A recent study has developed an automatic urinary stone detection system designed to help alleviate the burden on radiologists and improve patient outcomes. This system uses non-enhanced abdominopelvic CT scans and a cascade urinary stone map location-feature pyramid networks (USm-FPNs) combined with a ureter distance heatmap method. The development of such a system is important because urinary stones not only cause acute pain and discomfort but also increase the risk of chronic kidney disease.
The study involved 811 urology patients and 356 emergency patients. The urology patients' data was split into training and validation sets in a 3:1 ratio, while emergency patients' data served as an independent test set. Researchers assessed the system's performance using the free-response receiver operating characteristic curve and the precision-recall curve.
At the stone level, the findings were as follows:
- The system's mean of false positives per scan (mFP) was 1.88 with a sensitivity of 0.977 in the validation set.
- When the ureter distance heatmap was incorporated, the mFP decreased to 1.18, maintaining the same sensitivity.
At the patient level:
- The system achieved a sensitivity of 0.995 and precision of 0.990 in the validation set.
- In the real clinical setting of emergency patients, where 27.5% had stones, the mFP was 1.31, with a sensitivity of 0.977, reducing diagnostic time by over 20%.
Based on their results, the researchers emphasized the efficacy of their newly developed system, concluding that their findings suggest that "deep learning could be a valuable tool for urinary stone detection and location, which can help improve the efficiency of stone-assisted screening and triage of emergency patients with abdominal or flank pain."
The authors declare no competing interests.
