UROLOGY : CLINICAL UPDATES FROM THE PRACTICE
ROLE OF ARTIFICIAL INTELLIGENCE IN KIDNEY STONES AUTHORS Samarpit Rai , MD , & Murali Ankem , MD
10 LOUISVILLE MEDICINE
INTRODUCTION & BACKGROUND
Artificial intelligence ( AI ) is gaining momentum in every aspect of life , so it is no wonder it is taking impressive strides in clinical medicine as well . AI tries to emulate the human cognitive functions , i . e . reasoning , learning , memory and solutions to problems . 1 Traditional statistical approaches cannot handle big data and machine generated complex algorithms may be the future . In this article , we aim to introduce the role of AI in the diagnosis and management of urinary stone disease and a look into the future applications in the field of endourology .
DIAGNOSIS OF URINARY STONES
Computerized tomography ( CT ) with stone protocol is the imaging of choice for diagnosis of kidney stones and their composition . 2 The immense data generated with CT scans is used in constructing a convolutional neural network ( CNN ) to help in diagnosis of nephrolithiasis . The throughput in the emergency department could be facilitated by using CNN to expedite diagnosis without radiologists .
STONE LOCATION
Langkvist et al 3 used a set of 465 CT scans for ureteral stones and developed a CNN , which was tested in 88 CT scans . The sensitivity was 100 %, however , they had a high false positive rate of 3.7 per scan . Given the total number of slices analyzed ( 19,000 ), this false positive rate is not high , though it ’ s not yet ready for primetime .
Parakh and coworkers 4 reported 94 % and 96 % sensitivity and specificity , respectively , with the area under curve of 0.95 . One of the challenges they faced was differentiation of ureteral calculi from pelvic phleboliths .
STONE COMPOSITION
The treatment strategy for urinary stones depends on the composition of the stones . In some cases , this information is available in the form of stone analysis report . Without a previous stone analysis , urologists depend on the radiologic characteristics for choosing the optimal approach . Infrared spectroscopy is the gold standard for determining the stone composition and this was compared to digital images of stones that were passed by patients . Then a supervised learning algorithm , using a multiple decision tree based on stone texture and color , was developed to determine the stone composition . Serrat & colleagues 5 had 63 % success rate in predicting correctly with this model . Black et al 6 used digital images to train an advanced deep learning algorithm and predicted correctly 85 % of the time . Their accuracy was very high for uric acid stones at 94 %. Kazemi et al 7 implemented ensemble machine learning methods to forecast the stones ’ contents . They analyzed 41 clinical parameters and 40 classification algorithms and developed a predictive model , which yielded 97 % accuracy rate .
STONE FREE RATES
Using the clinical presentation , radiologic findings and previous history of treatments helps an experienced urologist come up with a treatment protocol , with prediction of outcomes after different operative and non-operative strategies . Machine learning models can be developed and used to predict the outcomes as well .