STORZ MEDICAL – Literature Databases
STORZ MEDICAL – Literature Databases
Literature Databases
Literature Databases

Choo MS et al, 2018: A Prediction Model using Machine Learning Algorithm for Assessing Stone-Free Status after Single-Session Shockwave Lithotripsy to Treat Ureteral Stones.

Choo MS, Uhmn S, Kim JK, Han JH, Kim DH, Kim J, Lee SH.
Department of Urology, Hallym University Dongtan Sacred Heart Hospital, Hwaseong-si, Geonggi-do, Republic of Korea; Department of Computer Engineering, Hallym University (SU, DHK, JK), Chuncheon, Gangwon-do, Republic of Korea.

Abstract

PURPOSE: The aim of this study was to develop and validate a decision support model using machine learning algorithm to predict treatment success after single-session SWL in ureteral stone patients. MATERIALS AND METHODS: Of the 1803 patients treated with SWL, we selected ureteral stone patients with an available preoperative CT scan. Treatment success after single-session SWL was defined as being free of stones or residual stone fragments less than 2 mm in length by CT scan or KUB two weeks later. A decision tree analysis using a machine learning algorithm identified the relevant parameters. A decision support model was developed to calculate the treatment success probability.
RESULTS: A total of 791 patients were enrolled, and the mean length and volume of the stones were 5.9±2.3mm and 89.3±140.0mm3, respectively. The overall treatment success rate after SWL was 64.4% (n=509). The rates for upper, middle and lower ureter stones were 59.8%, 65.5%, and 69.6%, respectively. In the decision tree analysis, volume, length, and the Hounsfield unit were the top three performance criteria factors, and the decision models were constructed with all possible combinations of factors. The model with 15 factors exhibited over 92% accuracy and an AUC of 0.951. The prototype of decision support model using machine learning algorithm is accessible at 'http://pisces.hallym.ac.kr/ESWL'.
CONCLUSIONS: We applied a machine learning algorithm, a subfield of artificial intelligence, to predict the outcome after single-session SWL for ureteral stones and developed a 92.29% accurate decision model with 15 factors and an AUC of 0.951.

J Urol. 2018 Jul 20. pii: S0022-5347(18)43554-9. doi: 10.1016/j.juro.2018.06.077. [Epub ahead of print]

 

0
 

Comments 1

Peter Alken on Friday, 21 December 2018 16:07

This paper is worth reading because it is a new promise to predict the future.
“Stone length as well as stone width and height measured from a CT image have been shown to be good predictors. In addition, uncommonly used urinalysis factors such as urine specific gravity, pH and microscopic red blood cells contributed to the accuracy of predicting treatment outcomes. Patient age and gender were also useful to predict success after single session SWL treatment. Regarding the study factors notably the proposed model used 15 factors and predicted the treatment outcome with 92.29% accuracy.”
In a step wise fashion factors were added to increase the precision of the predicting model. Beginning with only one factor, stone volume, factors like urinary pH or specific gravity of urine were included.
http://storzmedical.com/images/blog/Choo_MS.JPG
In the decisional tree, a figure shown in the supplementary material, urinary pH is added at some nodes with values > 5 or ≤ 5 and at others with values > 6,5 and ≤ 6,5; similar the specific gravity is introduced with values of > 1.028 and ≤ 1,028 or > 1,018 and ≤ 1,018 resp. It is difficult to imagine how these factors could influence stone disintegration. Additionally there is no information why these differing values were selected and at what time of day they were measured. We know that both values show significant physiological diurnal variations.
All in all I doubt that this nice statistical model will make it into clinical practice just as the other 9 models referenced in the article.
Unfortunately the access to the internet homepage “to make it easy to apply the prototype decision support model using the machine learning algorithm in clinical practice (prototype available at: http://pisces.hallym.ac.kr/ESW)” was not working at 26.10.2018.

This paper is worth reading because it is a new promise to predict the future. “Stone length as well as stone width and height measured from a CT image have been shown to be good predictors. In addition, uncommonly used urinalysis factors such as urine specific gravity, pH and microscopic red blood cells contributed to the accuracy of predicting treatment outcomes. Patient age and gender were also useful to predict success after single session SWL treatment. Regarding the study factors notably the proposed model used 15 factors and predicted the treatment outcome with 92.29% accuracy.” In a step wise fashion factors were added to increase the precision of the predicting model. Beginning with only one factor, stone volume, factors like urinary pH or specific gravity of urine were included. [img]http://storzmedical.com/images/blog/Choo_MS.JPG[/img] In the decisional tree, a figure shown in the supplementary material, urinary pH is added at some nodes with values > 5 or ≤ 5 and at others with values > 6,5 and ≤ 6,5; similar the specific gravity is introduced with values of > 1.028 and ≤ 1,028 or > 1,018 and ≤ 1,018 resp. It is difficult to imagine how these factors could influence stone disintegration. Additionally there is no information why these differing values were selected and at what time of day they were measured. We know that both values show significant physiological diurnal variations. All in all I doubt that this nice statistical model will make it into clinical practice just as the other 9 models referenced in the article. Unfortunately the access to the internet homepage “to make it easy to apply the prototype decision support model using the machine learning algorithm in clinical practice (prototype available at: http://pisces.hallym.ac.kr/ESW)” was not working at 26.10.2018.
Monday, 10 February 2025