qXR-CTR

The qXR-CTR device employs a deep-learning artificial intelligence algorithm to automatically analyze posterior-anterior chest X-rays, providing quantitative measurements of cardiac and thoracic diameters for calculating the cardiothoracic ratio. Its purpose is to assist physicians by offering an accurate and automated assessment tool for adult patients, enhancing radiological image processing without replacing clinical judgment.

qXR-CTR, developed by Qure.ai Technologies, is a deep-learning based software designed for the automated assessment of the cardiothoracic ratio (CTR) on posterior-anterior (PA) view chest X-ray (CXR) scans. Utilizing an artificial intelligence algorithm, it analyzes DICOM images to measure the maximal transverse diameter of the heart (CD) and the maximal inner transverse diameter (TD) of the thoracic cavity, subsequently calculating the CTR value. The system is intended for use by physicians or licensed practitioners in healthcare institutions like clinics and hospitals to aid in the assessment of adult patients (≥ 22 years of age), providing a structured report with a preview of the CXR, annotation lines for CD and TD, and the derived CTR, but is not meant for standalone clinical decision-making. The clinical performance of qXR-CTR was validated using 435 CXR scans from the US, with ground truth established by three experienced ABR thoracic radiologists. The software demonstrated accuracy with Root Mean Squared Error (RMSE) values of 7.55 mm for cardiac diameter, 5.43 mm for thoracic diameter, and 0.03 for the CTR. The Mean Absolute Error (MAE) for cardiac diameter was 5.66 mm (standard deviation 5.0 mm), and for thoracic diameter, it was 4.04 mm (standard deviation 3.63 mm), with an MAE for CTR of 0.02 (standard deviation 0.02). These results indicate the device's capability to provide reliable automated measurements to assist clinicians.

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. How does the qXR-CTR algorithm's performance vary across different patient demographics (e.g., age, sex, BMI) and varying image qualities?; What specific deep learning architecture is employed in qXR-CTR, and how was it trained and validated to ensure robustness against common CXR artifacts or unusual anatomies?; Beyond RMSE and MAE, what other performance metrics (e.g., sensitivity/specificity for detecting cardiomegaly, inter-observer variability compared to manual measurements) were evaluated or could be relevant for clinical utility?; How does the device handle edge cases, such as images with significant pleural effusion, lung opacities, or anatomical variants that might obscure cardiac or thoracic borders?; What is the processing time per image, and how does the software integrate into existing hospital Picture Archiving and Communication Systems (PACS) and radiology workflows?