AI in Prostate Cancer Screening & Diagnosis: A Registry-Based Study of ClinicalTrials.gov Trials

Abstract

Introduction

Artificial intelligence (AI) is increasingly applied in prostate cancer screening and diagnostic evaluation; however, the structure, methodological characteristics, and clinical positioning of AI-focused trials remain incompletely characterized. This study aimed to map the clinical trial landscape of AI applications in prostate cancer diagnosis using registry-based evidence mapping.

Methods

A registry-based evidence-mapping analysis was conducted using ClinicalTrials.gov. Trials registered up to 15 November 2025 were systematically identified using search terms related to prostate cancer and AI-based methodologies. Eligible studies included interventional and observational trials evaluating AI applications for diagnostic purposes. Data were extracted on study design, diagnostic modality, functional role of AI, comparator framework, and validation strategy. Descriptive statistics and cross-tabulation analyses were used to characterize patterns across studies. The study selection process was presented using a PRISMA-style flow diagram.

Results

A total of 84 trials met the inclusion criteria. Imaging-based AI applications predominated, accounting for 52.4% of studies, with magnetic resonance imaging (MRI) representing the most frequently investigated modality (34.5%). Biomarker-based (16.7%), multimodal (15.5%), and computational pathology (7.1%) approaches were less frequently reported. The most common functional applications were classification and risk prediction (48.8%) and lesion detection and segmentation (29.8%). Most studies employed prospective observational designs (84.5%) and frequently relied on stand-alone AI evaluation frameworks (39.2%). Histopathology or biopsy confirmation was the most commonly reported reference standard (56.0%). Only a limited number of trials incorporated workflow integration or clinical decision-support evaluation.

Conclusion

AI research in prostate cancer diagnostics appears to be primarily centered on imaging-based, early-phase, and performance-oriented studies. Current evidence suggests that AI systems are predominantly positioned as decision-support tools rather than fully integrated clinical solutions. Greater emphasis on multicenter validation, standardized reporting, and clinically relevant outcome evaluation may be required to support broader clinical implementation.

1. Introduction

Prostate cancer remains one of the most frequently diagnosed malignancies among men worldwide and continues to represent a major contributor to cancer-related morbidity and mortality.^1,2 The disease is characterized by marked biological heterogeneity, which complicates risk stratification and prognostic assessment at the time of diagnosis. Conventional diagnostic approaches, including serum prostate-specific antigen (PSA) testing and systematic transrectal ultrasound–guided biopsy, have long served as the primary tools for detection; however, both methods present important limitations. PSA testing has relatively low specificity and may produce a substantial number of false-positive results, leading to unnecessary biopsies and patient anxiety.^1,3 At the same time, systematic biopsy strategies are subject to sampling error and may fail to detect clinically significant tumors, while also identifying indolent lesions that may never become clinically relevant.^2,4 These diagnostic limitations contribute to persistent concerns regarding overdiagnosis and overtreatment of prostate cancer. Multiparametric magnetic resonance imaging (mpMRI) has improved lesion detection and localization and has become an important component of contemporary diagnostic pathways; however, interpretation of mpMRI remains highly operator dependent and is associated with substantial interobserver variability across readers and institutions.^1,5 These challenges highlight the need for more objective and reproducible diagnostic systems capable of improving consistency in prostate cancer detection and risk stratification.

Artificial intelligence (AI) has emerged as a promising approach for addressing several of these diagnostic challenges. Machine learning and deep learning techniques can analyze complex imaging, pathological, and clinical datasets to identify patterns associated with clinically significant prostate cancer. In imaging applications, AI systems have been developed to assist with lesion detection, automated segmentation, and risk prediction using multiparametric MRI datasets.^6,7 Similarly, digital pathology platforms employing deep learning algorithms have demonstrated the ability to detect cancerous regions on whole-slide biopsy images and support Gleason grading with performance approaching that of expert pathologists.⁸ In addition to imaging and pathology, AI-based models integrating biomarker and clinical data have been proposed to enhance risk prediction and assist clinical decision making. Collectively, these developments suggest that AI systems may function as supportive tools that complement specialist interpretation and potentially improve diagnostic efficiency and reproducibility.^7,9

Beyond individual applications, the integration of AI into healthcare has increasingly been conceptualized within broader digital health ecosystems, where data-driven technologies interact with clinical workflows, infrastructure, and decision-making processes.¹⁰ Within this context, AI-based diagnostic systems are evaluated not only for technical performance but also for their potential role in shaping diagnostic pathways and healthcare delivery. These developments reflect a broader transformation in prostate cancer care, where artificial intelligence is increasingly applied across imaging, pathology, and clinical decision-making to improve diagnostic accuracy and healthcare efficiency.^11,12

Despite the rapid expansion of AI-based diagnostic tools, several methodological challenges remain in evaluating these systems within clinical research. Many AI models are developed using retrospective datasets and are evaluated using internally derived test cohorts, raising concerns about limited external validation and generalizability across diverse clinical populations.^13,14 Dataset bias, demographic imbalance, and methodological heterogeneity further complicate the interpretation of reported performance metrics.^15,16 Variability in study design, outcome definitions, and validation strategies across studies also contributes to difficulties in comparing results and synthesizing evidence across the literature. These methodological limitations have prompted calls for more rigorous evaluation frameworks and prospective validation studies to support the clinical translation of AI diagnostic technologies.

In response to these concerns, several reporting and methodological frameworks have been proposed to improve transparency and reproducibility in AI-based medical research. Extensions of established clinical trial and prediction model reporting guidelines—including CONSORT-AI, SPIRIT-AI, TRIPOD-AI, and PROBAST-AI—have been developed to ensure clearer documentation of data sources, model development procedures, validation strategies, and human–AI interaction within clinical workflows.^17,18 Additional tools such as the Checklist for Artificial Intelligence in Medical Imaging (CLAIM) and QUADAS-AI aim to standardize methodological assessment in diagnostic accuracy studies involving AI technologies.^9,19 Although these frameworks represent important steps toward improving research quality, their adoption across AI diagnostic studies remains variable.

While a growing number of studies have evaluated AI models for prostate cancer detection and risk stratification, most investigations have focused on technical performance within imaging or pathology datasets rather than examining the broader clinical research landscape in which these technologies are evaluated. Registry-based analyses of clinical trial databases, such as ClinicalTrials.gov, have been widely used to characterize trends in medical research and to map emerging innovation landscapes across therapeutic and diagnostic domains.^20,21 However, the structure and methodological characteristics of registered clinical trials investigating artificial intelligence applications in prostate cancer diagnostics have not been systematically examined.

The present study therefore analyzes clinical trials registered on ClinicalTrials.gov to characterize the evolving landscape of artificial intelligence applications in prostate cancer diagnostics. Specifically, the study examines the diagnostic modalities in which AI systems are being investigated, the functional roles these systems perform within diagnostic workflows, and the methodological strategies used to evaluate them, including study design, comparator frameworks, and validation approaches. By providing a registry-based overview of ongoing and completed trials, this analysis aims to clarify how AI technologies are currently being studied in prostate cancer diagnostics and to identify methodological patterns that may inform future clinical evaluation and implementation of AI-assisted diagnostic systems. This study addresses this gap by providing a structured, registry-based characterization of how AI diagnostic systems are currently being developed, evaluated, and positioned within clinical research settings.

2. Methods

2.1. Study Design

This study was conducted as a registry-based evidence-mapping analysis to characterize the clinical research landscape of artificial intelligence (AI) applications in prostate cancer screening and diagnosis. Evidence mapping was used to systematically identify, categorize, and describe patterns in diagnostic modalities, functional roles of AI systems, and methodological approaches across registered clinical trials. The objective was to provide a structured overview of research activity and methodological variation rather than to perform quantitative synthesis or effect size estimation. In line with STROBE reporting considerations for cross-sectional studies, study identification and selection were conducted using a structured process and are presented in a PRISMA-style flow diagram to enhance transparency and reproducibility.

2.2. Data Source and Search Strategy

Clinical trial records were identified using the publicly accessible ClinicalTrials.gov registry. The search was performed on 15 November 2025, and the dataset was restricted to records available up to this date. The search strategy was designed to capture studies at the intersection of prostate cancer, artificial intelligence–related methodologies, and diagnostic applications. These concepts were implemented within the registry’s “Condition” and “Other Terms” fields using combinations of keywords related to prostate cancer and AI-based techniques.

The search incorporated terms such as “prostate cancer” alongside expressions including “artificial intelligence,” “machine learning,” “deep learning,” “radiomics,” and “algorithm.” The ClinicalTrials.gov search interface automatically expands queries to include indexed synonyms and related terms, allowing broader retrieval of relevant records. No restrictions were applied regarding study phase, recruitment status, geographic location, or language.

2.3. Eligibility Criteria

Trials were considered eligible if they investigated AI-based approaches applied to prostate cancer screening or diagnostic evaluation. This included studies involving imaging interpretation, biomarker analysis, or computational models designed for diagnostic classification or risk prediction. Both interventional and observational studies were included, provided that they involved human participants or human-derived clinical data.

Studies were excluded if they focused exclusively on therapeutic applications, such as treatment optimization, radiotherapy planning, or drug response prediction, or if they addressed non-oncologic urological conditions. Trials that did not provide sufficient methodological detail to determine the role of AI in diagnostic evaluation were also excluded.

2.4. Study Selection

All retrieved records were manually screened to assess eligibility. During the initial screening stage, the study title, brief summary, intervention description, and outcome measures were reviewed to determine whether the record addressed artificial intelligence (AI) applications relevant to prostate cancer screening or diagnostic evaluation. Studies were considered potentially eligible if the registry description indicated the use of artificial intelligence, machine learning, deep learning, radiomics, or related computational approaches applied to imaging, pathology, biomarker, molecular, or clinical data for diagnostic purposes. Records were excluded if the primary focus was therapeutic intervention, treatment planning, supportive care, patient communication, education, monitoring, or other non-diagnostic applications. Trials lacking sufficient methodological detail to determine the role of AI in diagnostic evaluation were also excluded.

2.5. Data Extraction

Data were extracted manually from each eligible trial using a structured coding approach based on registry-reported fields. Reviewed fields included the NCT identifier, study title, study status, brief summary, conditions, interventions, primary and secondary outcome measures, sponsor, study type, study design, enrollment, and study dates. These variables were used to assess trial relevance and to characterize the diagnostic modality, functional role of AI, comparator framework, and validation strategy described in the registry entry. Each record was reviewed in detail to ensure that extracted information accurately reflected the trial description provided in ClinicalTrials.gov.

2.6. Classification Framework

A structured classification framework was developed to organize the included trials according to the analytical objectives of the study. Diagnostic modalities were classified according to the dominant data source used by the AI system, including magnetic resonance imaging, positron emission tomography, ultrasound, computed tomography, computational pathology, biomarker or molecular data, multimodal systems, and clinical data–driven models. Where the primary modality was not immediately apparent, the study objective, intervention description, and outcome measures were used to determine the most appropriate classification.

Functional roles of AI systems were categorized as classification and risk prediction, lesion detection and segmentation, algorithm development and validation, workflow and image optimization, or clinical decision support. Methodological characteristics were further categorized according to study design, comparator framework, and reference standard used for evaluation. Each study was assigned to one primary category within each analytical domain based on the main objective and outcome description reported in the registry entry. This framework was applied consistently across all included studies.

2.7. Statistical Analysis

Extracted data were analyzed using descriptive statistical methods to summarize the characteristics of the included trials and to identify structural patterns across the dataset. Frequencies and proportions were calculated for all categorical variables. Cross-tabulation analyses were performed to examine relationships between diagnostic modalities and functional roles of AI, study design and comparator frameworks, and diagnostic modalities and validation reference standards. No inferential statistical analyses were performed because the study was designed to provide a descriptive and structural characterization of the registered research landscape.

3. Results

3.1. Trial Identification and Selection

The process of study identification and selection is illustrated in Figure 1. An initial search of the ClinicalTrials.gov registry for prostate cancer–related clinical trials yielded 7,278 records. After applying artificial intelligence–related diagnostic keywords, 123 trials were retrieved for further screening. These records were manually reviewed to assess their relevance to artificial intelligence–based diagnostic applications in prostate cancer. Following eligibility assessment, 84 trials met the inclusion criteria and were included in the final analysis.

Figure 1.

PRISMA 2020 flow diagram of study identification, screening, and inclusion

3.2. AI Application Domains and Diagnostic Modalities

The included trials encompassed multiple diagnostic platforms in which artificial intelligence (AI) was applied to prostate cancer detection and characterization. Imaging-based approaches represented the largest application domain, accounting for 45 of the included studies (53.5%). Within this category, magnetic resonance imaging (MRI)–based models were reported in 29 studies (34.5%), followed by positron emission tomography (PET)–based applications in 8 studies (9.5%), ultrasound-based systems in 6 studies (7.1%), and computed tomography (CT)–based approaches in 2 studies.

In addition to imaging-based approaches, biomarker and molecular diagnostic systems accounted for 14 studies (16.7%), where AI models were applied to biological data sources including blood, urine, genomic, or molecular biomarkers. Computational pathology applications were identified in 6 studies (7.1%), involving the analysis of histopathological images or biopsy-derived tissue samples. Multimodal AI systems combining multiple sources of clinical data were reported in 13 studies (15.5%), integrating imaging, molecular, or histopathological information within unified analytical frameworks. In addition, 7 studies (8.3%) utilized AI models derived primarily from clinical or electronic health record data. The distribution of AI application domains and specific modalities across the included studies is summarized in Table 1.

Table 1.

Artificial Intelligence Application Domains and Modalities in Included Prostate Cancer Trials (n = 84)

AI application domain	Specific Modality	Number of Trials
Imaging-Based AI	MRI	29
	PET/PET-CT/PET-SPECT	8
	Ultrasound	6
	CT/CBCT	2
Computational Pathology AI	Histopathology/Digital Pathology	6
Biomarker and Molecular AI	Blood, urine, genomic, or molecular biomarkers	14
Multimodal AI Systems	Combined imaging, biomarker, or clinical data	13
Clinical Data–Driven AI	Electronic health record or clinical prediction models	6
Total		84

Percentages are calculated from the total number of included studies (n = 84). Abbreviations: MRI = magnetic resonance imaging; PET = positron emission tomography; PET-CT = positron emission tomography–computed tomography; SPECT = single-photon emission computed tomography; CT = computed tomography; CBCT = cone-beam computed tomography.

3.3. Functional Classification of AI Applications

The functional classification of artificial intelligence (AI) applications across the included trials is summarized in Table 2. The most common application domain was classification and risk prediction, representing 41 studies (48.8%), where AI models were developed to estimate cancer probability, predict tumor aggressiveness, or stratify patient risk using imaging, biomarker, or clinical data. The second most frequent category was lesion detection and segmentation, comprising 25 studies (29.8%), in which AI systems were used to identify or delineate suspicious prostate lesions on imaging modalities such as magnetic resonance imaging (MRI), ultrasound, or positron emission tomography (PET).

Table 2.

Functional Classification of Artificial Intelligence (AI) Applications in Included Prostate Cancer Trials (n = 84)

AI application theme	Description	n (%)
Classification and Risk Prediction	Machine-learning systems designed to classify disease presence, predict malignancy risk, estimate tumor aggressiveness, or evaluate clinical outcomes using imaging, biomarker, or clinical features.	41 (48.8)
Lesion Detection and Segmentation	AI models developed to identify, localize, or segment prostate lesions on imaging modalities such as MRI, ultrasound, or PET.	25 (29.8)
Algorithm Development and Validation	Studies focused on the development, training, or external validation of AI models, including benchmarking across datasets or institutions.	8 (9.5)
Workflow and Image Optimization	Applications designed to improve image reconstruction, enhance image quality, automate segmentation pipelines, or streamline diagnostic workflows.	6 (7.1)
Clinical Decision Support Systems	AI tools integrating diagnostic outputs with clinical parameters to assist clinical interpretation, biopsy targeting, or management decisions.	4 (4.8)

Percentages are calculated from the total number of included studies (n = 84). Abbreviations: AI = artificial intelligence; MRI = magnetic resonance imaging; PET = positron emission tomography.

A further 8 studies (9.5%) focused primarily on algorithm development and validation, including model training, external validation, and benchmarking across datasets or institutions. Workflow and image optimization applications were reported in 6 studies (7.1%), addressing technical improvements such as image reconstruction, segmentation automation, or diagnostic workflow support. Clinical decision-support systems accounted for 4 studies (4.8%), where AI outputs were integrated with clinical parameters to assist interpretation or management decisions.

3.4. Methodological Characteristics and Evaluation Strategies

The methodological characteristics of the included trials are summarized in Table 3. Prospective observational designs represented the majority of the dataset, accounting for 71 studies (84.5%), while prospective interventional studies were identified in 7 trials (8.3%). Retrospective analyses were reported in 4 studies (4.8%), and one randomized controlled trial (1.2%) was identified. An additional observational study with unspecified design accounted for one trial (1.2%). These findings indicate that most studies were structured around prospectively collected data to evaluate artificial intelligence (AI) applications in clinical or diagnostic settings.

Table 3.

Methodological Characteristics and Evaluation Strategies of Included AI-Based Prostate Cancer Trials (n = 84)

Characteristic	Category	n (%)
Study design	Prospective observational	71 (84.5)
	Prospective interventional	7 (8.3)
	Retrospective	4 (4.8)
	Other/unspecified	2 (2.4)
Comparator type	Stand-alone AI evaluation or algorithm development	33 (39.2)
	AI compared with clinician interpretation	11 (13.1)
	AI compared with diagnostic tests or workflows	11 (13.1)
	Other comparative strategies	29 (34.5)
Reference standard for validation	Histopathology/biopsy confirmation	47 (56.0)
	Imaging-based reference	10 (11.9)
	Clinical outcomes or follow-up	8 (9.5)
	Composite or unspecified reference	19 (22.61)
Study status	Recruiting/ongoing	36 (42.9)
	Completed	21 (25.0)
	Not yet recruiting	9 (10.7)
	Terminated/withdrawn/suspended/Unknown	18 (21.4)

Percentages are calculated from the total number of studies included (n = 84). Reference standards were categorized according to the primary validation method described in the registry record. Abbreviations: AI = artificial intelligence.

Regarding evaluation strategies, stand-alone AI evaluation or algorithm development frameworks were the most frequently reported comparator approach, occurring in 33 studies (39.2%). Comparisons between AI systems and clinician interpretation were reported in 11 studies (13.1%), as well as compared AI approaches with existing diagnostic tests or clinical workflows studies. The remaining 29 studies (34.5%) employed a variety of alternative comparative or feasibility-oriented evaluation strategies.

With respect to validation methods, histopathology or biopsy confirmation served as the most common reference standard, reported in 47 studies (56.0%). Imaging-based references were used in 10 studies (11.9%), while clinical outcomes or follow-up endpoints were applied in 8 studies (9.5%). Validation based on technical or feasibility metrics or had no clear reference standard were reported in 19 studies (22.6%).

The study status of the included trials further reflected the evolving nature of AI diagnostic research. Recruiting or ongoing studies accounted for 36 trials (42.9%), while 21 trials (25.0%) had been completed. A further 9 studies (10.7%) were not yet recruiting, 12 trials (14.3%) had an unknown status, and 6 studies (7.1%) had been terminated, withdrawn, or suspended.

3.5. Structural Relationships in AI Diagnostic Trials

To further characterize the methodological and technological landscape of the included studies, additional cross-analyses were conducted examining relationships between key trial characteristics. These analyses explored the distribution of artificial intelligence (AI) functional themes across diagnostic modalities, the relationship between study design and comparator frameworks, and the validation reference standards used across different diagnostic approaches. The results of these structural analyses are presented in Tables 4-6.

Table 4.

Relationship Between Diagnostic Modality and AI Functional Theme in Included Prostate Cancer Trials (n = 84)

Specific modality	Classification and risk prediction	Lesion detection and segmentation	Algorithm development and validation	Workflow and image optimization	Clinical decision support systems	Total
MRI	9	12	3	4	1	29
PET/PET-CT/PET-SPECT	3	3	1	1	0	8
Ultrasound	1	4	1	0	0	6
CT/CBCT	1	1	0	0	0	2
Histopathology/Digital Pathology	3	2	1	0	0	6
Blood, urine, genomic, or molecular biomarkers	13	0	1	0	0	14
Combined imaging, biomarker, or clinical data	11	2	0	0	0	13
Electronic health record or clinical prediction models	0	1	1	1	3	6
Total	41	25	8	6	4	84

Abbreviations: MRI = magnetic resonance imaging; PET = positron emission tomography; PET-CT = positron emission tomography–computed tomography; SPECT = single-photon emission computed tomography; CT = computed tomography; CBCT = cone-beam computed tomography; EHR = electronic health record.

Table 5.

Relationship Between Study Design and Comparator Framework in Included AI-Based Prostate Cancer Trials (n = 84)

Study design	Stand-alone AI evaluation/development	AI vs clinician interpretation	AI vs diagnostic test/workflow	Other comparative strategy	Total
Prospective observational	32	8	10	21	71
Prospective interventional	0	1	0	6	7
Retrospective	1	2	0	1	4
Other/unspecified	0	0	1	1	2
Total	33	11	11	29	84

Abbreviations: AI = artificial intelligence.

Table 6.

Relationship Between Diagnostic Modality and Reference Standards Used for AI Validation in Included Prostate Cancer Trials (n = 84)

Diagnostic modality	Histopathology/biopsy	Imaging-based reference	Clinical outcomes/follow-up	Composite or unspecified reference	Total
MRI	16	9	1	3	29
PET/PET-CT/PET-SPECT	3	1	3	1	8
Ultrasound	6	0	0	0	6
CT/CBCT	0	0	1	1	2
Histopathology/Digital Pathology	4	0	0	2	6
Blood, urine, genomic, or molecular biomarkers	9	0	1	4	14
Combined imaging, biomarker, or clinical data	8	0	0	5	13
Electronic health record or clinical prediction models	1	0	2	3	6
Total	47	10	8	19	84

3.5.1. Modality–Function Relationships in AI Diagnostic Trials

The relationship between diagnostic modality and the functional role of artificial intelligence (AI) systems is summarized in Table 4. Across the included trials, the most frequent AI applications were classification and risk prediction and lesion detection and segmentation, which together accounted for the majority of studies. MRI-based systems represented the largest group of imaging modalities and included applications spanning multiple functional categories.

Non-imaging approaches were primarily associated with predictive tasks. Biomarker-based and multimodal AI systems were predominantly applied to classification and risk prediction, while imaging modalities such as MRI, ultrasound, and PET were more commonly used for lesion detection and segmentation tasks. Applications involving algorithm development, workflow optimization, and clinical decision support systems were reported less frequently across modalities. The full distribution of AI functional themes across diagnostic modalities is presented in Table 4.

3.5.2. Study Design–Comparator Relationships in AI Diagnostic Trials

The relationship between study design and comparator framework is summarized in Table 5. Overall, prospective observational studies constituted the majority of trials, representing 71 of the included studies. Within this design category, several evaluation strategies were used, including stand-alone AI evaluations, comparisons with clinician interpretation, and comparisons with existing diagnostic tests or workflows.

Across all study designs, stand-alone AI evaluations or algorithm development studies were the most frequently reported comparator framework, followed by alternative comparative strategies and comparisons with existing diagnostic tests or workflows. Comparisons between AI systems and clinician interpretation were reported in a smaller proportion of studies. The detailed distribution of comparator frameworks across study designs is provided in Table 5.

3.5.3. Modality–Reference Standard Relationships in AI Diagnostic Trials

The reference standards used for validation of artificial intelligence (AI) systems across diagnostic modalities are summarized in Table 6. Histopathology or biopsy confirmation was the most frequently reported reference standard overall, followed by composite or unspecified reference standards, imaging-based references, and clinical outcomes or follow-up endpoints. MRI-based studies included multiple types of reference standards, whereas ultrasound-based studies relied exclusively on histopathology or biopsy confirmation. PET-based studies also used several validation approaches, while computational pathology, biomarker-based, multimodal, and clinical data–driven studies showed variable distributions of reference standards across categories.

4. Discussion

4.1. Principal Findings

This study provides a registry-based perspective on how artificial intelligence (AI) diagnostic systems are currently structured and evaluated within clinical trial settings, offering insight into their stage of development and potential pathways toward clinical integration. The findings suggest that research activity is concentrated primarily in imaging-based diagnostic platforms, with magnetic resonance imaging (MRI) representing the most frequently studied modality. Biomarker-based models, computational pathology systems, and multimodal approaches integrating imaging, molecular, and clinical data were also identified, although in smaller proportions. Across the included trials, the most common functional roles of AI appeared to involve classification and risk prediction, followed by lesion detection and segmentation, indicating that current development may be centered largely on diagnostic assessment and lesion characterization rather than fully autonomous decision-making.

The methodological profile of the included studies also reflects this stage of development. Most trials were prospective observational investigations evaluating algorithm performance in diagnostic settings, while comparator frameworks varied across stand-alone algorithm evaluation, comparison with clinician interpretation, and comparison with established diagnostic tests or workflows. Histopathology or biopsy confirmation was the most frequently reported reference standard, indicating continued reliance on biopsy-based confirmation in the clinical evaluation of AI systems for prostate cancer diagnosis.

4.2. Diagnostic Modalities and Functional Orientation of AI Development

Imaging-based AI applications, particularly those using MRI, formed the largest segment of the trial landscape. This distribution is consistent with the central role of multiparametric MRI (mpMRI) in contemporary prostate cancer diagnostics and with the broader literature identifying MRI as a major platform for AI development in this field.^22-25 mpMRI provides high-resolution anatomical and functional information that is well suited to radiomics and deep learning approaches, and radiomic features derived from T2-weighted, diffusion-weighted, and dynamic contrast-enhanced sequences have been widely used in models aimed at lesion detection, localization, and risk estimation.^22,26,27 The prominence of MRI-based trials in the present analysis is therefore aligned with both its established position in prostate cancer diagnostic pathways and its suitability for data-intensive algorithm development.

The functional distribution of AI applications in the included trials follows a similar pattern. Classification and risk prediction represented the largest category, followed by lesion detection and segmentation, indicating that many current systems are designed to estimate disease probability, stratify risk, or identify suspicious lesions within imaging and related datasets. This emphasis is also reflected in the wider prostate cancer AI literature, where models are commonly developed to predict clinically significant disease, estimate tumor aggressiveness, or support lesion localization using radiomic, biomarker, and clinical features.^22,23,25 In imaging-based settings, these tools frequently generate lesion-level probability maps, segmentation masks, or risk scores that may support image interpretation and biopsy targeting.^7,27-29

Alongside this dominant imaging-centered pattern, the registry data also identified biomarker-based, computational pathology, and multimodal AI systems. This broader distribution is consistent with growing interest in genomic, transcriptomic, and other biomarker-driven models for disease detection and risk stratification, as well as efforts to combine imaging features with molecular and clinical variables in multimodal frameworks.^23,30-34 This emphasis on imaging-centered applications, particularly within MRI-based workflows, appears to be consistent with earlier work suggesting that machine learning approaches in prostate MRI are commonly directed toward lesion detection, segmentation, and assessment of tumor aggressiveness, while still requiring broader validation across institutions and imaging settings.³⁵ In the present registry analysis, however, these non-imaging approaches remained less frequently represented than MRI-centered systems.

4.3. Methodological Patterns and Clinical Evaluation

Most of the included trials were conducted in prospective observational settings, indicating that clinical AI research in prostate cancer is still oriented mainly toward prospective performance assessment rather than controlled testing of AI-assisted interventions within clinical decision pathways. Broader analyses of clinical AI research have reported a similar predominance of prospective observational evaluation before interventional or randomized testing, particularly during phases focused on feasibility and diagnostic performance under clinically relevant conditions.^33,36

The comparator frameworks reported in the registered trials are consistent with this pattern. A substantial proportion of studies involved stand-alone AI evaluation or algorithm development, whereas fewer studies directly compared AI systems with clinician interpretation or existing diagnostic workflows. Similar benchmarking-oriented approaches have been described in the wider AI literature, where internal validation and performance comparison often precede more clinically integrated evaluation.^19,31 Comparative studies involving clinician interpretation have become increasingly visible in diagnostic AI research and are often used to examine whether AI systems perform comparably to, or in combination with, expert readers in diagnostic tasks. However, many AI diagnostic systems remain evaluated primarily through benchmarking or performance comparison rather than through prospective assessment of their impact on clinical outcomes or healthcare delivery.^19,31

Studies comparing AI systems with established diagnostic tests or workflows occupy a related but distinct position within this evaluation landscape. Such designs may help clarify how algorithm outputs align with existing pathways of image interpretation and biopsy decision-making, although performance observed in benchmark settings does not necessarily translate directly into routine use, where differences in validation strategy, dataset characteristics, and clinical context may influence interpretation and limit comparability across studies.^37,38 Within the present analysis, the range of comparator strategies appears to reflect differences in study objectives as well as variation in the developmental stage of the AI systems under investigation.

4.4. Validation Standards in AI Diagnostic Trials

Histopathology or biopsy confirmation was the most frequently reported reference standard across the included trials. This is in keeping with the established role of biopsy and surgical pathology as the principal methods for confirming prostate cancer diagnosis and characterizing tumor features in clinical practice.^39,40 The predominance of histopathological validation in the registry data therefore reflects the continuing importance of pathology-based confirmation as the benchmark against which imaging, biomarker, and computational models are assessed.

At the same time, the included trials also incorporated imaging-based references, clinical outcomes or follow-up endpoints, and composite or unspecified validation approaches. This variability may reflect the evolving nature of AI evaluation in diagnostic settings. Imaging-derived validation strategies have been used in studies evaluating AI-assisted MRI and PET interpretation, particularly when lesion localization or disease characterization is assessed in relation to subsequent imaging findings or additional diagnostic procedures.^41,42 Clinical outcome-based validation has also been applied in studies linking imaging or molecular features with disease progression or treatment-related outcomes.⁴³ Composite reference standards that combine pathology, imaging, and follow-up data have likewise been proposed when diagnostic assessment extends beyond a single confirmatory source or when longitudinal information contributes to interpretation.⁴⁴

Radiomics-based approaches further illustrate this trend, as imaging features are increasingly integrated with clinical variables to support diagnostic and prognostic modeling, although broader multicenter validation may still be required to support reliable clinical translation.⁴⁵ More broadly, existing literature has emphasized the importance of multicenter prospective validation, standardized reporting, and clearer evaluation frameworks to improve comparability across studies and to support more consistent clinical implementation of AI-based diagnostic systems.^36-38

4.5. Implications for Clinical Practice and Research

The overall pattern in the registered trials places current AI development in prostate cancer largely within an adjunctive diagnostic role. The predominance of imaging-based platforms, classification models, lesion detection tools, and benchmarking-oriented study designs is more consistent with decision-support use than with autonomous clinical deployment. Similar trajectories have been described in the broader literature, where early implementation of clinical AI is often framed around supporting clinician interpretation in imaging- and pathology-intensive specialties rather than replacing it outright.^43,46,47

The presence of biomarker-based and multimodal systems in the registry also indicates that development is extending beyond image-only models. This may be relevant for future work seeking to combine radiological, pathological, molecular, and clinical information within integrated diagnostic frameworks.^33,34 Within the present analysis, however, relatively few trials appeared to address downstream effects on clinical decision-making, workflow integration, or patient-centered outcomes, leaving much of the current evidence concentrated in algorithm evaluation and diagnostic benchmarking rather than in implementation-oriented assessment.

This pattern is broadly aligned with recent literature indicating that AI applications in prostate cancer are increasingly directed toward workflow integration, biomarker development, and clinical decision support, although widespread clinical implementation may still depend on robust multicenter validation and collaborative development frameworks.¹²

4.6. Study Limitations

The findings of this study should be interpreted in light of several limitations. This analysis relied on information reported in the ClinicalTrials.gov registry, and the extracted data therefore reflect the level of detail available in trial registrations rather than the fuller methodological descriptions that may appear in published reports. Important technical details, including model architecture, feature engineering, preprocessing methods, and analytical pipelines, were not consistently available across registry entries.

The classification of diagnostic modalities, functional roles, comparator frameworks, and validation strategies was based on registry-reported information and may therefore have been influenced by variation in reporting completeness and terminology. Although consistent categorization criteria were applied, some studies may have included overlapping modalities or multiple functional aims that were simplified during classification. Because the analysis was limited to registered clinical trials, it captures the clinical investigation landscape rather than the full spectrum of AI research in prostate cancer. Algorithm development studies conducted exclusively on retrospective institutional datasets may not have been registered and therefore would not be represented. The analysis was also restricted to ClinicalTrials.gov and did not include other international trial registries. In addition, registry status reflects the information available at the time of data collection; because many studies remain ongoing, not yet recruiting, or incompletely updated, additional methodological details and outcome data may become available as the field continues to evolve.

5. Conclusion

This registry-based evidence-mapping study shows that clinical research on artificial intelligence in prostate cancer screening and diagnosis is expanding but remains concentrated in early-phase, performance-oriented investigations. Most registered trials focus on imaging-based applications, particularly MRI, and primarily evaluate AI for classification, risk prediction, lesion detection, and segmentation. The predominance of prospective observational designs and stand-alone algorithm assessments suggests that the field is still largely centered on technical validation rather than on implementation within real-world clinical pathways.

Overall, current evidence indicates that AI in prostate cancer diagnostics is being developed mainly as a decision-support technology rather than as an autonomous clinical tool. Although biomarker-based, pathology-based, and multimodal systems are emerging, broader clinical adoption will likely depend on stronger multicenter validation, more standardized reporting, transparent evaluation frameworks, and greater emphasis on clinically meaningful outcomes, workflow integration, and patient impact. These findings provide a structured overview of the current trial landscape and may help guide future research toward more robust and clinically relevant evaluation of AI-assisted diagnostic systems in prostate cancer.

Consent to Participate

This study analyzed publicly available registry data and published literature and did not involve human participants or identifiable personal information.

Footnotes

Acknowledgment

The project was funded by KAU Endowment (WAQF) at king Abdulaziz University, Jeddah, Saudi Arabia. The authors, therefore, acknowledge with thanks WAQF and the Deanship of Scientific Research (DSR) for technical and financial support.

ORCID iD

Mohammed M. Aldurdunji

Author Contributions

MMA conceptualized and designed the study. MMA & RHH developed methodology and data framework, formal analysis and interpretation conducted investigation and data curation. The original draft was prepared by MMA, while RHH, critically reviewed, revised and edited the manuscript. RHH supervised the study, coordinated project administration, and approved the final version of the manuscript. All authors read and approved of the final manuscript and agree to be accountable for all aspects of the work.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The project was funded by KAU Endowment (WAQF) at king Abdulaziz University, Jeddah, Saudi Arabia. The authors, therefore, acknowledge with thanks WAQF and the Deanship of Scientific Research (DSR) for technical and financial support.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

The datasets generated and analyzed during the present study are available from the corresponding author upon reasonable request.*

AI tool disclosure

ChatGPT (GPT-5, OpenAI) was used under author supervision to enhance the clarity, structure, and linguistic precision of the manuscript. The tool did not participate in data analysis, interpretation, or the generation of scientific findings. All conceptualization, results interpretation, and conclusions are entirely those of the authors.

Appendix

References

Padhani

Weinreb

Rosenkrantz

Villeirs

Türkbey

Barentsz

. Prostate Imaging-Reporting and Data System Steering Committee: PI-RADS V2 Status Update and Future Directions. Eur Urol. 2019;75(3):385-396. doi:10.1016/j.eururo.2018.05.035.

Stabile

Giganti

Rosenkrantz

, et al. Multiparametric MRI for Prostate Cancer Diagnosis: Current Status and Future Directions. Nat Rev Urol. 2019;17(1):41-61. doi:10.1038/s41585-019-0212-4.

Panebianco

Pecoraro

Fiscon

Paci

Farina

Catalano

. Prostate Cancer Screening Research Can Benefit From Network Medicine: An Emerging Awareness. NPJ Syst Biol Appl. 2020;6(1):13. doi:10.1038/s41540-020-0133-0.

Dushimova

Iztleuov

Chingayeva

, et al. Overdiagnosis and Overtreatment in Prostate Cancer. Diseases. 2025;13(6):167. doi:10.3390/diseases13060167.

Chaloupka

Pyrgidis

Ebner

, et al. Added Value of Randomised Biopsy to Multiparametric Magnetic Resonance Imaging‐targeted Biopsy of the Prostate in a Contemporary Cohort. Bju Int. 2024;133(5):548-554. doi:10.1111/bju.16248.

Bardis

Houshyar

Chang

, et al. Applications of Artificial Intelligence to Prostate Multiparametric MRI (mpMRI): Current and Emerging Trends. Cancers (Basel). 2020;12(5):1204. doi:10.3390/cancers12051204.

Penzkofer

Padhani

Türkbey

, et al. ESUR/ESUI Position Paper: Developing Artificial Intelligence for Precision Diagnosis of Prostate Cancer Using Magnetic Resonance Imaging. Eur Radiol. 2021;31(12):9567-9578. doi:10.1007/s00330-021-08021-6.

Froń

Semianiuk

Lazuk

, et al. Artificial Intelligence in Urooncology: What We Have and What We Expect. Cancers (Basel). 2023;15(17):4282. doi:10.3390/cancers15174282.

Ciccone

Garofano

Del

, et al. Improving early prostate cancer detection through artificial intelligence: Evidence from a systematic review. Cancers (Basel). 2025;17(21):3503. doi:10.3390/cancers17213503.

10.

Coccia

. Digital Pathology Ecosystem: Basic Elements to Revolutionize the Diagnosis and Monitoring of Diseases in Health Sector. In: Faghih

, ed. Digital Entrepreneurship:Exploring Alertness, Orientation, and Innovation in the Digital Economy. doi:10.1007/978-3-031-58359-9_5. Cham: Springer; 2024:111-134.

11.

Goldenberg

Nir

Salcudean

. A new era: artificial intelligence and machine learning in prostate cancer. Nat Rev Urol. 2019;16(7):391-403. doi:10.1038/s41585-019-0193-3.

12.

Baydoun

Jia

Zaorsky

, et al. Artificial intelligence applications in prostate cancer. Prostate Cancer Prostatic Dis. 2024;27(1):37-45. doi:10.1038/s41391-023-00684-0.

13.

Hadjiiski

Kenny

Chan

, et al. AAPM Task Group Report 273: Recommendations on Best Practices for AI and Machine Learning for Computer‐aided Diagnosis in Medical Imaging. Med Phys. 2023;50(2):e1-e24. doi:10.1002/mp.16188.

14.

Omiye

Gui

Daneshjou

Cai

Muralidharan

. Principles, Applications, and Future of Artificial Intelligence in Dermatology. Front Med. 2023;10:1278232. doi:10.3389/fmed.2023.1278232.

15.

Din

Agarwal

Grzeda

Wood

Modat

Booth

. Detection of Cerebral Aneurysms Using Artificial Intelligence: A Systematic Review and Meta-Analysis. J Neurointerv Surg. 2022;15(3):262-271. doi:10.1136/jnis-2022-019456.

16.

Zhao

Liu

Xie

, et al. What Benefit Can Be Obtained From Magnetic Resonance Imaging Diagnosis With Artificial Intelligence in Prostate Cancer Compared With Clinical Assessments. Mil Med Res. 2023;10(1):29. doi:10.1186/s40779-023-00464-w.

17.

Moons

Damen

Kaul

, et al. PROBAST+AI: An Updated Quality, Risk of Bias, and Applicability Assessment Tool for Prediction Models Using Regression or Artificial Intelligence Methods. BMJ. 2025;388:e082505. doi:10.1136/bmj-2024-082505.

18.

Hunik

Chaabouni

Laarhoven

, et al. Diagnostic Prediction Models for Primary Care, Based on AI and Electronic Health Records: Systematic Review. Jmir Med Informatics. 2025;13. doi:10.2196/62862. e62862-e62862.

19.

Belue

Türkbey

. Tasks for Artificial Intelligence in Prostate MRI. Eur Radiol Exp. 2022;6(1):33. doi:10.1186/s41747-022-00287-9.

20.

Gresham

Meinert

Gresham

Piantadosi

Meinert

. Update on the Clinical Trial Landscape: Analysis of ClinicalTrials.gov Registration Data, 2000–2020. Trials. 2022;23(1):858. doi:10.1186/s13063-022-06569-2.

21.

Rivera

Liu

Chan

, et al. Guidelines for Clinical Trial Protocols for Interventions Involving Artificial Intelligence: The SPIRIT-AI Extension. Nat Med. 2020;26(9):1351-1363. doi:10.1038/s41591-020-1037-7.

22.

Chen

Fan

Huang

Zhang

. Algorithms on the Rise: A Machine Learning–driven Survey of Prostate Cancer Literature. Front Oncol. 2025;15:1675459. doi:10.3389/fonc.2025.1675459.

23.

Zhu

Yang

Liu

, et al. Multimodal Fusion Of mpMRI Radiomics, Clinical Features, and Hematological Biomarkers Enhances Machine Learning‐Based Prediction of Biochemical Recurrence in Prostate Cancer Patients. Chem Biol Drug Des. 2025;106(6):e70206. doi:10.1111/cbdd.70206.

24.

Zhang

Wan

, et al. Predictive Value of Multiparametric Magnetic Resonance Imaging (T2-Weighted Imaging and Apparent Diffusion Coefficient) for Pathological Grading of Prostate Cancer: A Meta-Analysis. Int Braz J Urol. 2025;51(3):e20240509. doi:10.1590/s1677-5538.ibju.2024.0509.

25.

Lyu

Tong

Yang

, et al. Bibliometric and Visual Analysis of Radiomics for Evaluating Lymph Node Status in Oncology. Front Med. 2024;11:1501652. doi:10.3389/fmed.2024.1501652.

26.

Michelutti

Tel

Robiony

, et al. Updates, Applications and Future Directions of Deep Learning for the Images Processing in the Field of Cranio-Maxillo-Facial Surgery. Bioengineering. 2025;12(6):585. doi:10.3390/bioengineering12060585.

27.

Patel

Huang

Rashid

Varghese

Gholamrezanezhad

. A Narrative Review of the Use of Artificial Intelligence in Breast, Lung, and Prostate Cancer. Life. 2023;13(10):2011. doi:10.3390/life13102011.

28.

Feuerecker

Heimer

Geyer

, et al. Artificial Intelligence in Oncological Hybrid Imaging. Röfo - Fortschritte Auf Dem Gebiet Der Röntgenstrahlen Und Der Bildgeb Verfahren. 2022;195(02):105-114. doi:10.1055/a-1909-7013.

29.

Sun

Wang

, et al. Using an Artificial Intelligence Model to Detect and Localize Visible Clinically Significant Prostate Cancer in Prostate Magnetic Resonance Imaging: A Multicenter External Validation Study. Quant Imaging Med Surg. 2024;14(1):43-60. doi:10.21037/qims-23-791.

30.

Alarcón-Zendejas

Scavuzzo

Jiménez-Ríos

MÁ

, et al. The Promising Role of New Molecular Biomarkers in Prostate Cancer: From Coding and Non-Coding Genes to Artificial Intelligence Approaches. Prostate Cancer Prostatic Dis. 2022;25(3):431-443. doi:10.1038/s41391-022-00537-2.

31.

Ivanova

Pescia

Trapani

, et al. Early Breast Cancer Risk Assessment: Integrating Histopathology With Artificial Intelligence. Cancers (Basel). 2024;16(11):1981. doi:10.3390/cancers16111981.

32.

Goda

Abdel-Aziz

. Exploiting Artificial Intelligence in Precision Oncology: An Updated Comprehensive Review. J Transl Med. 2025;23(1):1397. doi:10.1186/s12967-025-07308-2.

33.

Manzoor

Gupta

Pinky

, et al. A Systematic Review of Multimodal Deep Learning and Machine Learning Fusion Techniques for Prostate Cancer Classification. medRxiv. 2025:25333235. doi:10.1101/2025.08.07.25333235.

34.

Urso

Badrane

Manco

, et al. The Role of Radiomics and Artificial Intelligence Applied to Staging PSMA PET in Assessing Prostate Cancer Aggressiveness. J Clin Med. 2025;14(10):3318. doi:10.3390/jcm14103318.

35.

Cuocolo

Cipullo

Stanzione

, et al. Machine learning applications in prostate cancer magnetic resonance imaging. Eur Radiol Exp. 2019;3(1):35. doi:10.1186/s41747-019-0109-2.

36.

Tiwari

Mishra

Kuo

. Current AI Technologies in Cancer Diagnostics and Treatment. Mol Cancer. 2025;24(1):159. doi:10.1186/s12943-025-02369-9.

37.

Sun

Zhang

Tong

. Application Progress of Artificial Intelligence in Tumor Diagnosis and Treatment. Front Artif Intell. 2025;7:1487207. doi:10.3389/frai.2024.1487207.

38.

Zhang

Wei

Nie

, et al. Research on the Developments of Artificial Intelligence in Radiomics for Oncology Over the Past Decade: A bibliometric and Visualized Analysis. Discov Oncol. 2025;16(1):763. doi:10.1007/s12672-025-02590-4.

39.

Mottet

van den Bergh

RCN

Briers

, et al. EAU-EANM-ESTRO-ESUR-SIOG Guidelines on Prostate Cancer-2020 Update. Part 1: Screening, Diagnosis, and Local Treatment with Curative Intent. Eur Urol. 2021;79(2):243-262. doi:10.1016/j.eururo.2020.09.042.

40.

Epstein

Zelefsky

Sjoberg

, et al. A Contemporary Prostate Cancer Grading System: A Validated Alternative to the Gleason Score. Eur Urol. 2016;69(3):428-435. doi:10.1016/j.eururo.2015.06.046.

41.

Lambin

Leijenaar

RTH

Deist

, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol. 2017;14(12):749-762. doi:10.1038/nrclinonc.2017.141.

42.

Gillies

Kinahan

Hricak

. Radiomics: Images Are More than Pictures, They Are Data. Radiology. 2016;278(2):563-577. doi:10.1148/radiol.2015151169.

43.

Topol

. High-performance medicine: the convergence of human and artificial intelligence. Nat Med. 2019;25(1):44-56. doi:10.1038/s41591-018-0300-7.

44.

Nagendran

Chen

Lovejoy

, et al. Artificial intelligence versus clinicians: systematic review of design, reporting standards, and claims of deep learning studies. BMJ. 2020;25:m689. doi:10.1136/bmj.m689.

45.

Chaddad

Tan

Liang

, et al. Advancements in MRI-Based Radiomics and Artificial Intelligence for Prostate Cancer: A Comprehensive Review and Future Prospects. Cancers (Basel). 2023;15(15):3839. doi:10.3390/cancers15153839.

46.

Kelly

Karthikesalingam

Suleyman

Corrado

King

. Key challenges for delivering clinical impact with artificial intelligence. BMC Med. 2019;17(1):195. doi:10.1186/s12916-019-1426-2.

47.

Hosny

Parmar

Quackenbush

Schwartz

Aerts

HJWL

. Artificial intelligence in radiology. Nat Rev Cancer. 2018;18(8):500-510. doi:10.1038/s41568-018-0016-5.