qXR detects abnormal chest X-ray findings. It can be used to separate normal from abnormal X-rays, for pre-read assistance, or as a radiology audit tool. qXR includes a proprietary algorithm that screens X-rays for signs of tuberculosis.
qXR is a chest X-ray screening tool built with deep learning. It classifies chest X-rays as normal or abnormal, identifies the abnormal findings, and highlights them on the X-ray.
qXR also generates a description of the X-ray findings, including name, size and location of the abnormality, that is used to pre-fill radiology reports.
The algorithms have been trained and tested using a growing database (over 2.5 Million) of X-rays from diverse sources.
The artificial intelligence algorithm underlying qXR is trained to detect not only classical primary pulmonary TB, but also atypical manifestations. It can be used to simultaneously screen for COPD, lung malignancies in high-risk populations, and certain cardiac disorders.
Complimented by an End-to-End software designed for screening programs, qXR for TB can screen results at the Point of Care, so that confirmatory diagnosis and notification of cases can happen on the same day.Learn More
qXR can screen a Chest X-ray and distinguish between Normal and Abnormal scans with a high degree of accuracy.
Integrated with a Radiology worklist, it can facilitate of reporting of all Abnormals first to improve turnaround for cases that need immediate attention.
qXR can detect and localize multiple findings in a Chest X-ray including abnormal classification, different types of lung parenchymal opacities, pneumothorax, pleural effusion, cardiac enlargement, and anatomical variations seen in the chest.
First published online on July 2018,the study has been updated to include validation of 10 different findings on the Qure 100k test set and an additional algorithm accuracy comparison against a 3 radiologist majority on 2000 scans.Read the paper on arXiv
Independent evaluation at Mass General Hospital, Boston, USA on 874 Chest X-rays for 4 different findings. qXR Algorithms were found to outperform Radiologists with different years of experience against Ground Truth set by Thoracic Radiologists and were able to quantify change over timeRead the paper on PLOS ONE
The list of abnormalities that the deep learning algorithms for qXR keeps growing and it is now part of multiple independent validation studies. The accuracy of 10 different findings is reported individually below, using majority opinion of 3 radiologists as ground truth. By altering the detection threshold, each algorithm can be operated at either a high-sensitivity or high-specificity operating point depending on the clinical setting.
|Abnormal finding||AUC (Confidence interval)||Operating point 1 (sensitive)||Operating point 2 (specific)|
|Normal(No Abnormality Detected)||0.856 (0.853 - 0.859)||0.90||0.56||0.66||0.90|
|Blunted CP Angle||0.947 (0.942 - 0.952)||0.90||0.88||0.88||0.90|
|Cardiomegaly||0.95 (0.947 - 0.954)||0.90||0.89||0.85||0.90|
|Cavity||0.964 (0.948 - 0.980)||0.93||0.90||0.90||0.93|
|Consolidation||0.941 (0.934 - 0.948)||0.90||0.88||0.87||0.90|
|Fibrosis||0.937 (0.928 - 0.946)||0.90||0.83||0.85||0.90|
|Hilar Enlargement||0.844 (0.828 - 0.860)||0.90||0.57||0.58||0.90|
|Nodule||0.920 (0.909 - 0.931)||0.90||0.77||0.80||0.90|
|Opacity||0.936 (0.933 - 0.939)||0.90||0.82||0.84||0.90|
|Pleural Effusion||0.957 (0.953 - 0.960)||0.90||0.88||0.88||0.90|
|Pneumothorax||0.967 (0.950 - 0.983)||0.95||0.85||0.90||0.90|
Write to us at email@example.com to request a demo.