (Interclass correlation coefficient (ICC)= 0.85; r = 0.87, P < 0.001; minimal mean bias with acceptable limits of agreements. The EF < 50% detection by the AutoEF algorithm was feasible with excellent sensitivity and specificity with a diagnostic accuracy of 88% ^[10]. This was subsequently proven in another multicentre study. Four hundred twenty-four participants requiring a TTE were recruited to have a focused cardiac US done by a novice or experienced user. The Kosmos device calculated AutoEF at the bedside, which was subsequently compared to cart-based TTE. The authors concluded that AI-assisted LVEF assessments provided highly reproducible LVEF estimations compared to formal TTE ^[11]. Similar findings were also noted in another study from two Japanese hospitals. The authors used high-end ultrasound machines to compare AutoEF by AI with Kosmos and Simpson’s biplane disk method. The ICC between the AutoEF and the standard techniques was excellent (0.81, p < 0.001) without clinically meaningful systematic bias (mean bias -1.5%, p = 0.008, limits of agreement ± 15.0%). Similar to previous studies, reduced EF <50%  was detected with a good sensitivity of 85%  and specificity of 81% ^[12]. In a study of 115 oncology patients, AutoEF calculation by oncology staff was feasible using AI-enabled  Kosmos. Detection of LVEF < 50% was possible with excellent accuracy. The authors concluded that this has the potential to expedite the clinical workflow of cancer patients ^[13]. In the context of a comprehensive LVDD assessment, CWD is essential to measure TR Vmax. However, its utility in this important parameter remains to be tested. 

Artificial intelligence (AI) software is utilised for various echocardiographic measurements in parallel developments. The us2.ai is an AI-driven vendor-independent cardiac ultrasound image processing platform approved by the United States Food and Drug Administration (FDA) ^[14]. Briefly, Us2.ai is a deep learning-based workflow that automates the entire process by rapidly analysing the digital imaging and communications in medicine (DICOM) files of a patient’s echocardiographic exam without the need for human intervention, from the classification of views, identifying cardiac phases to giving readouts of about 45 cardiac measurements. The software generates a comprehensive echocardiographic report, which, in addition to LV and right ventricular (RV) systolic function, gives a detailed report on LVDD grade as per ASE guidelines. This algorithm has been trained on a large dataset and externally validated. The automated measurements showed good agreement with measured values by expert sonographers, with a small mean error range of 9–25 mL for left ventricular volumes and LVEF and 1·8–2·2 for the E/e’ ratio. The us2.ai software reliably classified systolic dysfunction (LVEF <40%, AUC range 0·90–0·92) and diastolic dysfunction (E/e’ ratio ≥13, AUC range 0·91–0·91) ^[15]. This algorithm could diagnose and classify heart failure from DICOM images from a large electronic health record system in Scotland ^[16]. In another recent study, two-dimensional and Doppler echocardiographic images from patients with normal AV and all degrees of AS were analysed by Us2.ai with no human input. The us2.ai measurement of AV peak, mean velocity, and AV area by continuity equation closely matched human measurements across all AS severities. The authors concluded that this algorithm can improve the interpretation and diagnosis of AS by minimising interscan variability ^[17].In an external validation study, the automated GLS measurement by us2.ai showed good agreement with manual measurements. The automated GLS accurately identified patients with HF and had excellent AUC ^[18].

These two technological developments, handheld echo by Kosmos, Echonous Inc. and us2.ai, have been studied together recently. Novices with no prior echo experience underwent 2‐weeks of training to acquire echo images with trio AI guidance using the Echonous Kosmos with  Us2.ai software. All patients also had standard echo by cardiologists as the reference standard to detect reduced LV ejection fraction (LVEF) < 50%  in symptomatic patients. AI‐enhanced novice pathway yielded interpretable results in almost all patients and took a shorter time for the study. The AUC of the AI novice pathway was 0.880, with excellent sensitivity and specificity. The median absolute deviation of the AI‐novice pathway LVEF from the reference standard LVEF was acceptable ^[19]. A recent study demonstrated the feasibility of five novice nurses who visited 94 heart failure patients at home to detect cardiac dysfunction using POCUS by Kosmos, Echonous Inc. with us2.ai.  The sensitivity for the primary outcome LVEF <50% or LAVI >34 mL/m2 was 92% for AI-POCUS compared with 87% for NT-proBNP >125 pg/mL, with AI-POCUS having a significantly higher area under the curve (P = 0.040) ^[20]. 

Recently, there has been a lot of interest in the phasic function of LA measured by LAS. The reason can be LAS changes preceded alterations in LA volume, resulting in earlier identification of diastolic dysfunction, with prognostic implications in heart failure, atrial fibrillation and stroke ^[21,22].  LAS has been demonstrated to categorise LVDD ^[23] and is useful as a single parameter to grade LVDD and detect high LV filling pressures with accuracy ^[24-26].

Suppose these two Kosmos handheld echocardiography powered with us2.ai software can be combined in preoperative patients’ care. In that case, it can provide a quick, reliable, point-of-care assessment of LVDD for early diagnosis and appropriate treatment to improve outcomes. 

Objectives:   

To evaluate the feasibility of a handheld echocardiography machine with an artificial intelligence algorithm to assess LVDD using the ASE algorithm and compare results from the handheld echocardiography with the cart-based echocardiography.

Study groups:

The study involves a single group in whom all the proposed parameters would be measured.

Interventions:

None. It is a prospective observational study without any interventions planned. The information, test results, and other data collected as part of the subject’s standard clinical care will be provided to them in the usual manner. In addition, no data from the Kosmos echocardiography machine recordings will be entered into the subject’s electronic medical record or be used to guide their clinical care. No data collected from the Kosmos recordings as part of the study session, nor any results or analysis generated after that, will have any implications on further decision-making during their hospital care.

Observations:

Apart from the de-identified Kosmos device and us2.ai data, the observations are listed in the attached case record form, Appendix A.

Outcome measures:

The primary endpoint of the study will be the diagnostic accuracy of various parameters (namely E/A, E/e’, LA volume indexed, Vmax TR, LAS R) of LVDD and grade of LVDD obtained by us2.ai software using Kosmos handheld echo compared to these measurements on cart-based by echocardiography machine.

References:

1.     Philip B, Pastor D, Bellows W, Leung JM. The prevalence of preoperative diastolic filling abnormalities in geriatric surgical patients. Anesth Analg 2003;97:1214‐21. 

2.     Fayad A, Ansari MT, Yang H, Ruddy T, Wells GA. Perioperative diastolic dysfunction in patients undergoing noncardiac surgery is an independent risk factor for cardiovascular events. Anesthesiology 2016;125:72‐91. 

3.     Metkus TS, Suarez‐Pierre A, Crawford TC, Lawton JS, Goeddel L, Dodd‐O J, et al. Diastolic dysfunction is common and predicts outcomes after cardiac surgery. J Cardiothorac Surg 2018;13:67‐73. 

4.     aw R, Hernandez AV, Pasupuleti V, Deshpande A, Nagarajan V, Bueno H, et al. Effect of diastolic dysfunction on postoperative outcomes after cardiovascular surgery: A systematic review and meta‐analysis. J Thorac Cardiovasc Surg 2016;152:1142‐53.

5.     Nagueh SF, Smiseth OA, Appleton CP, Byrd BF 3rd, Dokainish H, Edvardsen T, et al. Recommendations for the evaluation of left ventricular diastolic function by echocardiography: An update from the American society of echocardiography and the European association of cardiovascular imaging. J Am Soc Echocardiogr 2016; 29:277‐314. 

6.     Subramaniam K, Boisen ML, Yehushua L, et al. Perioperative transtho- racic echocardiography practice by cardiac anesthesiologists—report of a “start-up” experience. J Cardiothorac Vasc Anesth 2021;35:222–32. 

7.     McIlroy DR, Lin E, Hastings S, Durkin C. Intraoperative transesophageal echocardiography for the evaluation and management of diastolic dysfunction in patients undergoing cardiac surgery: A survey of current practice. J Cardiothorac Vasc Anesth 2016;30:389‐97.

8.     Mahmood F, Jainandunsing J, Matyal R. A practical approach to echocardiographic assessment of perioperative diastolic dysfunction. J Cardiothorac Vasc Anesth 2012;26:1115‐23. 

9.     SachpekidisV,PapadopoulouS,KantartziV,etal.Anovelhandheldecho- cardiography device with continuous-wave Doppler capability: implications for the evaluation of aortic stenosis severity. J Am Soc Echocardiogr 2022;35:1273-80. 

10.  Papadopoulou S-L, Sachpekidis V, Kantartzi V, Styliadis I, Nihoyannopoulos P. Clinical validation of an artificial intelligence-assisted algorithm for automated quantification of left ventricular ejection fraction in real time by a novel handheld ultrasound device. Eur Heart J Digital Health 2022; 3:29–37.

11.  Motazedian P., Marbach J.A., Prosperi-Porta G., et al.  Diagnostic accuracy of point-of-care ultrasound with artificial intelligence-assisted assessment of left ventricular ejection fraction. npj Digital Medicine (2023)6:201 ; https://doi.org/10.1038/s41746-023-00945-1 

12.  Kagiyama N., Yukio Abe Y., Kusunose K. et al., Multicenter validation study for automated left ventricular ejection fraction assessment using a handheld ultrasound with artificial intelligence. Natures/Scientific Reports. 2024.14:15359  https://doi.org/10.1038/s41598-024-65557-5. 

13.  Papadopoulou SL., Dimitrios Dionysopoulos  D, Mentesidou V., et al., Artificial intelligence-assisted evaluation of cardiac function by oncology staff in chemotherapy patients. European Heart Journal - Digital Health (2024) 5, 278–287 https://doi.org/10.1093/ehjdh/ztae017. 

14.  Tromp, J. et al. A formal validation of a deep learning-based automated workflow for the interpretation of the echocardiogram. Nat. Commun. 2022 https://doi.org/10.1038/s41467-022-34245-1.

15.  Tromp J., Seekings P.J., Hung C-L., et al. Automated interpretation of systolic and diastolic function on the echocardiogram: a multicohort study. Lancet Digit Health 2022; 4: e46–54.

16.  Oo M.M., Gao C.,Cole C., et al. Artificial intelligence-assisted automated heart failure detection and classification from electronic health records. ESC Heart Failure (2024) DOI: 10.1002/ehf2.14828.

17.  Krishna H., Desai K., Slostad B., et al. Fully Automated Artificial Intelligence Assessment of Aortic Stenosis by Echocardiography. J Am Soc Echocardiogr 2023;36:769-77.  

18.  Myhre P.L., Hung C-L., Matthew J. Frost M.J., et al. External validation of a deep learning algorithm for automated echocardiographic strain measurements. European Heart Journal - Digital Health (2024) 5, 60–68 https://doi.org/10.1093/ehjdh/ztad072. 

19.  Huang W., Koh T., Tromp J., et al. Point‐of‐care AI‐enhanced novice  echocardiography for screening heart failure (PANES‐HF). Nature/ Scientific Reports (2024) 14:13503 | https://doi.org/10.1038/s41598-024-62467-4 . 

20.  Tromp J., Sarra C.,Nidhal B., et al. Nurse-led home-based detection of cardiac dysfunction by ultrasound: results of the CUMIN pilot study. European Heart Journal - Digital Health (2024) 5, 163–169 https://doi.org/10.1093/ehjdh/ztad079. 

21.  Ferkh A,Clark A, Thomas L. Left atrial phasic function: physiology, clinical assessment and prognostic value. Heart 2023;109:1661–9. 

22.  Thomas L, Muraru D, Popescu BA, et al. Evaluation of left atrial size and function: Relevance for clinical practice. J Am Soc Echocardiogr 2020;33:934–52.

23.  Singh A, Addetia K, Maffessanti F, et al. LA strain for categorization of LV diastolic dysfunction. J Am Coll Cardiol Imaging 2017;10:735–43.

24.  Borde D., Joshi S., Jasapara A., et al. Left Atrial Strain as a Single Parameter to Predict Left Ventricular Diastolic Dysfunction and Elevated Left Ventricular Filling Pressure in Patients Undergoing Off-Pump Coronary Artery Bypass Grafting. J  Cardiothorac  Vasc Anesth 2021; 35: 1618-25. 

25.  Cameli M, Sparla S, Losito M, et al. Correlation of left atrial strain and Doppler measurements with invasive measurement of left ventricular end- diastolic pressure in patients stratified for different values of ejection fraction. Echocardiography 2016;33:398–405; 

26.  Thomas L, Marwick TH, Popescu BA, et al. Left atrial structure and function, and left ventricular diastolic dysfunction: JACC state-of-the-art review. J Am Coll Cardiol 2019;73:1961–77.