Urology has always been in pole position to endorse new technologies (lasers, 3-dimensional [3D] printing and imaging, augmented and mixed reality, robotics) due to its diversity in clinical, surgical, oncologic and diagnostic applications.^1-6 Artificial intelligence (AI) is no exception, and it finds a natural dimension and fertile ground in our field. In this brief overview, we want to highlight the pros and cons of adopting AI in urology and its possible impact on our daily practice.

In the last few years, we have witnessed increasing usage of AI in health care that ranges from optimizing patient workflow and increased diagnosis accuracy as well as enhanced radiological and pathological image computer analysis to “precision medicine” granted by the big data analysis powered by smart data recording. However, before going further, a precise classification of the AI basic taxonomy is needed to provide a better understanding of this new field of technology (see figure).^7-12

Figure. AI taxonomy.

The applications of this new technology range from tumor detection and diagnosis to surgical training and prognosis prediction. In particular, deep learning (DL) model-based techniques have shown potential applications in urologic oncology since they offer a noninvasive characterization of the tumor using a group of quantifiable tumor metrics—such as the radiomics signature—which can be extracted from multimodality medical images.

The applications of machine learning (ML) and DL modeling to prostate cancer (PCa) are rapidly increasing.^7,13 ML algorithms can be used to recognize groups of genes responsible for tumor development and therefore could allow a more focused screening for specific individuals who express these genes. In terms of diagnosis, ML algorithms are now being tested to perform prostate segmentation (organ automated contouring and peripheral zone vs. central zone boundary) and tumor characterization. ML-based technology has shown potential to identify PCa location and extracapsular extension, assisting the surgeon with an augmented image-guided procedure.

Another example can be the assessment of bladder cancer (BCa).¹⁴ Currently, tumor stage evaluation is based on cystoscopy findings, transurethral resection of bladder tumor (TURBT) and imaging. A standard disease assessment method is direct visualization of BCa from cystoscopy and routine cross-sectional imaging of the abdomen and pelvis using computerized tomography or magnetic resonance imaging (MRI). Cystoscopy, TURBT and cross-sectional imaging represent the gold standard procedures for assessing the staging and other pathological information for the appropriate management of the patients. However, these approaches are not immune from limitations and could have significant impact on the quality of life and cost of care. A DL-based multi-omics approach could integrate patients’ pathological and radiological characteristics⁸ to understand the behavior of the disease. For example, this new approach could help in predicting the response to neoadjuvant chemotherapy, distinguishing between patients who are responding to it vs. those who may not benefit, and avoiding any unnecessary delay to radical cystectomy.

Small renal masses that are incidentally detected represent a heterogeneous group of potential tumors, and an accurate evaluation of malignant behavior should be performed before patients undergo treatment. Although in some cases a renal biopsy could be achieved, it could lead to complications. Radiomics ML-based analysis could help to discern between indolent vs. malignant renal masses, helping the urologist—in select cases—perform patient-tailored surveillance protocol or followup.^15,16

AI and ML have shown promising results in improving surgical skill in urology (“surgical AI”).⁹ Current efforts in this area already include the auto-segmentation of surgical activity within a surgical video such that areas of interest can be readily identified for further investigation.¹⁷ Similarly, robust efforts have already been seen in the prediction of patient outcomes after surgery, utilizing both patient factors and surgeon metrics as predictors.¹⁸ Finally, the automation of surgeon technical skills assessment is an area ripe for progress. A recent National Cancer Institute R01 award to our institution will make this aspiration a reality (1R01CA251579; Principal Investigator: P. I. Hung). Future work in surgical AI may include semi-automation or even full automation of surgical tasks.

New technological frontiers inevitably open new ethical issues. In a similar technology-driven context it is natural to wonder about the pros and cons of having dedicated human and artificial resources in a clinical department. We recently established the new AI Center within the Department of Urology at University of Southern California (USC) (“UroAI”). To our knowledge, this is the first such center in a urology department in the country. UroAI adds a vibrant, futuristic dimension to USC Urology with the mandate to inspire surgical-related AI research by leveraging innovations in novel algorithms and their clinical application to advance the field of urology. Our USC Urology team has been working on AI initiatives for quite some time now, including spacing from surgical AI, radiomics in urologic oncology and automated image recognition for MRI and histology.³ All 3 domains are current, active areas of research. (Please follow our Twitter account @urology_AI to get updates.)

Endorsing AI in the clinical and surgical settings represents a unique opportunity for interdisciplinary collaborations between computational scientists and medical experts, who can bring different perspectives in a new era of patient care.