Journal of Artificial Intelligence and Digital Health

INTRODUCTION

Over the past six decades, advancements in prostate cancer diagnostics have significantly improved early detection and treatment outcomes [1]. This malignancy, known as prostate cancer, is the second most frequently diagnosed and has a significant impact on men's health worldwide, presenting significant challenges in early detection and accurate prognosis [2]. Conventional diagnostic methods, including Prostate-Specific Antigen (PSA) testing and Digital Rectal Examination (DRE), frequently encounter challenges related to specificity and sensitivity. These methods tend to produce high false-positive rates, leading to unnecessary biopsies and overdiagnosis. Moreover, imaging modalities like multiparametric MRI (mpMRI) and ultrasound, while effective, are subject to inter-observer variability and constrained by the quality of imaging data [3].  These limitations hinder the ability to deliver precise, reliable, and early diagnosis, thereby impacting treatment outcomes and patient quality of life [1].

Progress in Machine Learning (ML) and Artificial Intelligence (AI) has led to valuable tools for overcoming these challenges by leveraging computational models to analyze complex, high-dimensional datasets. However, individual ML models often struggle to handle uncertainties inherent in medical data, such as variability in patient demographics and incomplete datasets [4][5]. Despite improvements in predictive accuracy with algorithms like XGBoost and deep learning, the interpretability of these models remains a barrier to clinical adoption. Thus, there is a pressing need for diagnostic frameworks that balance precision, robustness, and transparency. Among AI techniques, Interval Type-2 Fuzzy Logic (IT2FL) from Fuzzy logic systems (FLS), introduced by Zadeh in 1965, offers a robust framework for managing uncertainties and imprecise medical data. IT2FL enhances decision-making by incorporating linguistic variables and fuzzy inference systems, thereby providing a nuanced interpretation of patient data. By extending binary logic to include values between absolute truth and falsehood, fuzzy logic systems map input variables to crisp outputs and simulate human decision-making using linguistic variables and rules [6]. Unlike Type-1 Fuzzy Logic, IT2FL can handle more complex uncertainties, making it particularly suitable for medical applications [6-7].

Complementing IT2FL, the XGBoost algorithm has gained prominence for its superior performance in classification tasks. Known for its ability to reduce bias and variance, XGBoost delivers high predictive accuracy, this makes it especially well-suited for medical applications that demand high levels of precision. The integration of these two methodologies, IT2FL and XGBoost represents a hybridized model that combines the strengths of uncertainty management and predictive optimization. Artificial Intelligence (AI) has significantly advanced prostate cancer detection and prognosis, particularly in radiology and imaging techniques. AI applications, such as convolutional neural networks (CNNs) and machine learning (ML) models, have improved diagnostic precision and treatment planning. For instance, ML techniques like supervised and unsupervised learning, decision trees, and deep learning have been widely adopted for prostate cancer research [8]. CNNs have been particularly effective in enhancing prostate segmentation and aligning imaging with the Prostate Imaging Reporting and Data System (PI-RADS), as shown by [9]. These advancements surpass traditional diagnostic methods and human interpretation. Comparative studies highlight the superior performance of ML algorithms over conventional approaches. Boosting methods achieved the highest classification accuracy, as reported by [10].

Similarly, studies on hybrid models combining techniques like gradient boosting and fuzzy logic have shown improved reliability in handling medical data uncertainties [11]. The integration of AI in prostate cancer research has also expanded into predictive analytics. For example, hybrid AdaBoost-SVM algorithms have demonstrated exceptional diagnostic accuracy for breast cancer and similar potential for prostate cancer applications. The role of biomarkers and high-dimensional datasets in AI-driven diagnostic tools has grown, emphasizing the need for rigorous experiments to validate model efficacy [12]. Additionally, methods such as multiparametric MRI (mpMRI) and machine learning have enhanced risk stratification and biopsy accuracy, reducing false positives [13] . Challenges remain in implementing these technologies clinically due to data variability, computational requirements, and model interpretability. Future directions emphasize integrating explainable AI for enhanced trustworthiness, federated learning for data privacy, and expanding clinical trials to validate ML models. This comprehensive review identifies the potential of AI and ML in improving prostate cancer care while addressing existing limitations.

Li H, conducted an in-depth analysis of segmentation, machine learning, and deep learning methods to enhance the prediction of prostate cancer probability rates [14]. In Lee C, a reinforcement learning approach was utilized to improve the accuracy of prostate cancer prediction [15]. The study also explored various models and structural frameworks used for the precise identification and classification of prostate cancer. Similarly, Tataru OS examined recent advancements in AI for the development of automated prostate cancer diagnosis systems [16]. AI has demonstrated remarkable accuracy in medical imaging, aiding in the detection of prostate abnormalities and forecasting clinical outcomes such as patient survival and response to treatment. This systematic review investigates the potential of this hybrid approach, focusing on its application to prostate cancer detection and prognosis. Specifically, it explores the integration flow of IT2FL and XGBoost, evaluates the input parameters influencing model performance, and examines the significance of linguistic variables and XGBoost’s implementation. By analyzing existing literature from 2017 to 2024, this review purposes to provide comprehensive insights into the efficacy and applicability of hybrid models in prostate cancer diagnostics, setting the stage for future advancements in precision medicine.

While prior reviews such as [14], [15], and [16] have examined segmentation-based, reinforcement-learning-based, and general AI-driven approaches to prostate cancer diagnosis, none has focused specifically on the integration of Interval Type-2 Fuzzy Logic with XGBoost. This represents a gap in the literature, since existing reviews tend to treat uncertainty-handling techniques and gradient-boosting classifiers as separate research streams rather than as a combined diagnostic framework. The present review addresses this gap by systematically synthesizing studies relevant to the IT2FL-XGBoost hybrid, with the aim of clarifying how uncertainty management and predictive optimization can jointly improve prostate cancer detection and prognosis.

The integration of Interval Type-2 Fuzzy Logic (IT2FL) and XGBoost offers a novel approach to address these gaps. IT2FL provides a robust mechanism for managing data uncertainties through its ability to model imprecision and ambiguity using linguistic variables and fuzzy inference systems. This capability is particularly suited for medical applications where patient data often lack exact values. On the other hand, XGBoost, a gradient-boosting algorithm, is renowned for its superior classification performance, scalability, and ability to minimize both bias and variance. By combining these methodologies, the hybrid model hope to enhance diagnostic precision, manage data uncertainty, supports clinical decision-making and bridge theory and practice.

METHODOLOGY

This section offers an outline of the methodology employed in the literature review, including its objectives, research questions, inclusion and exclusion criteria, and the study framework. Subsection (A) outlines the steps taken during the literature review process. Subsection (B) addresses the formulation of research questions. Subsection (C) details the article selection strategy, while Subsection (D) includes Inclusion and Exclusion Criteria. This review followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines to ensure a transparent and reproducible review protocol.

Methodological Description

Table 1: Tabulated Literature on Detection Problem

Table 2: Summarized Literature on Diagnostic Problems

Table 3: Summarized Literature on Prediction Problems

Table 4: Summarized Literature on Classification Problems

Table 5: Gaps of Single Methodology in Previous Studies

Research Questions Steps

Preparation Steps

The preparation steps are essential for laying the groundwork for formulating well-structured and meaningful research questions. In this phase, the researcher identifies the study's primary focus and determines the investigation's scope.

RQ1: Why is ML relevant in prostate cancer detection and prognosis?

RQ1.1: What ML algorithms are adopted in the studies under review for prostate cancer detection and prognosis?

RQ1.2: Which regions of the world have utilized ML models for prostate cancer research?

RQ1.3: To what extent has the use of machine learning in prostate cancer detection translated into clinical implementation beyond theoretical models

RQ2: How many studies have been conducted on ML for prostate cancer detection and prognosis from 2017 to 2024?

RQ3: What are the most frequently used keywords in research on hybrid models for prostate cancer detection and prognosis?

RQ4: What evaluation metrics are most used to assess detection accuracy in prostate cancer models?

Article Selection Strategy

Execution Step

The Execution Step is the stage where the planned research activities are implemented, involving data collection and the application of the research design.

Search Strategy

The data collection was carried out using selective, term-based searches across articles from various sources published between 2017 and 2024. We searched research databases such as Google Scholar, IEEE Xplore, Elsevier, and Springer for the specified terms, focusing on titles and abstracts of the articles. The search process was guided by the research questions, and advanced search techniques like "AND" and "OR" were used to narrow down the results.

Digital Libraries Resources

Table 6 summarizes the digital library resources utilized for gathering data, with an emphasis on English-language materials obtained via keyword searches. Data collection occurred from January 2017 to September 2024 and included a variety of sources such as peer-reviewed journals, academic articles, e-books, conference proceedings, and online workshops. Major platforms accessed in this research included ACM Digital Library, Elsevier, Google Scholar, IEEE Xplore, Springer, ResearchGate, ScienceDirect, and Systematic Scholar, aided by the Google search engine.

Table 6:  Resource Libraries

Search Terms

Search terms are specific keywords or phrases used in databases, search engines, or digital libraries to find relevant literature, articles, or information related to a research topic. These terms include essential keywords that represent the main concepts of the research, as shown in:

1. (Interval Type-2 Fuzzy Logic OR IT2FL) AND (Prostate Cancer Detection OR Cancer Diagnosis) AND (Hybrid Models OR Fuzzy Logic Systems) AND (Medical Imaging OR MRI and Ultrasound)
2. (Machine Learning OR XGBoost) AND (Prostate Cancer Prognosis OR Cancer Risk Prediction) AND (Hybrid Models OR Integrated Systems) AND (Medical Diagnosis OR Non-Invasive Techniques)
3. (Fuzzy Logic Systems OR IT2FL Applications) AND (Cancer Detection OR Prostate Tumor Localization) AND (XGBoost OR Machine Learning Models) AND (Medical Data Uncertainty OR Diagnostic Accuracy)
4. (AI in Healthcare OR Hybrid AI Models) AND (Prostate Cancer Detection OR Cancer Staging) AND (Interval Type-2 Fuzzy Logic OR IT2FL) AND (XGBoost OR Predictive Models)
5. (Prostate Cancer Screening OR Diagnostic Tools) AND (Machine Learning OR XGBoost Applications) AND (Fuzzy Logic OR Uncertainty Handling) AND (Non-Invasive Imaging OR MRI-Ultrasound Fusion)

Study Selection Based on Inclusion and Exclusion Criteria

Figure 1 shows the Prisma flow diagram, illustrating the study selection process, detailing the number of records identified, screened, excluded, and included in the review. Searches were conducted across nine databases, while no records were retrieved from registers, websites, organizations, or citations. A total of 20 duplicates were removed, with 379 records automatically excluded. No records were screened or assessed for retrieval, and no reports were sought out or retrieved from other sources. During the assessment phase, several reports were excluded for specific reasons. Ultimately, 121 new studies and associated reports were included in the final analysis, highlighting the comprehensive effort to identify relevant literature.

The exclusion criteria for studies considered in this review include: i) Sample Size Thresholds - Studies with a sample size of fewer than 50 participants were excluded to ensure statistical reliability and generalizability of results. ii) Publication Quality - Articles published in non-peer-reviewed journals or lacking sufficient methodological rigor were excluded to maintain the credibility of the review. iii) Geographic Limitations - studies restricted to highly localized datasets without diverse demographic representation were excluded to enhance the applicability of findings across broader populations. iv) Timeframe - studies published outside the 2017-2024 window were excluded to focus on the most recent advancements in the field. v) Language - non-English studies were excluded due to resource constraints in translation and validation of findings. vi) Relevance - studies gthat did not specifically address prostate cancer detection or prognosis using machine learning, or that failed to include IT2FL or XGBoost as a component of the methodology, were excluded.

Figure 1: Prisma Flow Diagram of Study Selection

RESULTS AND DISCUSSION

RQ1: Why is ML relevant in prostate cancer detection and prognosis?

RQ1 is answered using the sub-questions RQ1.1 to RQ1.3.

RQ1.1: What ML algorithms are adopted in the studies under review for prostate cancer detection and prognosis?

This study employs a variety of ML algorithms to enhance diagnostic accuracy and predictive outcomes in prostate cancer research. The models utilized include semi-supervised learning, artificial neural networks (ANN), AdaBoost, boosting techniques, convolutional neural networks (CNN), decision trees, deep learning, fuzzy logic, k-nearest neighbors (KNN), recurrent neural networks (RNN), support vector machines (SVM), naive Bayes (NB), random forests, and XGBoost.

These diverse ML methods highlight the extensive application of advanced computational techniques in the field of prostate cancer research.

RQ1.2

Which regions of the world have utilized ML models for prostate cancer research?

Figure 2 and Figure 3 show that Europe and Asia have the highest number of countries utilizing machine learning models for prostate cancer research, reflecting strong research activity in these regions. North America also has significant involvement, primarily from the U.S. and Canada. Other regions, including the Middle East, Africa, South America, and Oceania, show some engagement but to a lesser extent, indicating that machine learning applications in prostate cancer research are more concentrated in Europe, Asia, and North America. This distribution likely reflects available resources, research funding, and technological advancements in these areas.

Figure 2: Region of Study with Several Countries

Figure 3: Distribution of Specified ML by Continent+

RQ1.3

To what extent has the use of ML in prostate cancer detection translated into clinical implementation beyond theoretical models.

The use of ML in prostate cancer detection has shown significant promise, particularly in tasks such as lesion detection, risk stratification, and treatment planning using imaging techniques like MRI. While ML models demonstrate high accuracy and potential for reducing observer variability, their clinical implementation remains limited due to challenges such as the need for extensive validation on diverse datasets, regulatory hurdles, and clinician training. Efforts to address these barriers include integrating ML tools into clinical trials, exploring explainable AI for trust and interpretability, and leveraging federated learning to ensure data privacy. As these advancements continue, the gap between theoretical models and real-world clinical applications is gradually narrowing, paving the way for broader adoption in prostate cancer care.

RQ2

How many studies have been conducted on ML for prostate cancer detection and prognosis from 2017 to 2024?

Figure 4 illustrates the number of studies conducted on ML for detection of prostate cancer and prognosis between 2017 and 2024. The trend shows a gradual increase in the number of studies from 2017, starting with 2 studies, and peaking in 2020 and 2021, with 14 and 13 studies, respectively. Following this peak, there is a decline in the number of studies, with 7 recorded in 2022, 9 in 2023, and 4 in 2024. This pattern indicates growing interest in the field during the earlier years of the period, followed by a tapering.

Figure 4: Machine learning used between 2017 and 2024

RQ3

 What are the most frequently used keywords in research on hybrid models for prostate cancer detection and prognosis?

Figure 5 present word cloud, showing a visualization of the most frequently used terms in the reviewed literature, highlighting key topics such as "prostate," "cancer," "machine learning," and "diagnosis".

Figure 5: provides a comprehensive visualization of terms frequently associated with research on prostate cancer detection and prognosis. The term "prostate" stands out as the most prominent, reflecting its centrality to the subject matter. Similarly, "cancer" is another dominant term, emphasizing the focus on the identification and treatment of malignant conditions affecting the prostate. The presence of "system" and "biopsy" suggests significant attention to diagnostic systems and procedures, such as biopsies, which remain a critical tool in confirming cancer diagnoses. Terms like "health" and "disease" highlight the broader context of medical research and the impact of prostate cancer on patient well-being. "Learning" and "artificial" indicate the growing reliance on machine learning and artificial intelligence technologies to enhance predictive models and decision-making processes in prostate cancer management. The appearance of "imaging" points to the pivotal role of medical imaging techniques, such as MRI and ultrasound, in detecting and monitoring the disease.

The frequent occurrence of "therapy," "PSA" (Prostate-Specific Antigen), and "survival" reflects the research interest in therapeutic interventions, biomarker-based screening methods, and survival outcomes. Words like "biomarker," "neural," and "deep" underscore the integration of biomarkers and deep learning techniques in predictive analytics and diagnostics.

Furthermore, the inclusion of terms such as "guidelines," "detection," and "predictive" signifies efforts to establish clinical best practices and improve early detection accuracy. "Radiotherapy," "management," and "decision" suggest a focus on treatment strategies and decision-support systems aimed at optimizing patient outcomes. The presence of words like "fuzzy" and "forest" indicates the adoption of advanced machine learning algorithms such as fuzzy logic and random forest models in research studies.

Additional terms such as "imaging," "Gleason," and "therapy" point to the use of imaging techniques, the significance of the Gleason grading system for classifying tumor severity, and ongoing advancements in therapeutic approaches. Lastly, terms like "survival," "disease," and "care" highlight the overarching goals of improving patient survival rates, understanding disease progression, and enhancing overall healthcare delivery.

In summary, the word cloud reveals a multidisciplinary research approach integrating computational methods, clinical diagnostics, and therapeutic strategies to improve prostate cancer detection, prognosis, and management.

Figure 5: Word Cloud Visualizing the Most Frequently Used Terms in The Reviewed Literature

RQ4

What evaluation metrics are most used to assess detection accuracy in prostate cancer models?

When assessing detection accuracy in prostate cancer models, several evaluation metrics are commonly employed To provide a thorough assessment of model performance, accuracy is defined as the ratio of correctly classified cases (both true positives and true negatives) to the total number of cases., which is crucial in cancer detection. Specificity, on the other hand, evaluates the model's effectiveness in identifying negative cases, minimizing false positives. The Positive Predictive Value (PPV) indicates the likelihood that patients identified as positive actually have the disease, while the Negative Predictive Value (NPV) reflects the probability that negative predictions are accurate. The F1 Score, the harmonic mean of precision and recall, provides a balance between these metrics, especially in cases with imbalanced datasets. Lastly, the Area Under the Receiver Operating Characteristic Curve (AUC-ROC) is utilized to assess the model's discriminative ability across various thresholds, offering insight into overall performance. These metrics collectively provide a robust framework for evaluating the effectiveness of prostate cancer detection models.

LIMITATIONS REPORTED IN THE STUDIES

The studies reviewed revealed several significant limitations that affected the performance and applicability of ML models in prostate cancer detection and prognosis. A recurring challenge was the quality and availability of medical data, which is crucial for developing accurate and reliable models. Many studies faced issues such as incomplete, imbalanced, or low-quality datasets, with missing values or a lack of diversity in patient populations. These data deficiencies often led to skewed model outcomes and limited the generalizability of the findings. Additionally, obtaining labeled data for supervised learning tasks was another common obstacle, with the high cost of acquiring detailed medical records or expert annotations being a limiting factor, especially for rare cancer cases. The computational complexity of advanced ML algorithms, particularly when using hybrid or deep learning models, was also highlighted, as these approaches often demand significant computational resources and longer processing times, posing a challenge for real-time clinical application.

Another limitation discussed in the studies was the interpretability of complex models. While advanced algorithms like deep learning offer high accuracy, they often function as "black boxes," making it difficult for healthcare professionals to understand how predictions are made. This lack of transparency raises concerns about the trustworthiness and clinical adoption of these models, as clinicians require an explanation for decisions that could directly impact patient care. Furthermore, the absence of standardized protocols for feature selection, model evaluation, and performance metrics made it challenging to compare the effectiveness of different ML approaches. Some studies also pointed out the need for better integration of these models into clinical workflows, emphasizing the importance of user-friendly interfaces that could help practitioners implement these tools effectively. Finally, the risk of overfitting was identified, particularly when models were trained on small or non-representative datasets, leading to overly optimistic performance estimates that may not reflect real-world conditions.

CONCLUSION

This systematic review synthesizes evidence on integrating Interval Type-2 Fuzzy Logic (IT2FL) and the XGBoost algorithm for prostate cancer detection and prognosis. Across the reviewed literature, IT2FL is reported to address uncertainties in medical datasets by providing a structured framework for handling complex and imprecise variables, while XGBoost is reported to improve classification accuracy by reducing bias and variance. Together, these findings suggest that combining the two techniques could offer a productive direction for improving diagnostic precision and supporting clinical decision-making, though this potential remains to be validated in prospective clinical studies.

The review analyzed 121 studies published between 2017 and 2024, drawn from bibliographic databases, academic search engines, and publisher platforms, including Google Scholar, IEEE Xplore, Elsevier, and Springer. Visualization tools including VOSviewer, Matplotlib, and Excel were used to identify trends across the reviewed literature. The reviewed studies collectively suggest that hybrid approaches combining uncertainty-handling and gradient-boosting techniques may improve early diagnosis, reduce false positives, and support treatment planning for prostate cancer, though the evidence base remains largely retrospective and has not yet been validated in large-scale prospective clinical trials.

Several limitations of this review should be acknowledged. The search was restricted to English-language sources published between 2017 and 2024, which may have excluded relevant earlier or non-English studies. In addition, some included sources, such as ResearchGate, host author-uploaded content rather than peer-curated indexing, which may affect the consistency of the underlying evidence base. Future research should prioritize prospective clinical validation of IT2FL-XGBoost hybrid approaches, the development of explainable AI techniques to improve model interpretability and clinician trust, and broader testing across diverse patient populations and healthcare settings, including under-resourced regions.

In conclusion, the reviewed literature suggests that combining IT2FL and XGBoost holds promise for advancing prostate cancer diagnostics, but realizing this potential in clinical practice will require further validation, standardization, and attention to interpretability before widespread adoption can be recommended.

ACKNOWLEDGEMENTS

The authors extend their heartfelt appreciation to TETFund for facilitating a conducive research environment through the TETFund Center of Excellence in Computational Intelligence Research at the University of Uyo. They also acknowledge the support provided by the management of the University of Uyo.

Ethical Considerations and Safeguards: As this study is a systematic review of previously published literature, it did not involve the collection of new patient or human subject data. All data analyzed were derived from publicly available, peer-reviewed publications, in accordance with standard academic and ethical practices for literature synthesis.

ETHICS APPROVAL

The study was conducted in compliance with institutional research ethics guidelines.

CONFLICTS OF INTEREST

The authors confirm that they have no conflicts of interest to disclose.