AI-based System Shows Promise in Tuberculosis Detection (IMAGE)
Caption
Examples of chest radiographs for which the deep learning system (DLS) provided the correct interpretation, corresponding to (A) tuberculosis (TB)–positive subjects and (B) TB-negative subjects. Blue outlines encircle salient regions via Grad-CAM (35) that most influence the positive prediction from the DLS and are shown when the DLS considered the image positive. Yellow outlines were annotated by a radiologist to indicate regions of interest; solid outlines indicate findings consistent with TB while dotted outlines indicate other findings. “Confident” indicates that the DLS predicted values close to 0 or 1, whereas “challenging” indicates that the DLS predicted values close to the operating point (0.45). CXR = chest radiograph, rads = radiologists.
Credit
Radiological Society of North America
Usage Restrictions
May use with credit.
License
Licensed content