Machine Learning Shows Promise for Urologic Oncology and Kidney Stone Management at AUA19

“Ongoing advances in machine learning, automation, and robotics will continue to change for the better the way we practice Urology,” said American Urological Association (AUA) President J. Brantley Thrasher, MD, FACS during his presidential address at last year's AUA annual conference.

Thrasher convincingly spoke of the influence that artificial intelligence (AI) would soon have on the specialty. This year’s conference in Chicago proved to be a testament to the truth of those words.

Machine learning, which was once relegated to the depths of academic computer science, has seen a surge in common consumer applications, ranging from self-driving cars to Siri in your pocket. Nearly a quarter of North America industries have made promises to scale up machine learning and nearly half around the world will scale up this year, including finance, transportation, manufacturing, retail, and of course, medicine.

For a quick summary, machine learning is an umbrella term for analytical techniques that use algorithms and statistical methods to learn from data, and then make new predictions or classifications. The ever-expanding amount of health care data, combined with the readily available computational capacity to analyze that data, has led to a boom in machine learning and its other associated buzzwords in our field. I claim no expertise in this area, just a personal interest. But personal interests aside, AI is a topic deserving of special attention due to its pervasiveness.

Urologists are quick to brandish themselves as early adopters of new technology. Not to be outdone in developing machine learning applications, they have made immense strides in its use for improving urologic care. From posters to the plenary, AUA19 boasts of new machine learning applications in Urologic Oncology, kidney stone management, surgical outcomes, and many other aspects of the field.

Urologic Oncology

In the field of prostate cancer, there is a plethora of research on how to identify patients who need a prostate biopsy. Prostate MRIs are frequently used to identify these patients, but some lesions in the prostate are still equivocal for cancer risk. Enter machine learning.

One presentation titled “A Machine Learning Model for Predicting Cancer Presence in Prostate Biopsy Targets using MRI” described how an algorithm was created using MRI lesion characteristics to identify, with a relatively high degree of accuracy, lesions that could harbor cancer. This essentially is a “virtual biopsy” and, although there is still work to be done, a potential use could be to identify patients with a low risk of cancer. Therefore, these patients may be spared an uncomfortable biopsy procedure (that is not without its own risks).

Another interesting presentation titled “Using Machine Learning Tools to Predict Prostate Cancer Upgrading After Robotic Radical Prostatectomy” compared radical prostatectomy pathology to the biopsy pathology using a machine learning algorithm to predict an upgrade of Gleason score between the biopsy and the prostatectomy. Two different techniques were employed, both of which outperformed conventional nomograms. From these two presentations alone, it is clear that machine learning will be used in multiple stages of prostate cancer management between pre-biopsy assessment to post-treatment prognostication.

One of the most shocking uses of machine learning was seen in bladder cancer diagnosis and surveillance. A late-breaking abstract presented at the plenary session described the development of an algorithm to identify papillary bladder cancer from cystoscopic video with impressive accuracy. Astonishingly, this algorithm identified and localized cancerous lesions in the bladder from cystoscopy footage just like you would see in a sci-fi movie. This was intended to aid in diagnosis as a standard cystoscopy may miss as much as 20 percent of cancers.

Kidney Stones

A number of presentations about machine learning focused on applications in stone management as well.

One presentation titled “Deep Learning Computer Vision Algorithms for Detecting Kidney Stone Composition: Towards an Automated Future” discussed an algorithm that uses images of kidney stones to classify them based on their composition with a high degree of accuracy, particularly for Calcium Oxalate Monohydrate stones. The authors even asserted that this has the potential to replace traditional laboratory analysis for determining the stone composition, offering a serious cost-saving impact.

Another presentation described the use of a machine learning algorithm to localize a kidney stone on fluoroscopic images, with a precision of 70 percent. Applications of this algorithm apparently include automatic stone targeting and real-time tracking during extracorporeal shockwave lithotripsy.

A myriad of other presentations expounded on the integration of machine learning and urologic care. (I'm only able to touch on a few of them here.) With such a breadth of applications, I suspect that next year’s AUA annual conference will have even more presentations about machine learning.

What does this mean for modern urologists? Are computers going to take over our jobs? Unlikely. At least, not anytime soon. There are still considerations that will need to be addressed. Some of these considerations are related directly to the accuracy of the machine learning models. I imagine that these will continue to improve and look forward to seeing these developments at future AUA conferences.

However, there are other considerations, including the quality of the data and the interpretation of the results. What is clear is that machine learning is meant to be a tool in the urologists’ armamentarium. Machine learning is hardly a replacement for clinical judgement. As this year’s AUA has shown, the applications of machine learning include improving patient selection for procedures, providing prognostic information, potentially reducing health care costs, and automating tasks. As such, machine learning is really an instrument that allows us to focus on our most important goal: improve the lives of our patients and, as Thrasher reminded us last year, the potential to “change for the better the way we practice Urology.”

Mahir Maruf, MD is a research associate in the Robert D. Jeffs Division of Pediatric Urology, Brady Urological Institute at the Johns Hopkins University School of Medicine and an incoming Urology resident at the University of Michigan. He Tweets at @MahirMaruf. He reports no conflicts of interests.


More from Op-Med