qXR-BT

qXR-BT is an automated image analysis software intended to assist physicians by identifying the position of breathing tubes and the carina on adult chest X-rays, serving as an adjunctive tool for placement confirmation. The device employs pre-trained convolutional neural networks to process DICOM-formatted chest radiographs, generating secondary images with visual overlays of these critical structures and their relative positions.

qXR-BT is an automated radiological image processing software designed to assist licensed physicians in confirming the correct placement of breathing tubes in adult patients. It operates as an adjunctive tool by generating a secondary digital chest X-ray image that highlights the tip of the breathing tube and the carina, a critical anatomical landmark for assessing tube position. The software processes standard frontal (AP/PA) chest X-rays provided in DICOM format and is engineered to integrate with existing hospital Picture Archiving and Communication Systems (PACS) and other radiology software systems or databases, enhancing current medical imaging workflows without replacing the original radiographic image. The core of qXR-BT's analytical capability lies in its use of pre-trained convolutional neural networks (CNNs), which are responsible for identifying the relevant structures. This AI-driven analysis is supported by pre-processing modules that prepare input DICOM images and post-processing modules that convert the CNN output into user-interpretable formats, including PDF reports with preview images and textual descriptions, and additional DICOM series with labeled overlays. Standalone performance testing conducted on 162 chest X-ray images, with ground truth established by three radiologists, demonstrated high accuracy: the mean absolute distance for carina localization was 2.15 mm, for breathing tube tip localization was 1.97 mm, and the mean absolute error for the distance between these two points was 1.98 mm, all meeting predefined success criteria.

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 performance of qXR-BT vary with different qualities of chest X-ray images (e.g., under/over-penetration, patient positioning, presence of artifacts)?; What are the false positive and false negative rates for the detection of the carina and breathing tube tip, particularly in challenging cases such as patients with anatomical variations or overlying pathologies?; How was the ground truth for carina and tube tip locations established using the three radiologists in the performance study, and what was the inter-rater variability among them?; Are there specific adult patient subgroups (e.g., based on BMI, co-morbidities, or specific thoracic conditions) where qXR-BT's accuracy might be reduced, and were these explored in the 162-image dataset?; What are the computational requirements (e.g., processing time per image, specific hardware or cloud infrastructure needs) for the qXR-BT software, and how might these impact integration into diverse clinical workflow speeds?