Editor’s Note: Artificial Intelligence (AI) is transforming the diagnostic and treatment approaches for urological tumors, especially with significant advancements in intelligent imaging and pathology, potentially becoming an indispensable tool for clinicians. Dr. Vasileios Sakalis, a urologist from the Hippokrateion General Hospital of Thessaloniki, Greece, presented a comprehensive review on the application of AI in urological tumors at the conference. During an on-site interview with “Oncology Frontier-Urology Frontier,” he shared further insights on the applications of AI.
Research Overview:
Role of Artificial Intelligence in Urological Tumors: A Systematic Review Overview (Abstract No.: A0850)
Background and Objectives:
Artificial Intelligence (AI) has great potential to enhance patient care. AI algorithms extract quantitative features from medical images and analyze large datasets aiming to improve diagnostic accuracy and treatment outcomes in various medical conditions. This study summarizes the role of AI in diagnosing and treating urological tumors.
Materials and Methods:
A systematic search was conducted up to July 2023 in databases including Embase, Medline, CDSR, and clinical trial registries. Included were systematic reviews (SRs), reviews, diagnostic, and interventional studies regarding the application of AI in malignant urological tumors; following Cochrane review methodologies; utilizing the ROBIS tool to assess method quality and bias risk (RoB). Results data are presented as ranges of AUC-ROC, Accuracy (Acc), Sensitivity (Sen), and Specificity (Spe).
Results:
75 studies were screened in full, with 52 included in the qualitative analysis. Risk of Bias (RoB) was considered low in 8 SRs, high in 33, and unclear in 11.
32 SRs evaluated AI as a diagnostic or prognostic tool in prostate cancer (PCa) using CT, MRI, and PET scans. AI models effectively differentiated between benign and malignant cases (AUC: 0.55-0.99, Acc: 88%-93%, Sen: 59%-95%, Spe: 58%-87%). AI outperformed human readings in distinguishing indolent PCa from clinically significant PCa (csPCa) in predefined MRI lesions (AUC: 0.68-0.95 vs AUC: 0.74-0.88), and in identifying csPCa in PIRAD 3 lesions (AUC: 0.78-0.96 vs AUC: 0.81-0.87). AI accurately predicted Gleason scores (AUC: 0.5-0.92, Acc: 77%-87%), biochemical recurrence (AUC: 0.68-0.88, Acc: 62%-91%), and lymph node involvement (AUC: 0.8-0.92, Acc: 70%-90%). Three SRs assessed AI models in recognizing pathological grades in biopsy specimens (AUC: 0.81-0.87, Acc: 88%-91%, Sen: 87%-100%), with consistency between AI and pathologists (κ coefficient: 0.3-0.83).
13 SRs evaluated AI models in diagnosing renal cancer (RCa). Using MRI and/or CT for distinguishing benign and malignant lesions showed a combined AUC of 0.52-0.97, with an accuracy of 70%-95%. AUC for predicting renal cancer subtype and nuclear grade was 0.78-0.04, with an accuracy of 73%-88%. Compared to radiologists alone, radiologists assisted by AI showed higher accuracy (77%-94% vs 69%-80%) and lower inter-observer variability.
7 SRs assessed AI in predicting recurrence (AUC: 0.7-0.92), progression (AUC: 0.8-0.97), complications post-cystectomy (AUC: 0.52-0.82), and overall survival (Acc: 85%-97%) in bladder cancer. AI showed superior diagnostic performance in predicting muscle-invasive tumors compared to radiologists alone (AUC 0.979 vs 0.856, Acc: 96.3% vs 90.1%).
Conclusions:
Artificial Intelligence enhances diagnostic accuracy, treatment outcomes, and patient prognosis in urology. With the ongoing development of AI, the management of urological tumors will become more precise and patient-centered.
Oncology Frontier-Urology Frontier: How accurate do you think current AI is in the field of urological tumors?
Dr. Vasileios Sakalis: Artificial intelligence has achieved significant accuracy levels, especially in using imaging
modalities like MRI to extract features. AI models are particularly precise and sensitive in identifying malignancies. However, accuracy slightly decreases when dealing with more complex data extraction, such as prognostic models or predicting Gleason scores.
Oncology Frontier-Urology Frontier: Can AI currently be used in clinical practice to assist doctors in diagnosis?
Dr. Vasileios Sakalis: AI’s ability to interpret quantitative imaging properties has been integrated into clinical practice, marking a significant advancement. Through machine learning and deep learning techniques, AI can uncover image features invisible to the human eye, merging them with clinical data to produce highly accurate diagnostic information.
Oncology Frontier-Urology Frontier: In terms of therapeutic areas, where can AI be used in urological oncology?
Dr. Vasileios Sakalis: AI models are actively utilized and assessed across all urological cancers, particularly in prostate, kidney, and bladder cancers, which are more common. While AI models for rarer cancers are under development, the most compelling evidence of AI’s effectiveness is in prostate cancer research, as evidenced by numerous systematic reviews.
Oncology Frontier-Urology Frontier: How can AI better predict patient outcomes to continuously adjust treatment plans and improve long-term survival?
Dr. Vasileios Sakalis: The European Association of Urology is exploring the potential of AI to create personalized treatment prediction models. Leveraging big data and AI can generate specific outcomes and evidence tailored to individual patients or patient groups, aiming to refine treatment approaches based on precise predictive insights.
Oncology Frontier-Urology Frontier: What are the most interesting research developments announced at this conference?
Dr. Vasileios Sakalis: Beyond my strong interest in AI and big data analysis for urological conditions, I am particularly focused on research into male lower urinary tract symptoms and neurogenic bladder. I am actively involved with the European Association of Urology, especially in the male LUTS panel, where my main interest lies in functional urology.
