Etiological DiagnOsis of caRdiac Diseases Based on echoCardiograpHIc Images and Clinical Data.

Hospices Civils de Lyon1,000 enrolled

Overview

Research hypothesis - Recent studies have shown that high-dimensional descriptors of the cardiac function can be efficiently exploited to characterize targeted pathologies. In this project, the investigators hypothesize that echocardiograms possess a wealth of information that is currently under-exploited and that, combined with relevant patient data, will allow the development of robust and accurate digital tools for etiological diagnosis. Objectives - Based on key advances recently obtained in image analysis, notably by members of the consortium, the objective of this project is to develop rigorous and explainable cardiac disease prediction models from echocardiography based on the transformer paradigm (AI). The strength of this study lies in the development of a strong AI framework to model the complex interactions between high-quality image-based measurements extracted from echocardiograms and relevant patient data to automatically predict etiological diagnosis of cardiac diseases

Study Type

OBSERVATIONAL

Enrollment

1,000

Conditions

Hypertension Arterial Hypokinetic Cardiomyopathies Left Ventricular Hypertrophy

Interventions

The origin of the pathology will have been previously diagnosed for each patient thanks to complementary examinations performed as part of routine care (e.g. cardiac CT, cardiac MRI, coronary angiography, thorough biology, nuclear medicine). This information will be used (i) to guide the learning of the AI method developed during the project from a sub-population (80% of the collected database will be used to train the algorithms); (ii) to serve as an evaluation criterion from a test sub-population (remaining 20% of the collected database)

Outcomes

Primary Outcomes

the comparison of the performance of the etiological diagnosis obtained by the artificial intelligence with the etiological diagnosis already established and validated by a physician from the complementary examinations performed on the targeted patients.

The origin of the pathology being previously diagnosed for each patient thanks to complementary examinations carried out in routine (e.g.: cardiac scanner, cardiac MRI, coronary angiography, thorough biology, nuclear medicine). This information will be used (i) to guide the learning of the AI method developed during the project from a sub-population (80% of the collected database will be used to train the algorithms); (ii) to serve as an evaluation criterion from a test sub-population (remaining 20% of the collected database). In addition, visualization tools will be developed to allow clinicians to analyze and interpret the results, particularly with respect to the decision mechanism performed by the algorithm to predict the origin of the pathology. In particular, attention maps will be displayed that will simply allow clinicians to see which data or part of the data was assembled in order to make the decision.

Time frame: Baseline