Artificial Intelligence in Functional Urology: How It may Shape the Future

Imad Bentellis; Sonia Guérin; Zine-Eddine Khene; Rose Khavari; Benoit Peyronnet


Curr Opin Urol. 2021;31(4):385-390. 

In This Article

Artificial Intelligence Applications for Basic Research in Functional Urology

Radiomics refers to a group of emerging technologies that generate image-driven biomarkers that reflect cellular and molecular properties of tissues.[7,8] These radiomic features can be associated with AI because of its better ability of handling a massive amount of data compared with the traditional statistical methods. As of now, there is very little data concerning functional urology. In a series of 40 adult spina bifida patients, Khene et al.[8] used texture analysis of the bladder wall on computed tomography (CT) and investigated its relationship with poor bladder compliance (PBC) and detrusor overactivity (DO) on urodynamics. A Lasso penalized regression identified texture parameters significantly associated with the urodynamic findings: skewness for PBC and kurtosis for DO. Delving into the association between texture parameters and specific urodynamic features may help to decipher the underlying mechanisms of some neurogenic lower urinary tract dysfunction.

The emergence of functional brain imaging in the early 2000s has revolutionized the understanding of the lower urinary tract neural control.[9] With imaging progresses over the past two decades, the amount of data drawn from functional magnetic resonance imaging (fMRI) has exploded. Functional connectivity analysis can reveal the complex interactions between different brain regions during storage and voiding. ML algorithms can be used to overcome the limitations of 'standard' imaging analysis by processing all connectivity information directly in the training and classification approach. In a recent study, Karmonik et al.[10] used fMRI evaluations in multiple sclerosis (MS) patients to identify brain regions linked to voiding dysfunction (VD). Four ML approaches were used to assess the relative functional connectivity strengths in various brain areas: random forests (RF), artificial neural networks (ANN), Generalized linear model, and partial least squares model (PLS). The best one (PLS) allowed to obtain the area under the ROC curve of 0.89 in the identification of the Voiding Initiation Network of these patients, followed by the RF model with 0.86. This finding is particularly interesting because it is known that voiding pattern of MS patients and structural MRI findings are not correlated.[11] This machine-learning approach used in functional MRI may thus open a new pathway to phenotype patients and help to tailor the therapeutic approach (e.g. using cortical stimulation techniques in those with demonstrated dysfunction of the voiding initiation network)