TOMTEC-ARENA

TOMTEC-ARENA (TTA2.50) is a medical image management and processing software that incorporates an artificial intelligence algorithm, specifically a Convolutional Neural Network trained via supervised learning, to provide semi-automated and editable echocardiographic measurements from B-MODE and DOPPLER data. The system is intended to assist physicians by enhancing the efficiency and consistency of quantifying cardiovascular, fetal, and abdominal structures and functions for diagnostic support.

TOMTEC-ARENA (TTA2.50) is a clinical software package designed for reviewing, quantifying, and reporting digital medical data to support physicians in diagnosing conditions related to cardiovascular, fetal, and abdominal structures and functions. A key update in this version is the introduction of semi-automated cardiac measurements powered by an artificial intelligence and machine learning (AI/ML) algorithm. This algorithm, a Convolutional Neural Network (CNN) developed using a supervised learning approach, processes B-MODE and DOPPLER echocardiographic datasets to suggest measurements. The AI/ML algorithm was trained by observing and learning from a large dataset of approximately 2,800 echocardiographic studies manually annotated by qualified healthcare professionals, and its performance was evaluated against a separate test pool of about 500 studies. When in use, the clinician selects an image frame, and the AI detector predicts the start and end caliper positions for a measurement. These AI-generated measurements are presented as editable graphical overlays, allowing the user to accept, modify, or reject the suggestion, or opt for manual measurement. This feature, integrated within the IMAGE-COM module, aims to streamline the quantification process without introducing new types of measurements.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. What specific Convolutional Neural Network (CNN) architecture is employed, and what was the rationale for its selection over other potential AI/ML models for echocardiographic measurement?; Can you provide more detail on the characteristics of the ~2,800 training studies and ~500 test studies, including demographic diversity, range of pathologies, and image quality, to understand the algorithm's generalizability?; What were the quantitative performance metrics (e.g., agreement with expert manual measurements, mean error rates for caliper placement) of the semi-automated measurements on the independent test dataset for both B-MODE and DOPPLER data?; How does the system handle images of suboptimal quality or rare anatomical variations not extensively represented in the training data, and are there confidence scores or alerts for potentially less reliable AI-generated measurements?; What is the typical workflow impact, such as time saved per study or reduction in inter-observer variability, when clinicians utilize the semi-automated measurement feature compared to fully manual quantification?