Clinical Validation of an Artificial Intelligence Tool to Predict Inversion Time

Clinical Validation of an Artificial Intelligence Tool to Predict Inversion Time

Introduction: Inversion-recovery (IR) magnetic resonance (MR) sequences are commonly used to perform late-gadolinium enhancement (LGE) imaging during cardiac magnetic resonance (CMR) scans. Inversion Time (TI), i.e. the time between the 180° inverting pulse and the 90°-pulse, must be manually input to obtain optimal myocardium nulling. Determinants of this value are patient's, sequence, and contrast characteristics, and the time after contrast injection. The identification of the correct TI is pivotal to quality images. The determination of TI is mostly based on experience, and it can be challenging in some diseases and for less experienced operators.

Aim of this study is to test in a clinical setting an Artificial Intelligence (AI) tool, which we developed to automatically predict TI in CMR post-contrast IR LGE sequences, named "THAITI".

THAITI performance will be evaluated in terms of 1) quality of images obtained using the AI-predicted TI with a 4-point Likert scale; 2) quality of images obtained using the AI-predicted TI in terms of Contrast-Enhancement ratio, i.e. the signal intensity of enhanced/remote myocardium in CMR-LGE images; 3) numbers of images that need to be reacquired; 4) average time duration of CMR-LGE imaging.

Inclusion Criteria:

• Patients in whom CMR-LGE is performed for a clinical reason
• Mixed cardiac conditions (including cardiomyopathies, ischemic heart disease, normal scans, focal and diffuse myocardial pathological processes)
• Both sexes
• Any age
• Availability of serum creatinine, measured within one month prior to CMR
• Provision of the written informed consent

Exclusion Criteria:

• Non-contrast CMR
• First-pass perfusion stress-CMR
• Absolute contraindication to CMR
• Inadequate overall image quality