Laser lithotripsy depends in part on fragmenting stones to sufficiently small size for spontaneous passage or extracting all large residual stone fragments to provide high stone free success rates, so a second repeat procedure is unnecessary. This preliminary study describes initial development of an optical system and software capable of tracking and labeling kidney stones. Machine learning and image processing techniques were implemented to track, label, and size, on a relative scale, stone fragments. Thresholds were placed on minimum stone fragment size (based on pixel sizes). For system validation, a series of still images from the laboratory setup and previously recorded clinical lithotripsy procedures were analyzed. A laboratory test was also conducted with homogenous background and video to determine number of true positives, false positives, and false negatives. In still images, 8/8 (100%) stones in the laboratory setup and 4/4 (100%) stones in clinical lithotripsy videos were correctly identified. A separate laboratory study correctly identified all five stones in each frame across a total of 110 video frames (550/550) (100%) with a total of 49/550 false positives (9%) and 0/550 (0%) false negatives, at a maximum frame rate of 50 Hz. This preliminary stone tracking study correctly identified stone fragments in laboratory frames for still and motion video and during still images from clinical lithotripsy procedures and experimental settings. Further development of software and optical tracking system is planned.
|