Background: Cardiac amyloidosis is characterized by deposition of misfolded protein in the myocardium causing mainly heart failure symptoms with preserved left ventricular ejection fraction. There are also specific clinical (bilateral carpal tunnel syndrome, polyneuropathy, skin bruising, ruptured biceps tendon…), biomarkers (disproportionally elevated NT-proBNP to the degree of heart failure, persistent elevated troponin, proteinuria..), electrocardiographic (reduced voltage of QRS, atrial fibrillation..) and echocardiographic features (concentric left ventricular hypertrophy, dilated atria, reduced global longitudinal strain with typical pattern of apical sparing, diastolic dysfunction…). Early diagnosis of the disease is crucial to identify patients that may benefit from appropriate treatment. Suspected cardiac amyloidosis on echocardiography or on cardiac magnetic resonance needs to prompt the request of serum free-light chain quantification and serum and urine immunofixation as well as single photon emission computed tomography using bone radiotracers.
Hypothesis: The use of artificial intelligence assisted algorithm applied to echocardiographic data may allow identification of suspected cardiac amyloidosis more precisely as compared to cardiologists with expertise in cardiac imaging.
Methods: This project proposal will comprised 3 different phases:
Phase 1: retrospective evaluation of clinically acquired echocardiographic data with reports indicating left ventricular hypertrophy (LV wall thickness ≥12 mm) and/or cardiac amyloidosis. This evaluation will consist of retrieval and analysis of echocardiographic data (around 20K studies) from 2022 to date. The data will be reanalysed by an experienced observer and a currently available artificial intelligence (AI)-based algorithm to detect suspected cardiac amyloidosis. The agreement between the algorithm and the observer will be tested. In those patients in whom the initial observer who reported the echocardiogram considered that there was suspicion of cardiac amyloidosis, the reports of additional test clinically required to confirm or rule out the diagnosis will be retrieved. Accordingly, the accuracy of the observer and the AI-based algorithm will be compared.
Phase 2. Based on the results of Phase 1, which will informed about the prevalence of the condition, the duration of the prospective assessment of AI-based algorithm assisted echocardiographic image analysis will be estimated to see if this would augment the capacity of cardiologists to pick up early patients suspected to have cardiac amyloidosis.
Phase 3. Pragmatic prospective AI-analysis assessment. Prospective analysis applying the AI-based algorithm to all patients presenting to the echo laboratory and following the clinically indicated diagnostic pathway to confirm or rule out cardiac amyloidosis according to contemporary guidelines. Patients will request to agree with the protocol and will sign informed consent.
Result measures:
The prevalence of cardiac amyloidosis will be analysed in the present project (Phase 1) and the agreement between the cardiologists experts in cardiac imaging and the artificial intelligence based algorithm will be reported. If it is demonstrated that the artificial intelligence-based algorithm provides a prevalence of true cardiac amyloidosis higher than the cardiologists, the impact in clinical practice is massive as a higher number of patients could be diagnosed and benefit from effective specific therapies, improving their clinical outcomes.
Quantify diagnostic suspicion ratio of amyloidosis of AI vs physician (positive & negative suspicion rates)
Quantify the Echo Red Flags: AI vs physician recognized amyloidosis red flags (Red Flags will be guideline directed: wall thickness concentric and also affecting the right ventricle, "sparkling" aspect of the interventricular septum, lipomatous hypertrophy of the interatrial septum, atrial dilatation, pericardial effusion, reduced MAPSE and TAPSE, apical cherry pattern on global longitudinal strain analysis of the left ventricle, thickening of the leaflets of the heart valves, including aortic stenosis)
Quantify the AI sensitivity & specificity. Determine the most specific and most sensitive Red Flags for the AI vs physician (Automated Us2.v2 parameters only)
Quantify the sensitivity & specificity of Us2.V2 vs Us2.ca algorithm
Diagnosis Time: Average time it takes for a specialist to diagnose by echo vs. the average time it takes for the AI algorithm to diagnose.
Analysis of ATTR cardiac amyloidosis patient characteristics (e.g. HFpEF, HFmrEF, HFrEF, Atrial Fibrillation, Aortic Stenosis, other co-morbidities)
Laboratory: The non-invasive cardiac imaging unit of the Heart Institute of the University Hospital Germans Trias i Pujol performs 17,000 echocardiograms yearly, being more than 11,000 transthoracic echocardiograms. The collaboration with the department of nuclear medicine and radiology is excellent having access to cardiac magnetic resonance (700 studies performed yearly, and the expectations for next year will be around 1,000) and single photon emission computed tomography.
Therefore, the feasibility of the project is high. The artificial intelligence-based algorithm for the analysis of the echocardiographic images is also in place and they can be analysed.
