A Multicenter Study to Optimize Microembolic Signal Classification Based on Double--Blind Multiparametric Assessment by Human Experts Using an Universal Graphical Interface [MESOMEGA]

Overview

Microembolic signals (MES) is a powerful predictor of future embolic events. This study aims to develop and validate a accurate model of classification of MES obtained by transcranial Doppler. monitoring of However, MES detection is technically demanding and requires expert interpretation. By providing a reproducible framework for MES interpretation, this work aims to facilitate MES integration into future clinical trials and decision-making.

Rationale The presence of microembolic signals (MES) is a powerful predictor of future embolic events. However, MES detection is technically demanding and requires expert interpretation. Aim We aim to develop and validate a supervised prediction model for MES classification using features extracted from transcranial Doppler (TCD) signals. The model is intended to support expert consensus and enhance classification concordance by utilizing standardized, pre-specified signal features. Sample size estimates Sample size was estimated using the pmsampsize R package. Based on five predictors, a 1:1 proportion of MES in final dataset, a maximum Nagelkerke R² of 0.75, a shrinkage factor of 90% (to minimize overfitting), and a mean absolute error in predicted probabilities ≤ 0.05, the required sample size is 850 clips. The calculations included an 80:20 training/testing split and a 10% dropout rate. Methods and Design The "Multicenter Study to Optimize Microembolic Signal Classification Based on Double-blind Multiparametric Assessment by Human Experts Using a Universal Graphical Interface" (MESOMEGA trial) is a prospective, randomized, double-blind, diagnostic validation study. All members of World Organization of Neurosonology, their national affiliated societies, and worldwide TCD users in the medical community will be invited to submit TCD monitoring 20-second clips of presumed solid MES or non-MES high-intensity transient signals recorded using a 2 MHz transducer from the proximal middle cerebral artery. Exclusion criteria include inseparable multiple MES (e.g., curtain) or any gaseous embolic form. Each clip will be independently assessed by two randomly allocated experts. Expert reading will be using TCDPlayer and will be blinded to clinical data, source information, and other assessments. They will manually annotate six predefined signal features: characteristic audible signal increase, characteristic wave-like of raw Doppler signals, Emboli-to-Background Ratio, Emboli-to-Mirror Ratio, signal duration, and average velocity of maximum intensity. Analysis will be completed within 90 days. A supervised decision tree model will be developed on the training dataset and validation set. Performance will be assessed using stratified k-fold cross-validation, reporting accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC). Following model development, a Delphi consensus process will be used to evaluate and validate model outputs, aiming for expert agreement on model acceptability and readiness for clinical application. The study will be conducted under appropriate ethical approval and in accordance with international report standards. The study will be conducted under ethical guidelines and approval. Study Outcomes The primary outcome is the classification of clips as MES or non-MES, using expert consensus as ground truth. The model will aim for ≥ 90% classification accuracy. Secondary outcomes include model performance without auditory parameter, interrater concordance and variability, and Delphi consensus strength. Discussion This study will assess the performance of a supervised decision tree model for MES classification and benchmark it against prior MES detection approaches. By providing a reproducible framework for MES interpretation, this work aims to facilitate MES integration into future clinical trials and decision-making.