Urinary stone segmentation: A computationally efficient yet consistently effective approach using image processing

The study evaluated the effectiveness of a new method for urinary stone segmentation using image processing. The research included 531 stone samples from 287 patients analyzed using non-contrast-enhanced computed tomography. The proposed method consisted of three steps: stone edge detection, pixel selection based on Hounsfield unit values, and noise removal. The method achieved high agreement with reference data with a median Dice coefficient of 0.86 in the test set. The reproducibility of the method was exceptionally high with a median Dice coefficient of 1.0. On the comparison test set with 125 samples, the method outperformed fixed thresholding and deep learning-based methods. The processing time for one sample was less than one second. The proposed semi-automatic approach proved to be a reliable, simple, and robust method for urinary stone segmentation.