1 Introduction
It is estimated that almost 1.3 million new cases of prostate cancer (PCa) and 359,000 associated deaths worldwide in 2018, accounting for 7.1% of the total new cancers diagnosed worldwide, ranking as the second most frequent cancer and the fifth leading cause of cancer death in men.[1,2] [1,3]
Therefore, reliable and early detection of PCa has become an important priority in the field of urologic oncology. For the past 25 years, Prostate-specific antigen (PSA) has always been the gold standard for the diagnosis of PCa, followed by transrectal ultrasound (TRUS)-guided biopsy, which has resulted in decreased PCa mortality by 20% to 30%,[4] [5,6] [7] [8]
However, one of the main limitations of the mp-MRI is that its interpretation requires experienced radiologists capable of analyzing data extracted from the different MR sequences, which may lead to high inter- and intra-reader variability in diagnosis.[9]
In the past decade, several computer-aided systems[10–13] [14] [15]
The aim of the study is to conduct a systematic review and meta-analysis to:
1) evaluate the diagnostic accuracy of CAD system based on MRI images of the prostate and provides a malignancy assessment;
2) determine which classifier of CAD system is superior for the diagnosis of PCa;
3) determine whether the performance of the CAD system depends on the specific regions of the prostate.
2 Methods
This research protocol has been developed according to the Preferred Reporting Items for Systematic reviews and Meta-Analyses Protocol (PRISMA-P),[16] [17]
The protocol has been registered in PROSPERO (ID: CRD42019132543).
2.1 Eligibility criteria
2.1.1 Types of study
We will include all studies that investigated diagnostic accuracy of CAD systems based on MRI in adult patients with suspected PCa. Included studies should have sufficient information to build a 2 × 2 contingency table (true positive [TP], false positive [FP], true negative [TN], false negative [FN]). Case–control studies will be excluded when the control group entails healthy volunteers as they are not representative of the population in which CAD will be performed.
2.1.2 Participants
We will include studies that evaluate patients 18 years of age or older and with suspected PCa.
2.1.3 Setting
Our study will include participants from different clinical settings, such as hospital wards, emergency departments, and intensive care units.
2.1.4 Index test
We will include studies that CAD system was used to diagnose PCa, and study data was based on MRI.
2.1.5 Reference standards
Biopsy should serve as the reference standard.
2.2 Exclusion criteria
We will exclude the studies in which the information of a 2 × 2 contingency table are lacking, and cannot be calculated from the text or appendices; and duplicated articles, review articles, editorials, case reports, summaries, animal and cell studies, meta-analysis, letters, editorials, comments, and other irrelevant article types will be also excluded.
2.3 Search strategy
Following databases will be systematically researched for relevant studies: Cochrane library, PubMed, EMBASE, and Chinese Biomedicine Literature Database (CBM) from their inception. There will be no restrictions placed on document language or publication status. A search strategy will be developed to define subject headings and keywords for all searches. Specific search strategies (e.g., for PubMed) are as follows: (“prostatic neoplasm∗” OR “prostate neoplasm∗” OR “prostate cancer∗” OR “prostatic cancer∗” OR “prostate tumor∗” OR “prostatic tumor∗”) AND (“artificial intelligence” OR “deep learning” OR “computer-assisted” OR “machine learning” OR “neural network∗” OR “artificial inligence” OR “AI” OR “computational intelligence” OR “machine intelligence” OR “computer reasoning” OR “automated”) AND (“diagnosis” OR “diagnos∗” OR “detection” OR “sensitivity” OR “specificity” OR “accuracy”, “positive likelihood” OR “negative likelihood” OR “ROC”).We will also contact leading authors and experts in the field of PCa for additional studies via email. The bibliographies of relevant reviews and included studies will be used to identify additional references for review. Finally, we will transfer all relevant titles and abstracts to Endnote Web for selection.
2.4 Study selection
After the removal of duplicate results, the selection of potential articles reviews will be done first by title and then by abstract by 2 independent authors (LMX and CLJ). At this stage, we will exclude studies that were not described as CAD for PCa diagnosis based on MRI. Then, the full text of each potential study will be assessed for inclusion. Disagreements will be resolved through discussion and consensus, or by consulting a third member (L-Y) of the review team. The details of study selection process will be presented in the PRISMA flow chart (shown in Fig. 1 ).
Figure 1: Preferred reporting items for systematic reviews and meta-analyses flow chart of study selection process.
2.5 Data extraction
According to the characteristics of included studies, 2 reviewers will independently extract the following information:
Basic characteristics of included studies: first author, year of publication, country, patient numbers, patient ages, study design, PSA (ng/ml), testing set, reference standard;
The details of different CAD systems: field strength, classifier, Steps of CAD System, Imaging sequence used in system.
Diagnostic data: true positive (TP), TN, FP, FN, Accuracy, Sensitivity, and Specificity.
If there are any discrepancies, they will discuss and resolve by consensus with a third reviewer.
2.6 Methodological quality assessment
Two authors (LMX and LHJ) will independently evaluate the methodological quality of each eligible study using the quality assessment of diagnosis accuracy study (QUADAS-2) tool;[18]
2.7 Quality of the evidence
A Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach for diagnostic tests has now been developed, which provides guidance on how to translate accuracy data into a recommendation involving patient-important outcomes.[19]
2.8 Data analysis
We will first extract the 2 × 2 contingency table (TP, FP, TN, FN). Some of the primary studies did not directly give all the data in the 2 × 2 tables, we will calculate the missing data based on the existing data in the text or appendices in each primary study using the calculator in Review Manager 5.3. Using these tables, we determined the true-positive rate (TPR; sensitivity) the true-negative rate (TNR; specificity), a descriptive forest plot and summary receiver operating characteristic (SROC) curves will be derived by Review Manager 5.3. And the stata12.0 software will be also used to develop forest plot so as to present the sensitivity and specificity and their pooled results. SROC curves are defined by sensitivity (y-axis) and specificity (x-axis), respectively, and each data point represents 1 particular study, and the area under the curve (AUC) is the final comparison indicator. The criteria for AUC classification are 0.90 to 1 (excellence), 0.80 to 0.90 (good), 0.70 to 0.80 (fair), 0.60 to 0.70 (poor), and 0.50 to 0.60 (failure).[20]
2.9 Assessment of heterogeneity
Initially, to examine heterogeneity, we will visually inspect forest plots of each study's sensitivities and specificities as well as ROC curves related to the individual study results. Statistical heterogeneity will be evaluated informally from forest plots of the study estimates and more formally using the χ2 test (P <.1, significant heterogeneity) and I2 statistic (I2 >50% = significant heterogeneity). In addition, different diagnostic thresholds of included studies may lead to heterogeneity; we will use the Spearman correlation coefficients to test whether there is a threshold effect. When there is a threshold effect, sensitivity and specificity will be negatively correlated, and the results will present a “shoulder-arm” point distribution on the SROC curve.
2.10 Subgroup analysis
We will conduct subgroup analyses according to:
a) the type of classifier of CAD systems used to determine which classifier of CAD system is superior for the diagnosis of PCa;
b) The specific regions of the prostate (peripheral zone, transitional zone, and central gland), to investigate whether the CAD diagnostic accuracy depends on the prostate zoon.
2.11 Assessment of publication bias
If a sufficient number of studies are identified, we will investigate publication biases by Deek's funnel plot.[21]
2.12 Patient and public involvement
Neither patients nor public got involved.
3 Discussion
Although there is some evidence on the accuracy of CAD in the diagnosis of PCa, evidence is limited and was not systematically reviewed. To the best of our knowledge, this is the first study that will systematically review CAD for PCa diagnosis based on MRI. Greater scientific rigour is necessary when establishing a diagnostic strategy that represents current evidence accurately, and we will also conduct subgroup analyses according to the type of classifier of CAD systems used and the different prostate zoon.
We will conduct a systemic review of CAD system based on MRI for the diagnosis of PCa using appropriate methodologies and quality assessment tools that may feed into an evidence-based clinical practice. This will be the first systematic review to directly compare the diagnostic accuracy of CAD system based on MRI to a reference standard of PCa.
The major limitation is that the results from this systematic review will be highly dependent on the quality of the underlying primary studies, which will be mainly retrospective studies. Another possible limitation of this study, is its susceptibility to publication and small sample biases, and may not be generalisable to other settings.
Author contributions
Data curation: Meixuan Li, Huijuan Li.
Formal analysis: Xiaoqin Wang, Huijuan Li, Yumeng Song.
Funding acquisition: Jieting Liu.
Methodology: Fuxiang Liang, Meixuan Li, Liang Yao.
Project administration: Bing Song.
Resources: Jieting Liu.
Software: Meixuan Li.
Writing – original draft: Fuxiang Liang, Meixuan Li.
Writing – review & editing: Fuxiang Liang, Meixuan Li, Liang Yao, Liujiao Cao, Shidong Liu, Bing Song.
References
[1]. Bray F, Ferlay J, Soerjomataram I, et al. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2018;68:394–424.
[2]. Ferlay J, Colombet M, Soerjomataram I, et al. Estimating the global cancer incidence and mortality in 2018: GLOBOCAN sources and methods. Int J Cancer 2019;144:1941–53.
[3]. Siegel RL, Miller KD, Jemal A. Cancer statistics, 2018. CA Cancer J Clin 2018;60:277–300.
[4]. Schröder FH, Hugosson J, Roobol MJ, et al. Screening and prostate-cancer mortality in a randomized European study. N Engl J Med 2009;360:1320–8.
[5]. Caster JM, Falchook AD, Hendrix LH, et al. Risk of pathologic upgrading or locally advanced disease in early prostate cancer patients based on biopsy gleason score and psa: a population-based study of modern patients. Int J Radiat Oncol Biol Phys 2015;92:244–51.
[6]. Cohen MS, Hanley RS, Kurteva T, et al. Comparing the Gleason prostate biopsy and Gleason prostatectomy grading system: the Lahey Clinic Medical Center experience and an international meta-analysis. Eur Urol 2008;54:371–81.
[7]. Hoeks CM, Barentsz JO, Hambrock T, et al. Prostate cancer: multiparametric MR imaging for detection, localization, and staging. Radiology 2011;261:46–66.
[8]. Tan N, Margolis DJ, Lu DY, et al. Characteristics of detected and missed prostate cancer foci on 3-t multiparametric mri using an Endorectal Coil correlated with whole-mount thin-section histopathology. AJR Am J Roentgenol 2015;205:W87–92.
[9]. Litjens GJ, Barentsz JO, Karssemeijer N, et al. Clinical evaluation of a computer-aided diagnosis system for determining cancer aggressiveness in prostate MRI. Eur Radiol 2015;25:3187–99.
[10]. Bonekamp D, Kohl S, Wiesenfarth M, et al. Radiomic machine learning for characterization of prostate lesions with MRI: comparison to ADC values. Radiology 2018;289:128–37.
[11]. Giannini V, Rosati S, Regge D, et al. Specificity improvement of a CAD system for multiparametric MR prostate cancer using texture features and artificial neural networks. Health Technol 2017;7:71–80.
[12]. Kwak JT, Xu S, Wood BJ, et al. Automated prostate cancer detection using T2-weighted and high-b-value diffusion-weighted magnetic resonance imaging. Med Phys 2015;42:2368–78.
[13]. Roethke MC, Kuru TH, Mueller-Wolf MB, et al. Evaluation of an automated analysis tool for prostate cancer prediction using multiparametric magnetic resonance imaging. PLoS One 2016;11:e0159803.
[14]. Litjens GJS, Vos PC, Barentsz JO, et al. Automatic computer aided detection of abnormalities in multi-parametric prostate MRI. Proc SPIE Int Soc Opt Eng 2011;7963.
[15]. Iussich G, Correale L, Senore C, et al. Computer-aided detection for computed tomographic colonography screening: a prospective comparison of a double-reading paradigm with first-reader computer-aided detection against second-reader computer-aided detection. Invest Radiol 2014;49:173–82.
[16]. Shamseer L, Moher D, Clarke M, et al. Preferred reporting items for systematic review and meta-analysis protocols (PRISMA-P) 2015: elaboration and explanation. BMJ 2015;350:g7647.
[17]. McInnes MDF, Moher D, Thombs BD, et al. Preferred reporting items for a systematic review and meta-analysis of diagnostic test accuracy studies: the PRISMA-DTA statement. JAMA 2018;319:388–96.
[18]. Whiting PF, Rutjes AW, Westwood ME, et al. QUADAS-2: a revised tool for the quality assessment of diagnostic accuracy studies. Ann Intern Med 2011;155:529–36.
[19]. Hsu J, Brożek JL, Terracciano L, et al. Application of GRADE: making evidence-based recommendations about diagnostic tests in clinical practice guidelines. Implement Sci 2011;6:62.
[20]. Bradley AP. The use of the area under the ROC curve in the evaluation of machine learning algorithms. Pattern Recognition 1997;30:1145–59.
[21]. Deeks JJ, Macaskill P, Irwig L. The performance of tests of publication bias and other sample size effects in systematic reviews of diagnostic test accuracy was assessed. J Clin Epidemiol 2005;58:882–93.
