Invited Commentary: Prostate Cancer Diagnosis—Challenges and Opportunities for Artificial Intelligence

HomeRadioGraphicsVol. 41, No. 6 PreviousNext Genitourinary ImagingFree AccessRadiology-Pathology CollectionInvited Commentary: Prostate Cancer Diagnosis—Challenges and Opportunities for Artificial IntelligenceAndrei S. Purysko Andrei S. Purysko Author AffiliationsFrom the Section of Abdominal Imaging and Nuclear Radiology Department, Imaging Institute and Glickman Urological and Kidney Institute, Cleveland Clinic, 9500 Euclid Ave, Mail Code JB-322, Cleveland, OH 44145.Address correspondence to the author (e-mail: [email protected]).Andrei S. Purysko Published Online:Oct 1 2021https://doi.org/10.1148/rg.2021210187MoreSectionsPDF ToolsImage ViewerAdd to favoritesCiteTrack CitationsPermissionsReprints ShareShare onFacebookTwitterLinked In See also the article by Alcalá Mata et al in this issue.IntroductionThe diagnosis and management of prostate cancer (PCa) have substantially evolved over the past 4 decades, and so has our understanding of the biology of this condition. The introduction of prostate-specific antigen (PSA) screening and the use of transrectal US (TRUS)–guided biopsy of the prostate have contributed to earlier diagnosis of PCa, but at the expense of costly and harmful overdetection and overtreatment of clinically insignificant cancers (1). In recent years, MRI and MRI-guided prostate biopsy have been incorporated into diagnostic pathways because of mounting evidence showing that these methods not only increase the detection of clinically significant cancers that are often missed by TRUS-guided biopsy, but can also mitigate the detection of clinically insignificant cancers (2). However, there are concerns about the actual benefit of MRI in clinical practice, primarily stemming from reported inconsistencies in image quality and interpretation (3). Concerns also exist regarding the interpretation of prostate biopsy specimens. Revisions of the original Gleason scoring system by the International Society of Urogenital Pathology (ISUP) have improved our ability to identify the histologic patterns of aggressive cancers, but pathologic assessment remains labor intensive, requires a high level of expertise, and is subject to interobserver variability (4).In their article in this issue, Alcalá Mata et al (5) deliver a timely and comprehensive review of the state of the art of artificial intelligence (AI) methods for PCa diagnosis and how this technology is poised to overcome many of the limitations of traditional and contemporary diagnostic pathways. The societal burden associated with PCa is substantial, and it is increasingly recognized that accurate, cost-effective, risk-adjusted strategies for PCa screening, biopsy triage, and management selection are urgently needed. The ability of AI methods to analyze a large volume of data is beneficial in this context, given the large number of existing biomarkers and the broad range of available management options. As described by Alcalá Mata and colleagues, the initial experience with AI methods in the form of natural language processing tools and machine learning (ML) techniques, including deep learning (DL), convolutional neural networks (CNNs), and radiomics, has shown encouraging results in many aspects of the diagnostic workup for PCa (6,7).With respect to MRI, AI methods have already been incorporated into clinical practice for a few important applications. Specifically, semi-automated prostate gland segmentation and coregistration of MRI and US images by using AI algorithms have become essential for gland volume calculation and software-assisted MRI-guided biopsies, respectively (8). Many other applications for AI methods listed by Alcalá Mata and colleagues are under active investigation. For instance, research has shown that for lesion characterization (ie, discriminating benign from malignant and significant from insignificant PCa on the basis of histologic features), the performance of ML models combined with radiomics is approaching the performance of humans (6,7). Research has also demonstrated similar results for lesion detection when using DL and CNN methods (6,7). For these applications, AI models have the potential to increase the radiologist’s efficiency by reducing interpretation time; they can also reduce interreader variability by improving the performance of less experienced readers. These improvements are critical, considering the growing demand for imaging in the diagnostic pathway. Similar benefits are also expected to be seen in the interpretation of pathologic images once digital pathologic assessment becomes increasingly adopted.These models do have important limitations that are noteworthy; most algorithms have been developed by using relatively small cohorts of patients that lack diversity in terms of cancer risk and prevalence and MR scanner technology used to generate the images (6). Before large-scale deployment into radiologic clinical workflows occurs, it is essential that these models are subjected to rigorous external validation and that their benefits are demonstrated in prospective studies. Furthermore, the development of user interfaces and mechanisms to monitor the functionality of these models will be needed to ensure patient safety.While the use and applications of MRI in the diagnostic pathway for PCa are expected to grow, particularly with AI models, other imaging technologies that are not addressed by Alcalá Mata and colleagues also deserve to be mentioned. Prostate-specific membrane antigen (PSMA)–targeted radiotracers for PET have shown promising results for PCa staging and detection of recurrent disease. The role of this method in the initial diagnosis of PCa is still uncertain, but the technique may prove useful in detecting PCa in patients at high risk who have had negative biopsy and MRI results (9). Microultrasound is another novel technology that has piqued interest in recent years, with initial studies showing that the performance of this method for detecting clinically significant PCa compares favorably with that of MRI; additionally, microultrasound has a much lower cost than MRI and can be performed in the urologist’s office (10).There is little doubt that AI will eventually facilitate the integration of clinical, radiologic, and pathologic data, as described by Alcalá Mata and colleagues, and that this integration will enable a personalized PCa diagnosis in the near future. I commend the authors for their great contribution to the PCa literature.Disclosures of Conflicts of Interest.— A.S.P. Activities related to the present article: disclosed no relevant relationships. Activities not related to the present article: payment to institution for a service contract with Profound Medical; research in support of Blue Earth Diagnostics; consulting agreement with Koelis; lectured for the American College of Radiology; holds a patent. Other activities: disclosed no relevant relationships.AcknowledgmentI am grateful for the editorial assistance of Megan M. Griffiths, ELS, scientific writer for the Imaging Institute, Cleveland Clinic, Cleveland, Ohio.The author has disclosed no relevant relationships.References1. Ilic D, Djulbegovic M, Jung JH, et al. Prostate cancer screening with prostate-specific antigen (PSA) test: a systematic review and meta-analysis. BMJ 2018;362k3519. Crossref, Medline, Google Scholar2. Drost FH, Osses DF, Nieboer D, et al. Prostate MRI, with or without MRI-targeted biopsy, and systematic biopsy for detecting prostate cancer. Cochrane Database Syst Rev 2019;4(4):CD012663. Medline, Google Scholar3. de Rooij M, Israël B, Tummers M, et al. ESUR/ESUI consensus statements on multi-parametric MRI for the detection of clinically significant prostate cancer: quality requirements for image acquisition, interpretation and radiologists’ training. Eur Radiol 2020;30(10):5404–5416. Crossref, Medline, Google Scholar4. Van Booven DJ, Kuchakulla M, Pai R, et al. A Systematic Review of Artificial Intelligence in Prostate Cancer. Res Rep Urol 2021;13(31):39. Google Scholar5. Alcalá Mata L, Retamero Díaz JA, Gupta RT, Ramos Guillén C, García Figueras R, Luna A. Artificial intelligence–assisted diagnosis: radiologic-pathologic correlation. RadioGraphics 2021;41(6):1676–1697. Google Scholar6. Cuocolo R, Cipullo MB, Stanzione A, et al. Machine learning for the identification of clinically significant prostate cancer on MRI: a meta-analysis. Eur Radiol 2020;30(12):6877–6887. Crossref, Medline, Google Scholar7. Twilt JJ, van Leeuwen KG, Huisman HJ, Fütterer JJ, de Rooij M. Artificial Intelligence Based Algorithms for Prostate Cancer Classification and Detection on Magnetic Resonance Imaging: A Narrative Review. Diagnostics (Basel) 2021;11(6):959. Crossref, Medline, Google Scholar8. Kongnyuy M, George AK, Rastinehad AR, Pinto PA. Magnetic Resonance Imaging-Ultrasound Fusion-Guided Prostate Biopsy: Review of Technology, Techniques, and Outcomes. Curr Urol Rep 2016;17(4):32. Crossref, Medline, Google Scholar9. Bagguley D, Ong S, Buteau JP, et al. Role of PSMA PET/CT imaging in the diagnosis, staging and restaging of prostate cancer. Future Oncol 2021;17(17):2225–2241. Crossref, Medline, Google Scholar10. Sountoulides P, Pyrgidis N, Polyzos SA, et al. Micro-Ultrasound-Guided vs Multiparametric Magnetic Resonance Imaging-Targeted Biopsy in the Detection of Prostate Cancer: A Systematic Review and Meta-Analysis. J Urol 2021;205(5):1254–1262. Crossref, Medline, Google ScholarArticle HistoryReceived: June 24 2021Accepted: June 29 2021Published online: Oct 01 2021Published in print: Oct 2021 FiguresReferencesRelatedDetailsCited ByMachine Learning Algorithm Accuracy Using Single- versus Multi-Institutional Image Data in the Classification of Prostate MRI LesionsDestieProvenzano, OleksiyMelnyk, DanishImtiaz, BenjaminMcSweeney, DanielNemirovsky, MichaelWynne, MichaelWhalen, Yuan JamesRao, MurrayLoew, ShawnHaji-Momenian2023 | Applied Sciences, Vol. 13, No. 2Artificial intelligence in prostate cancer: Definitions, current research, and future directionsRose S.George, ArkarHtoo, MichaelCheng, Timothy M.Masterson, KunHuang, NabilAdra, Hristos Z.Kaimakliotis, MahmutAkgul, LiangCheng2022 | Urologic Oncology: Seminars and Original Investigations, Vol. 40, No. 6Accompanying This ArticleArtificial Intelligence–assisted Prostate Cancer Diagnosis: Radiologic-Pathologic CorrelationOct 1 2021RadioGraphicsRecommended Articles PI-RADS Steering Committee: The PI-RADS Multiparametric MRI and MRI-directed Biopsy PathwayRadiology2019Volume: 292Issue: 2pp. 464-474The Current State of MR Imaging–targeted Biopsy Techniques for Detection of Prostate CancerRadiology2017Volume: 285Issue: 2pp. 343-356PI-RADS: Where Next?Radiology2023Volume: 307Issue: 5Abbreviated Biparametric Prostate MR Imaging in Men with Elevated Prostate-specific AntigenRadiology2017Volume: 285Issue: 2pp. 493-505Prostate Cancer: PI-RADS Version 2 Helps Preoperatively Predict Clinically Significant CancersRadiology2016Volume: 280Issue: 1pp. 108-116See More RSNA Education Exhibits Prostate Cancer: What We Already Know About What Is on the HorizonDigital Posters2020MRI: Transrectal Ultrasound Fusion-Guided Biopsy of the ProstateDigital Posters2019Magnetic Resonance Imaging of the Prostate Before and After Focal TherapyDigital Posters2020 RSNA Case Collection Locally advanced, metastatic prostate adenocarcinomaRSNA Case Collection2020Peripheral Zone Prostate Cancer with PseudocapsuleRSNA Case Collection2022Anterior transition zone prostate adenocarcinomaRSNA Case Collection2020 Vol. 41, No. 6 Abbreviations Abbreviations: AI artificial intelligence PCa prostate cancer Metrics Altmetric Score PDF download