Automated Machine Learning Segmentation and Measurement of Urinary Stones on CT Scan

Efficient and Accurate Computed Tomography–Based Stone Volume Determination: Development of an Automated Artificial Intelligence Algorithm

“…9,[18][19][20] As such, there is a growing appeal to the application of automated CT-based stone volume determination algorithms. 18,[21][22][23][24][25][26][27] To this end, AI offers clinical utility, as it precludes the need for manual measurements while increasing the accuracy of the determination by eliminating the high interobserver variability seen with linear measurements and the inaccuracies of employing the ellipsoid formula. 9 Prior studies investigated the utility of AI in urolithiasis diagnostics, but they primarily focused on detecting the presence of stones rather than assessing stone volume.…”

“…Only a few studies have trained AI to focus on the more complex matter of volumetric stone burden assessment. 18,26 The AI of Babajide et al, while efficient in identifying the presence of stones, fell short in terms of volumetric precision (ie, a Dice score of 0.66). 26 Subsequently, Elton and colleagues designed a 13-layer CNN; this excellent program had only a slightly inferior accuracy for stone volume determination compared to our program (R[0.97).…”

“…18,26 The AI of Babajide et al, while efficient in identifying the presence of stones, fell short in terms of volumetric precision (ie, a Dice score of 0.66). 26 Subsequently, Elton and colleagues designed a 13-layer CNN; this excellent program had only a slightly inferior accuracy for stone volume determination compared to our program (R[0.97). 18 Unfortunately, their work did not include a determination of Dice score (ie, intersection-over-union) of the testing dataset, as this is a better metric for image-based AI algorithm assessments.…”

Theranostic roles of machine learning in clinical management of kidney stone disease

“…To deal with noisy CT, Elton et al [56] have employed CNN (U-Net model) for automated detection and volume quantification of small stones in coronal CT images. Babajide et al [57] have analyzed the efficacy of a machine learning method to detect and characterize kidney stones automatically compared with manual diagnosis. The data have shown that the machine learning algorithm more accurately approximates the stone boundary with both sensitivity and specificity of 100 % [57] .…”

“…Babajide et al [57] have analyzed the efficacy of a machine learning method to detect and characterize kidney stones automatically compared with manual diagnosis. The data have shown that the machine learning algorithm more accurately approximates the stone boundary with both sensitivity and specificity of 100 % [57] .…”