*Review* **Outcomes Related to Percutaneous Nephrostomies (PCN) in Malignancy-Associated Ureteric Obstruction: A Systematic Review of the Literature**

**Francesca J. New 1, Sally J. Deverill <sup>2</sup> and Bhaskar K. Somani 1,\***


**Abstract:** Background: Malignant ureteric obstruction occurs in a variety of cancers and has been typically associated with a poor prognosis. Percutaneous nephrostomy (PCN) can potentially help increase patient longevity by establishing urinary drainage and treating renal failure. Our aim was to look at the outcomes of PCN in patients with advanced cancer and the impact on the patients' lifespan and quality of life. Materials and Methods: A literature review was carried out for articles from 2000 to 2020 on PCN in patients with advanced malignancies, using MEDLINE, EMBASE, Scopus, CINAHL, Cochrane Library, clinicaltrials.gov, and Google Scholar. All English-language articles reporting on a minimum of 20 patients who underwent PCN for malignancy-associated ureteric obstruction were included. Results: A total of 21 articles (1674 patients) met the inclusion criteria with a mean of 60.2 years (range: 21–102 years). PCN was performed for ureteric obstruction secondary to urological malignancies (*n* = −633, 37.8%), gynaecological malignancies (*n* = 437, 26.1%), colorectal and GI malignancies (*n* = 216, 12.9%), and other specified malignancies (*n* = 205, 12.2%). The reported mean survival times varied from 2 to 8.5 months post PCN insertion, with an average survival time of 5.6 months, which depended on the cancer type, stage, and previous treatment. Conclusions: Patients with advanced malignancies who need PCN tend to have a survival rate under 12 months and spend a large proportion of this time in the hospital. Although the advent of newer chemotherapy and immunotherapy options has changed the landscape of managing advanced cancer, decisions on nephrostomy must be balanced with their survival and quality of life, which must be discussed with the patient.

**Keywords:** prostate cancer; nephrostomy; quality of life; survival; decision making

#### **1. Introduction**

Malignancy-associated ureteric obstruction occurs in a variety of pelvic cancers, often as a late manifestation, which can be secondary to locally advanced disease or nodal metastases. Treatment consists of various options ranging from ureteric stent insertion (retrograde or antegrade), to percutaneous nephrostomy (PCN), to other forms of urinary diversion. While these procedures can help to improve renal function, they also risk complications and can have a profound effect on the quality of life (QoL). Stenting can consign the patient to stent symptoms (which may include frequency, urgency, pain, haematuria, and dysuria), and regular stent changes (typically every 6–12 months) under a general anaesthetic but is generally believed to be better for QoL than long-term PCN, although give the underlying disease this might be challenging [1].

Unfortunately, in the context of locally advanced pelvic cancers, there are often scenarios whereby a patient will start with a retrograde ureteric stent (RUS), but subsequently, as this fails, it necessitates PCN insertion. In the event that a RUS change or drainage fails, the decision to proceed with PCN often marks disease progression. Without treatment of

**Citation:** New, F.J.; Deverill, S.J.; Somani, B.K. Outcomes Related to Percutaneous Nephrostomies (PCN) in Malignancy-Associated Ureteric Obstruction: A Systematic Review of the Literature. *J. Clin. Med.* **2021**, *10*, 2354. https://doi.org/10.3390/ jcm10112354

Academic Editor: Konstantinos Stylianou

Received: 31 March 2021 Accepted: 21 May 2021 Published: 27 May 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

malignant ureteric obstruction, the patient will deteriorate over time with symptoms of uraemia, fluid overload, electrolyte disturbances, flank pain, urinary infections, reduction in alertness, renal failure, and subsequent death [2]. Patients with advanced malignancies, who present with acute renal failure (ARF) due to malignant ureteric obstruction, are often poor surgical and/or anaesthetic candidates, and therefore PCN, which can be done under local anaesthesia (LA), is often preferred. Similarly, it is not always possible to insert primary retrograde stents in the context of locally advanced pelvic malignancies [3–5].

Percutaneous nephrostomy has a high rate of technical success; however, periprocedural complications can occur. These may include sepsis, bleeding or vascular injury, perirenal haematoma, and injury to surrounding structures such as colon, liver, and lung [3]. Furthermore, PCN can block, dislodge, develop line or component fracture, become infected, or colonised with bacteria, and patients can develop skin reactions, cellulitis, or abscesses [3]. Such complications can result in multiple readmissions to hospital, often needing a change in PCN, which can also significantly impact their QoL [1]. Emergency readmissions also happen if the PCN falls out completely, needing a new nephrostomy placement as a matter of urgency [6]. Patients with advanced cancers who develop infections secondary to nephrostomy are at a high risk of deterioration, especially if they are receiving immunosuppression such as chemotherapy or immunotherapy.

Most studies looking at malignancy-associated ureteric obstruction cover an extremely heterogenous population, with multiple different aetiologies and presentations. Treating malignant ureteric obstruction is an ever-changing landscape, and as newer cancer treatments become available, this continues to evolve. We aimed to review the quality of evidence available to date in this group of patients, establishing outcomes of PCN in malignancy-associated ureteric obstruction, assessing the risk of complications, life expectancy, QoL and potential indicators of favourable versus poorer outcomes.

#### **2. Materials and Methods**

#### *2.1. Study Population*

Population: Adults with malignancy-associated ureteric obstruction. Intervention: Percutaneous nephrostomy. Comparator: Not applicable for this study. Outcome: Life expectancy, QoL, and outcomes related to PCN.

#### *2.2. Inclusion Criteria*

Studies reporting on patients with advanced malignancies with ureteric obstruction. English-language studies reporting on a minimum of 20 patients.

#### *2.3. Exclusion Criteria*

PCN insertion for benign disease. Studies that included primary ureteric stenting as the only treatment option. Case reports, laboratory studies, or review articles.

#### *2.4. Search Strategy and Study Selection*

The systematic review was performed as per the Cochrane guidelines and the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) checklist [7]. The database searched were MEDLINE, EMBASE, Scopus, CINAHL, Cochrane Library, clinicaltrials.gov and Google Scholar from January 2000 to December 2020. The search terms included 'Nephrostomy', 'percutaneous nephrostomy', 'PCN', 'urinary drainage', 'stent', 'ureteric stent', 'prostate', 'ovarian', 'cervical', 'bowel', 'malignancies, malignancy or cancer', and 'pelvic, gynaecological, colorectal, urological'. Boolean operators (AND, OR) were used with the above search terms to refine the search. Two reviewers (S.D. and F.N.) independently identified all the studies that matched the inclusion criteria and any discrepancies were resolved by consensus with the senior author (BKS).

#### *2.5. Data Extraction and Analysis*

The primary outcome measures were complications after PCN, time spent in the hospital after PCN, and survival times after their first PCN. Secondary outcomes were QoL after PCN and differences in outcomes based on the cancer sub-type. Information was collected on the year of publication, type of malignancy, patient demographics, and outcomes of PCN. Data were collected using Microsoft Excel 2019 (version 19.0). A narrative review was done due to heterogeneity of the studies and data available.

#### **3. Results**

#### *3.1. Literature Search and Included Studies*

After an initial search of 110 articles, 21 studies (1674 patients) met the inclusion criteria for the final review (Figure 1) [3,6,8–26]. A full breakdown of the patient demographics can be seen in Table 1.

**Figure 1.** PRISMA flowchart of the included articles.


**Table 1.**

Demographics,

 survival, cancer sub-type and time spent in the hospital.


**Table 1.** *Cont*.


**Table 1.** *Cont*.

#### *3.2. Patient Characteristics*

There were 1674 patients with a mean age of 60.2 years (range: 21–102 years), although two studies did not state the mean or median age [6,7], and two studies stated the median age [8,9]. The majority of studies were retrospective in nature (*n* = 17), with one prospective study [8] and four where the type of study was not specified (Table 1) [10–13].

PCN was performed for ureteric obstruction secondary to urological malignancies (*n* = 633, 37.8%), gynaecological malignancies (*n* = 437, 26.1%), colorectal and gastrointestinal (GI) malignancies (*n* = 216, 12.9%), and other specified malignancies (*n* = 205, 12.2%) (Table 1) [13]. Fourteen studies documented the length of survival post nephrostomy insertion for the different cancer subtypes [3,8,9,11,12,15–23].

#### *3.3. Primary Outcomes*

#### 3.3.1. Survival Times after PCN

The reported mean survival time varied from 2.6 to 8.5 months post initial PCN insertion, with an average survival time of 5.9 months (Figure 2, Table 1). Five studies documented median survival time as 5.2 months (range: 2–7 months) [8–11,22], and three did not document the survival time post PCN insertion [6,13,24].

**Figure 2.** (**A**): Patients who died on the same admission as nephrostomy (PCN) insertion (**B**): Median survival post PCN insertion.

Romeo et al. [18] documented the survival times post PCN insertion with 40% dead at 6 months and a further 24.4% at 1 year, while Aravantious documented that 67% of the patients were dead within 6 months of a PCN insertion [21] (Table 1). A prostate cancer study by Nariculam and colleagues in 2009 found that the overall mean time to death post PCN was 7.5 months, but if patients developed ureteric obstruction while already on hormones, the mean survival decreased to 4.5 months. In the context of newly diagnosed and hormone-naïve patients, the survival increased to a mean of 16 months (range: 1–38 months) [23]. Similarly, Harris et al. found that survival was longer for the hormone-naïve group (226.5 days) when compared to 100.2 days in the castrate-resistant prostate cancer group [20].

In the context of bladder cancer, Ekici et al. looked at 23 patients with malignant ureteric obstruction due to bladder cancer, including patients with new diagnosis of locally advanced disease, disease recurrence post cystectomy, and those with metastatic disease. There was a mean survival of 4.9 months (range: 1–14 months). Eighteen (78%) died of disease progression or irreversible renal failure after malignant ureteric obstruction during the study period [17].

Romero et al. found that prognosis was worse in patients over 52 years old and in patients with bladder cancer or hormone refractory prostate cancer, rather than cervical cancer, but patient numbers were small (*n* = 43), so this may not be generalisable [18]. Misra et al. reported a median survival post PCN insertion as only 78 days (range: 4–1137 days) and also described that the subset of bladder cancer patients seemed to do more poorly [12]. In contradiction to these findings, Jalbani described an improved median survival in urogenital malignancies (bladder and prostate) of 350 days (range: 150–700 days) when compared to non-urogenital malignancies, except lymphoma (gynaecological, colorectal, breast, and gallbladder cancers) where the median survival was only 25 days (range: 7–80 days) [8].

Folkard et al. found that the average survival time post PCN was 139 days, and there was no significant difference between the cancer subgroups in terms of survival time post nephrostomy. They also showed that a greater improvement in renal function did not improve the survival time. A large proportion of their patients (65.7%) did not undergo further oncological treatment post PCN as they became too frail for it [25].

#### 3.3.2. Prognostic Indicators

Alawneh et al. found that the factors associated with a shorter survival time were type of malignancy, bilateral hydronephrosis, serum albumin <3.5 mg/dL, presence of metastases, ascites, or pleural effusion. Survival was better if patients had only one risk factor, with median survival 17.6 months vs. 1.7 months if four risk factors were present. The overall 12-month survival in their paper was 33.7% [9]. Ishioka [15] found that the factors associated with a poorer prognosis included colorectal cancer, three or more events related to metastatic disease, degree of hydronephrosis, and serum albumin <3 g/dL.

Lienert et al.'s [10] prognostic indicators were consistent with previously discussed studies; a serum albumin <3 mg/dL and three or more events related to dissemination of cancer were factors significantly associated with shorter mean survival. Moreover, a sodium <135 mEq/L was found to be a significant prognostic factor. In this study, degree of hydronephrosis was not found to be a significant prognostic factor.

Nariculum et al. [23] showed that the mean survival for newly diagnosed patients (hormone-naïve) was 16 months (range: 1–38 months), compared to patients who developed ureteric obstruction while on hormones, where the mean survival was only 4.5 months (range: 10 days to 17 months). This was also shown by Harris et al., who showed that hormone-naïve patients survived longer at 226.5 days, compared to 114.3 days in hormone-responsive groups and 100.2 days in the hormone-resistant group. Another prognostic factor was the failure of renal function to improve despite nephrostomies, and if the post-procedure urea and creatinine went below 15 mmol/L and below 250 μmol/L, respectively, then the mean survival time was 192.4 days, but if the renal function did not improve, then the mean survival was only 30.7 days [20].

Romero et al. showed that the poor prognostic factors in their study were age above 52 years and patients with bladder and hormone refractory prostate cancer [18]. Misra also showed that patients with bladder cancer had a worse prognosis [12]. In contrast, Radecka et al. [3] and Jalbani et al. [8] showed an improved survival in patients with

bladder cancer. De Souza et al. demonstrated that the finding of hypotension unrelated to septic symptoms was a risk factor for progression to death [24].

#### 3.3.3. Complications of PCN

Nineteen studies commented on the complication rates (Table 2). The overall complication rate ranged from 7% to 87%. The majority of the complications were minor, including urinary tract infection, haematuria, skin infection, malposition/dislodgement of PCN tubing and self-limiting fever. There was, however, a reasonably high rate of kinking, dislodgement, or loss of nephrostomy requiring reinsertion. There were some major complications described, including two patients who required a nephrectomy due to severe infection and peri-renal abscesses [9].

#### **Table 2.** Complications of percutaneous nephrostomy (PCN) insertion.


McDevitt et al. specifically looked at the number of routine vs. emergency PCN changes. Out of 87 PCN exchanges or reinsertions, only 33% were routine and 67% were for emergency reasons such as infection, obstruction, displacement, or mechanical complications [6].

Insertion of the initial PCN has good rates of technical success. Aravantinos et al. reported a 2.5% failure rate, with no serious complications, a minor temperature rise of 55%, and a transfusion rate of 2.9%; however, they commented on pre-existing anaemia, and therefore this may not be related to the PCN insertion itself. They also reported that a small proportion of patients (4.4%) needed staged a second nephrostomy tube due to persistent uraemia despite a unilateral nephrostomy tube [21].

#### 3.3.4. Bilateral vs. Unilateral PCN

One point of interest was whether in order to improve QoL in patients with bilateral hydronephrosis secondary to malignant ureteric obstruction, a unilateral nephrostomy was sufficient. Thirteen studies commented on whether they inserted unilateral or bilateral nephrostomies. In prostate cancer, one study reported that the mean survival for unilateral nephrostomy patients was better (157.6 days) than for those who required bilateral nephrostomies, whether they were placed simultaneously or staged [20]. This could be due to the fact that they also demonstrated that a worse prognosis is linked with bilateral hydronephrosis. In one study of mixed malignancies, 92% of the patients had bilateral hydronephrosis and their aim was to trial unilateral PCN. Only 4.4% patients required a second-stage nephrostomy due to persistent uraemia despite having a unilateral nephrostomy [21].

#### 3.3.5. Quality of Life after PCN

There are no validated questionnaires specifically looking at QoL with nephrostomies in cancer patients [27]. A wide range of methods for determining quality of life with a nephrostomy were used throughout the studies. Aravantinos et al. [21] used the QoL questionnaire EORTC-QLC-C30 [28] and found that QoL improved at 1 month, and of the different cancer subgroups, it was better in the prostate cancer subgroup. Wilson et al. used the criteria of Grabstald and McPhee to define 'useful quality of life' and found 17/32 (53.1%) did not fulfil such criteria, and the subgroup of bladder cancer patients had poorer outcomes [19]. Misra used the Watkinson criterion (if the patient was able to leave hospital for 6 weeks or more), finding that 64% would have satisfied this criterion [12]. In the studies that measured QoL, only around half of the patients achieved an adequate QoL post PCN insertion.

#### 3.3.6. In-Hospital Stay after PCN

The time spent in hospital following PCN insertion was highly variable and poorly reported (Table 1). Romero found that the percentage of lifetime left that was spent in hospital was 17.7%, and 57.7% of those discharged from hospital had to be readmitted (either due to disease progression or complications from PCN) [18]. Wilson reported a mean hospital stay of 29 days from PCN insertion to death or end of study period, and each patient was readmitted an average of 1.6 times until death [19]. Misra reported a median hospital stay post PCN of 23 days (range: 3–89), with 29% of a patient's end of life spent in hospital [12]. Folkard had a mean hospital stay of 14 days post PCN; however, 39% of the patients were readmitted, and 20% spent their remaining life in hospital [25].

Many patients with advanced malignancies die in hospital despite PCN insertion, and nine studies reported the percentage of patients who died on the same hospital admission as their PCN was placed [8,12,16,18–20,22,24,25]. The mean percentage of patients who died on the same hospital admission as their PCN insertion was 30.8% and ranged from 12.5% to 70% (Figure 2).

#### **4. Discussion**

#### *4.1. Findings of Our Study*

The mean survival time varied from 2.6 to 8.5 months post initial PCN insertion across the studies, with an average survival time of 5.9 months (Figure 2, Table 1). The majority of studies agreed that hormone-naïve prostate cancer had a longer survival time post PCN insertion, whereas bladder cancer, cervical cancer, and hormone refractory prostate cancer all had shortened life expectancies. Poor prognostic indicators throughout the studies were patients who had already undergone cancer treatment, presence of multiple metastasis, type of cancer, degree of hydronephrosis, and a low serum albumin concentration. The number of days spent in hospital post PCN insertion were high (Table 1) and a third of the patients (range: 12.5–70%) died on the same admission while they were admitted to hospital (Figure 2).

#### *4.2. Patient Counselling*

The ethics of palliative urinary decompression have been debated, and many factors must be taken into account, such as the type and stage of malignancy, the ability for further palliative treatment, patient's quality and quantity of life along with their preference. Malignant ureteric obstruction from pelvic malignancies often presents a significant treatment dilemma for urologists. While PCN insertion is relatively safe, patients with advanced malignancies tend to have a higher risk of PCN-related complications (Table 2) and spend a large proportion of their time in hospitals. PCNs should only be pursued after thoughtful counselling regarding further treatment options and likely disease prognosis.

#### *4.3. Quality of Life*

There are no validated questionnaires specifically looking at QoL with nephrostomies in cancer patients [27]. A wide range of methods for determining QoL with a nephrostomy were used throughout the studies, ranging from whether the patient ever left hospital at all, to whether they left hospital for 6 weeks or more (Watkinson criteria [29]), to scoring them on four criteria; of little or no pain, full mental capacity, few complications related to PCN insertion, and the ability to return home (Grabstald and McPhee criteria [19]), to using EORTC-QLC-C30 questionnaires [28]. It is difficult to ascertain whether QoL is worse after PCN insertion due to the procedure, or the progression of the cancer; hence a standardised questionnaire would be useful in ascertaining this and could aid patients in making the decision on whether or not to proceed with a nephrostomy [27].

#### *4.4. Costs of Replacement of PCN*

McDevitt et al. looked at patients who had nephrostomies placed for malignant ureteric obstruction, and the causes of PCN exchanges during the follow-up period. There were 87 exchanges performed, and of those, 29/87 (33.3%) were routine elective changes, but 58/87 (66.7%) were unplanned and due to complications, such as infection (21/87, 33%), obstruction (23/87, 26%) or mechanical complications (14/87, 16%). The cost of emergency exchange vs. routine exchange was modelled to be higher, and they therefore hypothesised that decreasing the length of time to routine exchange from 90 days to 60 days would decrease the amount of readmissions for emergency exchange or replacement, which would decrease the overall cost [6].

#### *4.5. Conversion of PCN to Ureteric Stents*

In some cases, where PCN has been inserted primarily, it may be possible to convert it to an indwelling ureteric stent, usually via antegrade stenting. Wilson and colleagues reported that in 34.4% of cases, they were able to have PCN converted to an indwelling stent [19], and Misra et al. reported that 56% of all PCNs were subsequently antegradely stented and rendered nephrostomy free [12]. Folkard reported that 65% of PCNs were converted to stents.

#### *4.6. Limitations*

Almost all of the studies were retrospective, and with historic data, which made it difficult to apply them to today's cancer patients with recent advances in cancer treatment. These studies cover a heterogeneous population with some having a variety of different primary cancers, while others focus on a single cancer type, which makes interpretation difficult. As novel immunotherapy and chemotherapy options emerge, the ability to predict prognosis is more guarded, and newer information is needed to aid decision making. There were no data from situations where patients presented with hydronephrosis and the decision was not to perform PCN, and how their QoL and length of life compared to those with PCN.

Since the studies reported included a wide time interval (from 2003 to 2020), it should be appropriate to take into account that some malignancies have improved treatment options with potential benefits to prognosis and quality of life. For example, in colorectal cancer, starting from 2004 several drugs have been introduced (cetuximab, bevacizumab, and panitumumab) with advantage on cancer-specific survival. Similar improvements have been reported in prostate cancer from 2011 with new hormone-based therapies (abiraterone and enzalutamide) in metastatic castration-resistant patients, and from 2015 in metastatic hormone-sensitive patients. This treatment may also affect the quality of life and the number of days spent in hospital. Moreover, in selected cases, the option of a new treatment line can justify the insertion of ureteric stent or nephrostomy.

The retrospective nature of the included papers with different inclusion criteria makes it liable to selection bias and hence difficult to draw meaningful comparisons. Given that almost a third of the patients died on the same hospital admission as their PCN insertion suggests that a high number of reported PCNs were performed for palliative reasons. The decision on nephrostomy would have to be individualised for a given patient and must take into account their medical condition and underlying disease status.

#### *4.7. Areas of Future Research*

Prognosis of patients with malignant ureteric obstruction is mostly dependent on further treatment strategies. In recent years, there has been a big leap in oncological therapies, many of which are reliant on good renal function. In many situations now, where there is malignant ureteric obstruction, a patient may still have further options for palliative chemotherapy, immunotherapy or novel hormone therapies. However, if there are no options in reserve, the prognosis is poor with or without nephrostomies, and end-of-life care should be discussed with the patient and relatives, rather than proceeding with invasive interventions that have no impact on disease progression. Complications and death due to locally invasive cancer should be weighed against complications and death due to uraemia.

#### **5. Conclusions**

There is little doubt about the benefits of percutaneous nephrostomy for patients with a new diagnosis of disease, allowing improvement of renal function to allow staging investigations. However, in patients in the end stages of their cancer, PCN insertion should only be placed after thoughtful counselling regarding further treatment options available and disease prognosis, given that with advanced malignancies, many patients have a short life expectancy, spending most of their time in the hospital with a poor quality of life.

**Author Contributions:** Conceptualization, B.K.S.; methodology, B.K.S., F.J.N., S.J.D.; formal analysis, F.J.N., S.J.D.; data curation, F.J.N., S.J.D.; writing—original draft preparation, S.J.D., F.J.N.; writing review and editing, F.J.N., B.K.S.; supervision, B.K.S.; All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interests.

#### **References**


## *Review* **Artificial Intelligence and Its Impact on Urological Diseases and Management: A Comprehensive Review of the Literature**

**B. M. Zeeshan Hameed 1,2,3,4, Aiswarya V. L. S. Dhavileswarapu 5, Syed Zahid Raza 6, Hadis Karimi 7, Harneet Singh Khanuja 8, Dasharathraj K. Shetty 9, Sufyan Ibrahim 3,10, Milap J. Shah 1,3, Nithesh Naik 3,4,11,\*, Rahul Paul 12, Bhavan Prasad Rai <sup>13</sup> and Bhaskar K. Somani 1,3,14**


**Abstract:** Recent advances in artificial intelligence (AI) have certainly had a significant impact on the healthcare industry. In urology, AI has been widely adopted to deal with numerous disorders, irrespective of their severity, extending from conditions such as benign prostate hyperplasia to critical illnesses such as urothelial and prostate cancer. In this article, we aim to discuss how algorithms and techniques of artificial intelligence are equipped in the field of urology to detect, treat, and estimate the outcomes of urological diseases. Furthermore, we explain the advantages that come from using AI over any existing traditional methods.

**Keywords:** urology; artificial intelligence; machine learning; urinary incontinence; kidney stone disease; fertility; reproductive urology; renal cell carcinoma; hydronephrosis; urinary reflux; urolithiasis; endourology; pediatric urology; prostate cancer; bladder cancer

## **1. Introduction**

Advances made in digital technologies, electronic health records, and computing power are producing vast amounts of data in the medical field [1]. With expanded channels, quantity, and quality of data, physicians encounter new obstacles while performing data analysis to establish a reliable diagnosis, planning individualized care, and forecasting the

**Citation:** Hameed, B.M.Z.;

S. Dhavileswarapu, A.V.L.; Raza, S.Z.; Karimi, H.; Khanuja, H.S.; Shetty, D.K.; Ibrahim, S.; Shah, M.J.; Naik, N.; Paul, R.; et al. Artificial Intelligence and Its Impact on Urological Diseases and Management: A Comprehensive Review of the Literature. *J. Clin. Med.* **2021**, *10*, 1864. https://doi.org/10.3390/jcm10091864

Academic Editors: Kent Doi, Andreas Skolarikos and Emilio Sacco

Received: 15 March 2021 Accepted: 8 April 2021 Published: 26 April 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

future. Thus, physicians are now relying on artificial intelligence (AI) to build automated models to enhance patient treatment across all aspects of healthcare [2].

In the healthcare industry, AI refers to all the applications, systems, algorithms, and devices that help physicians in providing healthcare based on computer systems and big data. Medical data are ideally used for advising doctors and patients during the decisionmaking process and identifying the most suitable treatment. The role of AI here is to create new methods for analyzing labor-intensive data, which involves the usage of disciplines of AI. Along with providing improved patient care, it will also enhance efficiency and research and development (R&D), in addition to highlighting disease patterns and correlations earlier than what would be possible via traditional methods. In recent times, AI has seen an explosion in investment and application in the field of medicine, as there is cumulative evidence that it may enhance the delivery of healthcare [3]. This article discusses how AI algorithms and techniques are used in the medical field to detect, treat, and estimate the outcomes of urological diseases and further explains the advantages of using AI over any existing methods.

#### **2. Materials and Methods**

#### *2.1. Search Strategy and Article Selection*

A non-systematic review of the literature associated with urology and artificial intelligence that was published between the years 2010 and 2020 was conducted in October 2020 using PubMed and MEDLINE, along with Scopus and Google Scholar. The search strategy involved using a search string based on a set of keywords that included the following: urology, artificial intelligence, machine learning, urinary incontinence, kidney stone disease, fertility, reproductive urology, renal cell carcinoma, hydronephrosis, urinary reflux, urolithiasis, endourology, pediatric urology, prostate cancer, and bladder cancer.

Inclusion criteria:


Exclusion criteria:


The review of the literature was performed in compliance with the guidelines for inclusion and exclusion criteria. The assessment of titles and abstracts followed by the screening and assessment of the full article text was done according to the inclusion criteria for the selected articles. Further, a manual review of the references list for the chosen articles was conducted to screen for any supplementary work of interest. After a discussion, our authors successfully resolved the disagreements regarding the eligibility for a consensus decision.

#### *2.2. What Is Artificial Intelligence?*

AI emphasizes constructing an autonomous computer that will effectively execute activities done by humans, using sophisticated non-linear mathematical simulation systems with simple building blocks that replicate human neurons. It begins by searching for ways in which a human mind perceives, understands, and executes cognitive functions. The human mind is capable of intelligence, creativity, language recognition, memory, pattern identification, vision, reasoning, and the creation of ties among facts. AI aims to replicate the aforementioned skills to perform wide-ranging functions, from small, manageable tasks like object recognition to complex tasks like forecasting. AI strategies include learning from known data without bias, dependent only on statistical models, and estimating unknown data about the future, thereby making the task of decision-making smarter and easier [2].

The ultimate goal of AI is to build a machine that can perceive its environment and perform tasks to maximize its probability of success. The process of achieving this goal is quite complex and involves various AI subfields such as machine learning (ML), artificial neural networks (ANNs) and deep learning (DL), natural language processing (NLP), computer vision, predictive analytics, evolutionary and genetic computing, expert systems, vision recognition, and speech processing, of which most are used in medicine and healthcare today. Thus, some of them need defining for further discussion on the clinical impact of artificial intelligence on various sub-specialties of urology. Figure 1 shows the relationship between artificial intelligence (AI), machine learning (ML), and deep learning (DL).

**Figure 1.** The relationship between artificial intelligence (AI), machine learning (ML), and deep learning (DL).

Machine learning is the process of teaching a computer to make accurate predictions with the help of algorithms that are trained and made to learn from past experiences in a model that maps features to the corresponding outcome variables. The primary aim of ML is to allow the computer to automatically learn when data are fed. An artificial neural network is the basis of deep learning and a subfield of machine learning. ANNs are defined as highly structured information processing units that, along with their synaptic strengths, called weights, mimic the computational abilities of the human brain and nervous system. The neurons are arranged in a series of layers where the weights are modified gradually during the learning process to yield minimum to no error in the input–output mapping. A neural network that has a significant number of layers is called a deep learning network. Being a subfield that holds paramount importance in AI, neural networks have naturally found promising applications in medicine and healthcare, including cardiology, electromyography, electroencephalography, therapeutic drug monitoring for patient care, and sleep apnea.

Decision trees are one of the predictive modeling approaches used in ML, constructed in an algorithmic approach to identifying ways of splitting the dataset based on different conditions. A simple way to describe a decision tree's working would be to assume a decision node with two or more possible choices. A random forest is an algorithm built with a large number of decision trees that operate as an ensemble. These algorithms are widely adopted in the healthcare industry to determine the patient's most favorable choice, such as telehealth services.

Another AI subfield that plays a critical role in healthcare is natural language processing, which is concerned with the interaction between the computer and human languages. The biggest challenge in clinical research is to deal with data that are lacking in volume or detail, which is a result of data previously being recorded in narrative clinical documentation. Some of AI's most promising uses in healthcare include predictive analytics, precision medicine, diagnostic imaging of diseases, and clinical decision support.

#### *2.3. Applications of AI in Urology*

Urology is a field that rapidly expanded through the history of medicine and is continually growing by adopting newer technology to achieve better patient outcomes [4]. Urology being a healthcare segment that deals mainly with male and female urinary tracts and male reproductive organs, the underlying diseases and conditions in these specific areas could become severe if not addressed earlier. Figure 2 shows the role of artificial intelligence in urology.

**Figure 2.** Role of artificial intelligence in urology.

AI has been widely adopted in the field for early diagnosis, for providing an effective treatment plan, and in surgical specialties. AI is playing an important role and helping physicians in decision making for patients with urological disorders (Figure 3). In the past 5 years, there has been an emergence of studies affirming the safe and effective augmented-reality (AR) experiences in urology. Modern urologists are using a robotic arm with seven degrees of freedom to remove the kidney remotely, using augmented reality with image overlay [5]. AR is significantly improving the integration of information into the surgical workflow, making minimally invasive procedures less complicated for surgeons. It is bringing innovative approaches in medical education as well as surgical interventions, aiding richer and more interactive experiences. Similarly, there are other technologies combined with AI that impact the field to a great extent. Within urology, there are several sub-specialties, among which urologic oncology, reproductive urology, renal transplant, and pediatric urology are some specialties that have leveraged AI to provide better patient care through developments in diagnostics, treatment planning, and surgical skill assessment [5]. The application of AI in these subfields is discussed below.

**Figure 3.** Artificial intelligence in decision making in patients with urological disorders.

#### **3. Diagnosis**

#### *3.1. Urologic Oncology*

It is a sub-specialty of urology that is associated with the diagnosis and treatment of cancers in the urinary tract of the human body and male reproductive organs. Urological cancers are relatively common, with prostate, bladder, and kidney cancers among the 10 most prevalent cancers diagnosed in the United States.

#### *3.2. Prostate Cancer*

The data that are widely used for developing AI algorithms are clinicopathological data of patients abstracted from their electronic medical records (EMRs) because of their high evaluability. With clinical data from 944 Korean patients for predicting organ-confined prostate cancer and non-organ-confined disease, Kim et al. [6] developed a set of ML applications (Table 1). In comparison, Partin tables achieved an accuracy of 66% when using the same dataset. This study highlighted that one can achieve better forecasting results using ML algorithms than using standard statistical models.

Researchers have suggested methods of using AI to simplify the diagnosis and classification of prostate cancer, which has become possible due to the advances in medical imaging and the evidence surrounding it. Using various radiomic features from multiparametric MRI (Magnetic resonance imaging), AI applications have been equipped for detecting prostate cancer [7,8] or for estimating multiparametric MRI Gleason scores [9,10] (Table 1). What also makes AI better than traditional diagnostic standards is its ability to get trained by and learn from complex, multi-variable, big data, thereby improving over time. The ML models displayed an average performance increase of 33–80% for MRI-negative biopsy-positive and 30–60% for MRI-positive biopsy-negative patients when developed using Prostate Imaging Reporting and Data Systems. Fehr et al. [10] observed that ML

algorithms had an advantage over unimodal classifiers as they performed more effectively in both identifying the disease and forecasting the correct Gleason score.



Area Under the ROC Curve (AUC); Neural Network (NN); Support Vector Machine (SVM); Prostate Specific Antigen (PSA); Naive Bayes (NB); Bayesian Networks (BNs); Classification and Regression Tree (CART); Random Forest (RF); Quadratic Discriminant Analysis (QDA); Magnetic resonance imaging (MRI); Logistic Regression (LR); Recursive Feature Elimination (RFE); Hematoxylin and Eosin (HE).

> Prostate cancer diagnosis depends on the pathologists reviewing specimen slides as well as assessing the same using Gleason scoring, and while the entire procedure takes a lot of time, it can cause intra-observer bias, depending on the experience of the pathologists. AI-assisted image analysis in clinical pathology combines automated image recognition, examination, as well as evaluation of digitalized tissue specimen images, allowing automatic and standardized pathology diagnosis (Table 1). Kwak et al. [11] developed an AI application for detecting the disease in optical pathology images of varying resolutions. The algorithm was able to achieve an accuracy of >97% on the same using segmented prostate specimen images. The aforementioned group also developed ANNs with the nuclear morphology of prostatic epithelial cells for the detection of cancer [12]. They were able to achieve an AUC (Area under the ROC Curve) score of 0.97 for the diagnosis of prostate cancer, surpassing diagnostic methods using handcrafted nuclear engineering technologies. Nguyen et al. [13] developed an ML algorithm to classify the Gleason score of prostate cancer. The classifier has different AUC scores when considering cancer and non-cancer specimens in distinguishing between epithelial tissue and stromal tissue, specifically 0.97 for the former and 0.87 for the latter. In addition, when provided

five characteristics of histology, the algorithm achieved an AUC of 0.82 in distinguishing Gleason 3 vs. 4 cancer [13].

#### *3.3. Urothelial Cancer*

Bladder cancers, also known as urothelial carcinomas, begin in the cell lining of the bladder (i.e., non-muscle-invasive bladder cancer) and can spread to the muscle wall and beyond, to other tissues (i.e., muscle invasive or metastatic bladder cancer). They are highly curable when detected and treated early. Similar to prostate cancer, radiomic imaging and urinary metabolite markers have been used to diagnose urothelial cancer using AI techniques (Table 2). Xu et al. [14] developed ML algorithms with radiomic mpMRI characteristics for distinguishing between bladder tumor and normal bladder wall. Garapati et al. [15] used morphological and textural features of CT (Computed Tomography) urography for determining the stage of bladder cancer. The algorithm was successful in achieving an AUC of 0.7–0.9 in predicting the stage of cancer when using these radiomic attributes. Shao et al. [16] trained decision trees based on urinary metabolic markers to diagnose bladder cancer. They were able to achieve an accuracy of 76.6%, a sensitivity of 71.8%, and a precision of 86.6%. Ikeda et al. [17] used the technique of transfer learning, which enables anomaly detection by using gastroscopic images, to extract important features that apply to cystoscopic images. The dataset used contained 22 cystoscopic images, and the model was compared to results from actual urologists and medical students, who were divided into groups based on their expertise levels. The median time taken by the AI was 5 s as compared to 634 s by the group of observers and achieved 0.930 as the maximum score for Youden's index.



Computed Tomography (*CT*); convolutional neural network (CNN).

#### *3.4. Renal Cancer*

Detection of renal cell cancer (RCC) in its early stages is crucial for its effective treatment, which can be clinically difficult once it spreads. Clinicians can use metabolomics data along with Raman spectra for building AI models, which are effective in the diagnosis of RCC during or before surgery (Table 3). Zheng et al. [18] attempted to identify RCC using a cluster of nuclear-magnetic-resonance-based serum metabolite biomarkers. The authors started with using ANNs to a group and categorized serum metabolites as healthy or RCC and then estimated the detection of RCC in patients individually. Furthermore, ANNs were used for testing patients with RCC who had undergone nephrectomy. The expectation was that an individual patient who was previously classified as RCC would now be healthy after going through a nephrectomy. Haifler et al. [19] used shortwave Raman spectroscopy for distinguishing intra-operatively between healthy and malignant renal tissue. Training an AI model using Raman spectra from RCC and standard tissue samples could improve the identification of benign versus malignant tissue during surgery; the identification currently relies on a frozen section of the pathological specimen [19].


Sparse Multinomial Logistic Regression (SMLR).

#### *3.5. Hydronephrosis/Urinary Reflux*

Radiomic imaging technologies are used along with AI to diagnose clinically relevant hydronephrosis and/or urinary reflex. Blum et al. [20] used ML techniques to create a model that is capable of detecting hydronephrosis based on renogram features. The analysis successfully displayed a higher precision in detecting hydronephrosis when compared with just half-time and 30 min clearance. Cerrolaza et al. [21] used ultrasound features to develop ML methods that help in predicting renal obstruction (halftime > 30 min). Logvinenko et al. [22] used ultrasonography results to estimate vesicoureteric reflux (VUR) on the emptying after the cystourethrogram. They found that the AI model worked marginally better than the multivariate logistic regression.

#### *3.6. Reproductive Urology*

Statistics reveal that around 70 million couples globally are failing to conceive, and male infertility is held responsible for 50% of these cases. Various factors contribute to reproductive problems in men, such as genetic mutations, lifestyle choices, and medical illnesses. Considering such factors, many investigators have paired predictive analytics with AI techniques in their studies to demonstrate how AI could be of assistance in reproductive urology. In the studies by Gil et al. [23] and Candemir et al. [24], AI networks and algorithmic models were used to predict semen quality by considering variables such as lifestyle and environmental factors. Both studies displayed high accuracies, the first study showing an accuracy of ~86% for sperm concentration and 73–76% for motility and the second showing an accuracy of ~90%. These predictive models for semen quality could certainly be used as a tool for screening men with fertility issues to effectively expose any underlying seminal disorders. Among the men, 10–20% undergoing infertility evaluation are found to be suffering from azoospermia, a medical condition in men that causes impotency due to inadequate or no sperm production [25]. Akinsal et al. performed a retrospective study to predict the subset of azoospermic patients that should undergo additional genetic evaluation by applying logistic regression analyses and ANNs [26]. The model identified azoospermic patients with chromosomal abnormalities and those without chromosomal abnormalities with an accuracy of 95%. Exploiting AI to identify individuals with potential genetic abnormalities may mitigate the expense and time lag of formal genetic testing. Apart from predicting semen quality, AI has also been applied in various investigations to determine potential biomarkers for infecundity. In a study by Vickram et al., three different models of ANNs were employed to predict the biochemical parameters for male infertility, of which the backpropagation neural network (BNN) showed minimum error [27]. Men with infertility issues are asked to undergo semen analysis in which most of the parameters, such as sperm motility and concentration, are measured manually. To avoid these time-consuming procedures and the available expensive alternate procedures, Thirumalarjaju et al. introduced an AI-based approach using ANNs that was successful in producing the desired results in analyzing sperm morphology. The network identified abnormal semen samples with a staggering accuracy of 100% [28].

#### *3.7. Urolithiasis*

There has been a drastic alteration in the way urolithiasis cases are handled now compared with how they were handled in the past, and this approach will be highly influenced by AI techniques [29]. The future of AI in this field could provide complete management for urolithiasis: prevention, diagnosis, and treatment. Kazemi et al. [30] introduced a novel decision support system based on ensemble learning for the early detection (prevention) of kidney stones and explained the underlying mechanisms to determine the type of kidney stones. Various AI algorithms such as the Bayesian model, decision trees, ANNs, and rule-based classifiers were used in this system to understand the complex biological features involved in predicting kidney stones, with the system yielding an accuracy of 97.1%. Längkvist et al. [31] built a CNN (convolutional neural network) model for the detection of ureteral stones in high-resolution CT scans. This model was able to classify stones with a specificity of 100%, where the false positive was found to be 2.68 per scan and the AUC–ROC (receiver operating characteristic curve) was 0.9971.

#### *3.8. Pediatric Urology*

Pediatric urology handles congenital birth disabilities and disorders in newborn and young children. Though AI is yet to be wholly accepted and explored in this field, it certainly has brought new possibilities to light. About 1–3% of infants suffer from VUR, a condition that could potentially affect the bladder and kidneys if not diagnosed and treated earlier. One of the initial applications of AI in pediatric urology was the use of ANN architectures for the prognosis of VUR. To avoid a painful procedure for VUR detection, such as voiding cystourethrogram (VCUG), that exposes children to radiation, Papadopoulos et al. proposed an ML framework called Venn prediction for detecting VUR [32]. The model exhibited better sensitivity compared with other techniques. Likewise, another novel ML model was suggested to predict the future risk of febrile urinary tract infections (UTIs) related to VUR [33]. The predictive model performed with a reasonable degree of certainty in recognizing children most likely to benefit from VCUG, thus enabling personalized treatment.

#### *3.9. Endourological Procedures*

Endourology is another area in urology where AI is used to reach novel directions in planning and surgical interventions. Some of the previously mentioned minimally invasive procedures also come under this subfield. Images captured during cystoscopy play a pivotal role in the identification of bladder diseases. Ikeda et al. [17] introduced a support system based on CNNs for the proper diagnosis of bladder cancer using 2102 cystoscopic images. The built model separated the images of normal tissue from those of tumor lesions with high accuracy (area under ROC: 0.98; maximum Youden index (YI): 0.837; sensitivity: 89.7%; and specificity: 94%).

#### **4. Outcomes Prediction**

Patient outcome predictive analysis requires developing statistical methods that can interpret data to forecast outcomes for a particular patient. We can use either statistical modeling techniques or new methods emerging in the field of AI. These methods have the potential to handle the lack of accuracy and complexity that is typical in clinical and biological data. Additionally, AI techniques can handle the analysis of big data that are too big or too complex for standard statistical models more efficiently [34].

#### *4.1. Prostate Cancer*

Clinicopathological characteristics of individual patients are used to develop AI algorithms to forecast the outcome. Wong et al. [35] used clinicopathological characteristics of each patient to develop ML algorithms that can estimate the biochemical recurrence following prostatectomy (Table 4). They developed three different ML algorithms that were trained on a dataset of 338 patients to achieve an accuracy between 95% and 98% and an AUC between 0.9 and 0.94. In comparison to the conventional Cox regression analysis, these methods had better predictive efficiency. Tissue morphometric data [36], imaging radiomic features [37,38], and tissue genomic profiling [39,40] are also among the methods that are used for outcome forecasting of a patient. These studies have successfully demonstrated that AI has a higher accuracy when it comes to outcome prediction than other already existing methods.

Apart from the medical causes, surgical performance can also affect patient outcomes. Hung et al. [41,42] created and tested AI algorithms to find out the duration that a patient will have to stay in the hospital and the recovery of urinary control following robotic radical prostatectomy (Table 4). The algorithms were able to achieve an accuracy of 87.2% in the estimation of hospital stay and a C-index of 0.6 for estimating urinary control.



Regularized Extreme Learning Machine (RELM); MicroRNA (miRNA); Glycoprotein (GP).

#### *4.2. Urothelial Cancer*

Urothelial cancers have a high chance of recurrence. AI systems for forecasting cancer recurrence and patient survival have been engineered [43–46] (Table 5). Lam et al. [43] and Wang et al. [44] used clinicopathological evidence to create and test a significant number of AI algorithms to estimate the 5-year survival after radical cystectomy. Their work results obtained are equivalent to those obtained by other statistical methods. Sapre et al. [45] proposed using an ML classifier with urinary microRNA to diagnose bladder cancer in patients. The classification results by this research achieved an AUC between 0.8 and 0.9 in observing a clinically relevant disease, while also reducing the requirement for cystoscopy by 30%. Bartsch et al. [46] used gene expression profiling to develop AI strategies to forecast the recurrence of non-muscle-invasive bladder cancer. Such experiments have demonstrated the possibility of the potential uses of AI for the treatment of urothelial carcinoma.

**Table 5.** Studies using AI to predict outcomes of urothelial cancer.


Automated Performance Metrics (APM).

#### *4.3. Urolithiasis*

Percutaneous nephrolithotomy (PCNL) and shockwave lithotripsy (SWL) are commonly recognized therapeutic methods for urolithiasis; however, the rates of success may differ significantly and might include repeat procedures in case the treatment is unsuccessful. Aminsharifi et al. [47] used ANNs to forecast a stone-free PCNL rate with an accuracy of 82.8% and the need to repeat PCNL with an accuracy of 97.7%. Mannil et al. [48] focused their study on the individual patient, using the patient's body mass index (BMI), along with the 3D texture and scale of the stone, also accounting for the skin-to-stone distance to estimate the performance of SWL. The authors developed and tested five AI algorithms, each with varying 3D textural permutations of patient characteristics, to register AUC values between 0.79 and 0.85, which was an increment from the AUC score of 0.58 that was achieved when using only patient characteristics. For a different report, 3D texture analysis was used to estimate the number of shock waves needed for effective SWL [49]. Against other statistical models, AI displays the most accurate predictions of the number of shock waves needed (<72 or ≥72), with an AUC of 0.838 recorded. Both of Mannil et al.'s [48,49] experiments demonstrated that using AI along with advanced textural analysis methods is practical, reproducible, and predictive of SWL performance.

#### *4.4. Renal Transplant*

With renal transplantation (RT) being the best available therapy for end-stage renal failure (ESRF), some hindrances are faced in the procedure that can be dealt with by analyzing the survival of transplant patients. The availability of medical data and improving AI techniques have made this challenging prospect more achievable.

The current trend of AI in RT revolves around ensemble learning, where multiple models are combined to achieve better predictive performance. Ethan et al. [50] proposed an ensemble model of ML algorithms for the effective allocation of kidneys by using 18 different predictive variables. The survival model exhibited a higher index of concordance (0.724) than the other existing models (0.68) used for determining recipient priority in the allocation system. Recently, a risk prediction score named iBox has been developed by an international team of French researchers for forecasting the risk of allograft failure after RT [51]. This robust system outperforms the current golden standard (estimated glomerular filtration rate and proteinuria) to monitor kidney recipients. The forecasts of this method, validated on more than 7500 patients, are extremely accurate in decision making, independent of the healthcare environment, medical conditions, clinical action, or actual patient treatment.

Though RT is a better option over dialysis, the recipient's kidney is always at a risk of rejection, and hence early identification of such complications is necessary. Abdeltawab et al. [52] came up with a non-invasive method for the timely diagnosis of acute RT rejection. The authors developed a novel deep-learning-based computer-aided diagnostic system drawn upon both imaging and clinical biomarkers. With its sensitivity of 93.3% and 92.3% specificity in distinguishing between non-rejected and discharged renal transplants, the proposed method produced an accuracy of 92.9%. Using RT survivor statistics, Kyung et al. [53] conducted a retrospective study and built a predictive model to evaluate graft survival in RT receivers. Their survival decision tree model performed better compared to the conventional decision tree and Cox regression models, with indexes of concordance of 0.80, 0.71, and 0.60–0.63, respectively.

#### **5. Treatment Planning**

#### *5.1. Prostate Cancer Radiotherapy*

Brachytherapy for prostate cancer involves a systematic preparation by a brachytherapist, a time-consuming process that can have varied results, depending on the observer [54]. There has been a high degree of research involving the use of ML algorithms to rapidly build recovery schedules for brachytherapy [54,55]. The time required to create and test the algorithms was found to be much shorter (0.8 vs. 17.9 min; *p* = 0.002), while the dosimetry

metrics predicted were close to that of the qualified brachytherapist [55]. The accuracy of the dosimetry may be influenced because of different geometrical complexities during external radiotherapy. AI algorithms were developed by Guidi et al. [56] to handle such issues related to avoiding radiation injuries. CT images are used to train the AI algorithms in the radiotherapy planning and recovery phase of the treatment, which are used to compare scheduled and performed radiation therapy, helping patients who thereby benefit from receiving individualized care.

#### *5.2. Cancer Drug Selection*

AI interventions will be of assistance in the choice of adequate medications for cancer diagnosis and treatment. Saeed et al. [57] used ML technologies to measure and assess their activity with more than 300 forms of drugs in castration-resistant prostate cancer cells. Navitoclax family inhibitor Bcl-2 was described as highly active in patients with prostate cancer resistant to castration.

#### *5.3. Surgical Skill Assessment*

The evaluation of medical expertise and success is usually carried out by manual peer examination, allowing professionals to evaluate the surgical success or to monitor surgical performance. Such evaluations are often unreliable and increase the uncertainty due to different definitions of success by various observers. Endoscopic instruments offer direct visualization that is integrated with video cameras. These data, along with other types of information, including the movement of the surgical instruments, can also be collected. Such imagery and output data from the surgical robot can be used to test surgical output automatically using AI techniques. Figure 4 shows the procedural representation of a general biopsy using AI techniques. Anatomical landmark identification is an important metric in the assessment of advanced surgical skills. Nosrati et al. [58] and Baghdadi et al. [59] used ML algorithms to study the color and textural features from visualization of the surgical sites' anatomical features during partial nephrectomy and radical prostatectomy.

**Figure 4.** (**a**) Identification/Segmentation of the region of interest. (**b**) Classification of histopathological images using a deep learning technique.

Tracking the movements and actions of the surgical instruments is also an important metric for performance assessment. Ghani et al. [60] looked at the movements of instruments to determine surgical skills and techniques. The authors collected data on the movements of the instruments either manually or by using motion trackers, which were then fed to an ML algorithm to determine the expertise level of the surgeon, achieving a precision between 83.3% and 100% [60].

#### **6. Robotic Surgery**

Apart from assessing the surgical skill, as discussed in the previous section, AI also plays a key role in improving new surgical techniques such as minimally invasive procedures involving surgical robots. Determining the best practices by analyzing the patterns and aiding in reducing technical errors are the primary tasks of AI in robotic surgery. Its performance in each sub-specialty of urology is discussed below.

#### *6.1. Urologic Oncology*

Recent advances in robotic urologic surgery and minimally invasive procedures have enabled approaches to treating prostate cancer, such as laparoscopic prostatectomy and robotic-assisted surgery. Robotic prostate surgery is an extremely precise procedure that provides excellent cancer control and is considered safe in experienced hands.

Radical cystectomy has been the surgical standard to treat patients suffering from muscle-invasive bladder cancer. Though there is a significant reduction in the estimated blood loss (EBL), the blood transfusion rate, and the length of stay in robotic-assisted radical cystectomy (RARC) compared to those in open radical cystectomy (ORC), the complications and the positive margin status have been found to be similar [61–68]. Although the role of RARC is controversial, it has become an acceptable alternative to open surgery by some guideline organizations, including the European Association of Urology [62].

#### *6.2. Reproductive Urology*

Etafy et al. [69], in a study, validated that robot-assisted microsurgical procedures are now safe and practicable in dealing with male infertility. More than 500,000 American men opt for vasectomy as a method of contraception annually, of which 2–6% will eventually undergo vasectomy reversal [70]. Studies have shown that robot-assisted vasovasostomy (RAVV) yields comparable results to that of the pure microsurgical technique [71]. Though the former approach is not superior, it offers a few additional advantages over normal surgical procedures. These benefits include the elimination of tremors, multiview magnification, additional instrument arms, and enhanced dexterity with articulating instrument arms.

#### *6.3. Pediatric Urology*

In pediatrics, robotic surgery remains controversial due to both cost and the lack of published high-level evidence. Ballouhey et al. [72] discussed how size difference in children cannot be a limiting factor for performing robotic surgery (patients with body weight of >15 kg or <15 kg yielded similar results). Robot-assisted laparoscopic pyeloplasty (RALP) is the standard treatment of ureteropelvic junction obstruction in older children and has even been performed in infants and redo procedures. In a study by Avery et al. [73], among the 60-patient cohort with a mean age of 7.3 months, 91% showed improvement or resolution of hydronephrosis after pyeloplasty, with 11% facing post-operative complications and 2 patients requiring redo procedures. Redo robotic pyeloplasty is deemed a safe and effective approach for recurring ureteropelvic junction obstruction, reporting up to 100% success rates and 0% complication rates [74]. Along with RALP (Robot-assisted laparoscopic pyeloplasty), robot assistance in nephrectomy [75], ureteroureterostomy [76], ureteral reimplantation [77], and other procedures has yielded affirmative results and unlocked new possibilities in the field of pediatric urology.

#### *6.4. Renal Transplant*

Robot-assisted RT (RART) is another application of AI that is highly recommended for obese and high-risk ESRF patients as it delivers low complication rates and excellent graft function over conventional surgery [78]. RART is considered to be a safe, feasible, and reproducible option when performed by surgeons with practice in both robotic and conventional RT surgery.

#### **7. Discussion**

In this article, we explored how AI can help us in the diagnosis, outcome prediction, and other treatment processes of urological diseases, even when provided with a heterogeneous and complex dataset. The growth in the granularity of data due to the huge spike in data collection over the recent years makes interpretation and pattern identification difficult for traditional statistical models, which are restricted by the limitation of using fixed correlations that work on the assumption that the data will have linear relationships. AI is much more robust and flexible when it comes to working with different data types and dealing with noise, missing data, and infrequent visits by the patient. It can even handle high-dimensionality data, while making minimum assumptions.

Although using AI can be tricky, the results and accuracy achieved when it is used correctly exceed those observed with the standard statistical models. It can also help in simplifying manually performed procedures and thus reducing the variation in outcomes due to human ability, bias, and methodological mistakes or inefficiencies. Therefore, AIbased models help clinicians in getting early, reliable, and personalized data that can help in the decision making.

It is observed that AI achieves a higher accuracy for most tasks, but it cannot be used to answer every question. Sometimes, standard statistical models can outperform AI models. Kattan et al. [79] compared ML estimation and Cox proportional risk regression methods based on three separate datasets of urological results. Cox regression could correspond with or surpass the ML model predictions. Neural networks have freely used parameters for the transformation of feature and class prediction, the neural networks being accurate and adapted to the maximal values of these free parameters. A well-constructed conventional model can outperform an ML model built lousily. Another issue with using ML-based models is something called a black box. When we make a deep neural network, the model builds non-linear, non-monotonic response functions, which despite having remarkable accuracy might be harder to explain, which makes the performance of these networks more empirical than theoretical.

Several clinicians and researchers have discussed the role of AI in healthcare and in treating certain urological conditions [80,81]. The approach adopted in this review provides a comprehensive view with an aim to address all possible aspects of AI in the field of urology. The studies reviewed by us vary in their training features, algorithms used, and the observed endpoints, which makes the task of quantitative analysis more difficult. In addition, these studies lack generalizability across different datasets as we have the results only for that particular dataset. Some of them also do not give a comparison with the standard statistical models, which limits our ability to understand how AI techniques are better than other models.

Real-life usage of AI technologies in the field of medicine is still a long way into the future. They face high levels of quality control and regulatory obstacles. The US FDA (United States Food and Drug Administration) has issued the first AI system assessment guidelines [82], which show that adaptive architecture should provide real-life evidence in clinical studies to assess the efficacy of AI techniques. AI models are data driven; they learn from the data that are given to them, and therefore require continuous training to maximize their utility and accuracy.

#### **8. Conclusions**

AI has come a long way in making exponential progress in healthcare over the past decade. There are still a lot of challenges and hurdles that need to be addressed before these techniques can be completely trusted to be used in the medical field. Though the future of AI in the field of urology is bright, considering it has already provided excellent solutions to handle various health issues through early diagnosis and personalized treatment, there is still a lot of room for improvement and growth when it comes to delivering solid results to positively influence more number of lives on an individualized basis.

**Author Contributions:** Concept and Study design: B.M.Z.H., B.K.S., N.N. and B.P.R.; Methods and experimental work: S.Z.R., H.K., R.P. and H.S.K.; Results analysis and conclusions: B.M.Z.H., B.K.S., N.N., R.P. and B.P.R.; Manuscript preparation: S.D.A.V.L., S.I., M.J.S. and D.K.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare that they have no conflict of interest.

#### **References**


MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Journal of Clinical Medicine* Editorial Office E-mail: jcm@mdpi.com www.mdpi.com/journal/jcm

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18

www.mdpi.com

ISBN 978-3-0365-2709-3