AI detects urinary tract stones on CT scans

Convolutional neural networks (CNNS) can detect urinary tract stones on unenhanced CT scans with significant accuracy, according to new findings published in Radiology: Artificial Intelligence.

The authors explored unenhanced abdominopelvic CT exams for more than 500 patients with suspected urolithiasis, with a radiologist reviewing each case and serving as the reference standard. The exams took place on one of two scanners from the same facility.

Two CNNs were then developed—the first identified the urinary tract, the second detected the stones—and the team came up with nine variations to study which deep learning model would be the most effective. While some of the variations were trained with the ImageNet dataset, for instance, others used the GrayNet dataset, which takes the ImageNet dataset and then adds enriched CT images. Also, some model variations focused one of the two scanners (S1 or S2) while others focused on both scanners (SB).

Stones were present in 279 of the 535 patients included in the study. The age and sex of the patient did not seem to make a significant impact on the likelihood of a stone being present. The mean stone size was 6 millimeters (mm).

Overall, the area under the ROC curve (AUC) for the GrayNet-SB variation was 0.954. This was the highest AUC, though “not statistically significantly different” than ImageNet-SB (0.936) or Random-SB (0.925) models. GrayNet-SB also had a higher accuracy (95%) than ImageNet-SB (91%) or Random-SB (88%).  

“In this study, we have developed a cascading CNN model, enriched with modality-specific (CT) radiology images, that detects stone within the urinary tract at unenhanced abdominopelvic CT with a high accuracy,” wrote Anushri Parakh, MD, department of radiology at Massachusetts General Hospital in Boston, and colleagues. “In the realm of emergency radiology, high sensitivity is necessary and turnaround time or lack of resources may pose challenges. On a patient level, the current model achieved a 94% sensitivity and 96% specificity for stone detection where two of three false-negative examinations comprised small-sized stones. On the basis of stone location, the performance was high for all locations, with one scan false-negative for kidney and ureter and two scans false-negative for bladder.”

Parakh et al. also noted that the enriched medical images were an effective way to improve the overall performance of a CNN, as one can see in the higher AUC and accuracy of the GrayNet-SB model compared to the ImageNet-SB model.