Histopathological evaluation and grading for prostate cancer: current issues and crucial aspects

Abstract

A crucial aspect of prostate cancer grading, especially in low- and intermediate-risk cancer, is the accurate identification of Gleason pattern 4 glands, which includes ill-formed or fused glands. However, there is notable inconsistency among pathologists in recognizing these glands, especially when mixed with pattern 3 glands. This inconsistency has significant implications for patient management and treatment decisions. Conversely, the recognition of glomeruloid and cribriform architecture has shown higher reproducibility. Cribriform architecture, in particular, has been linked to the worst prognosis among pattern 4 subtypes. Intraductal carcinoma of the prostate (IDC-P) is also associated with high-grade cancer and poor prognosis. Accurate identification, classification, and tumor size evaluation by pathologists are vital for determining patient treatment. This review emphasizes the importance of prostate cancer grading, highlighting challenges like distinguishing between pattern 3 and pattern 4 and the prognostic implications of cribriform architecture and intraductal proliferations. It also addresses the inherent grading limitations due to interobserver variability and explores the potential of computational pathology to enhance pathologist accuracy and consistency.

INTRODUCTION

Even with the presence of significant interobserver variation, histopathological grading remains the most trusted method for predicting both the recurrence and mortality of prostate cancer. In addition, it is the principal approach used to determine the appropriate treatment options for patients.1 Several grading systems have been used over time, but since the late 1990s, the Gleason grading system, conceived in 1966 and based on morphological patterns, was gradually adopted worldwide, replacing a multitude of competing grading systems.2 Within low- and intermediate-risk prostate cancer, one of the most critical aspects of histopathological grading is the correct identification and quantification of pattern 4, comprising poorly formed/fused glands as well as glomeruloid and cribriform architecture.3 However, the reproducibility in recognizing poorly formed/fused glands remains poor, especially if poorly formed glands are intermingled with well-formed pattern 3 glands, with important implications for patient management and inclusion in active surveillance protocols. On the other hand, higher reproducibility among uropathologists has been observed in the recognition of glomeruloid and cribriform architecture,4 and the latter has been demonstrated to be associated with the worst prognosis among the different subtypes of pattern 4.5,6 Intraductal carcinoma of the prostate (IDC-P) has also been demonstrated to be associated with high-grade cancer at radical prostatectomy and portends a poor prognosis.7 IDC-P can be indistinguishable from cribriform prostate cancer without immunohistochemistry (IHC) for basal cell markers. Reporting IDC-P in association with invasive cancer constitutes one of the main differences between the International Society of Uropathology (ISUP) and the Genitourinary Pathology Society (GUPS).8 The precise identification and classification of prostate cancer by the pathologist, as well as the evaluation of the tumor’s extension on biopsy, are crucial for determining the appropriate treatment and management of the patient. In this review, we summarize some important concepts on prostate cancer grading with a focus on some key aspects related to low- and intermediate-risk prostate cancer such as the distinction between pattern 3 and pattern 4 and the prognostic implications of cribriform architecture as well as intraductal proliferations; we also discuss about the intrinsic limit of grading represented by interobserver variability and computational pathology as an opportunity to improve pathologist’s accuracy and consistency.

FROM GLEASON SCORE TO GRADE GROUPS

The Gleason grading system, conceived in 1966 and based on the evaluation of architectural patterns, was adopted worldwide since the late 1990 replacing many different grading systems.1 Different studies, over the years, have focused on the prognostic impacts of different tumor architectural growth patterns and their prognostic value. In 2005, it was decided that large cribriform glands should be classified as Gleason grade 4, whereas small cribriform glands could continue to be designated as Gleason grade 3.9 Given the lack of interobserver reproducibility on diagnosing cribriform grade 3 glands, it was then decided to include all cribriform glands within Gleason grade 4, together with glomeruloid architecture, during the 2014 ISUP consensus conference.10 The ISUP also endorsed the five-tier prognostic grade grouping developed by Pierorazio et al.11 in 2013 as per Gleason score ≤6, 3 + 4, 4 + 3, 8, and 9–10 (Table 1); such grading system was also recommended by the World Health Organization (WHO) and the American Joint Committee on Cancer (AJCC) tumor–node–metastasis (TNM) system. As expected, when compared to a four-tier Gleason score grouping (i.e., Gleason scores 6, 7, 8, and 9–10), the five-tiered ISUP grading system showed better prognostic value for both clinical and pathological outcomes following prostatectomy12,13 and pathologists currently report not only the traditional Gleason score but also the corresponding grade group (GG) of the tumor. Besides grading, a very important issue with crucial implications for the inclusion of patients into active surveillance programs is represented by the quantification of tumor involvement on biopsies. In fact, discontinuous tumor involvement (DTI) is a common finding in prostate needle core biopsies. In this regard, a study from Johns Hopkins showed that adding the intervening benign tissue between two separate foci of cancer correlates with findings at radical prostatectomy better than when only neoplastic tissue is added to calculate the percent.14 A meta-analysis showed that compared to additive measurement (with the subtraction of intervening benign tissue), linear measurement (including intervening benign tissue) of DTI predicted more accurately aggressive diseases in the radical prostatectomy.15

Table 1.

Grade groups

GG	Gleason score
1	≤3+3
2	3+4
3	4+3
4	4+4; 3+5; and 5+3
5	4+5; 5+4; and 5+5

GG: grade group

There is a strong association between the total number of positive cores and the percent of tumor involvement with the tumor volume in the gland and outcome after surgery or radiotherapy.16,17

THE IMPORTANCE OF THE IDENTIFICATION AND QUANTIFICATION OF PATTERN 4

The distinction between pattern 3 and pattern 4 is critical since the presence of pattern 4 excludes a patient from the very low- and low-risk groups of the National Comprehensive Cancer Network guidelines.18 Patients in the favorable intermediate-risk group may be eligible for active surveillance, depending on the proportion of the tumor exhibiting pattern 4 and other clinical and imaging factors. The presence and extent of Gleason pattern 4 are critical factors, so recent guidelines stress the importance of consistent and accurate definition and reporting of this parameter.19,20 Both genitourinary pathology organizations GUPS and ISUP recommend indicating the percentage of pattern 4 in GG2 and GG3 carcinoma. This is crucial as the prognosis and treatment approach can differ significantly between patients with GG2 carcinoma with only a small amount of pattern 4, and those with a GG2 tumor nearing GG3 in terms of percentage of pattern 4.21 Similarly, by indicating the percentage of Gleason pattern 4, pathologists can effectively communicate that a GG3 tumor with 90% pattern 4 across several biopsy samples is likely to behave more similarly to a GG4 tumor. On the other hand, it may be appropriate for a patient to choose active surveillance if biopsies show a low volume of GG2 carcinoma with <5% Gleason pattern 4, considering factors such as life expectancy, other health conditions, and risk acceptance. Moreover, consistently indicating the percentage of Gleason pattern 4 can help diminish the perceived extent of discordance between pathologists when they have differing evaluations of cases that hover between GG2 and GG3. In fact, it should be easily comprehensible that two pathologists might occasionally have a valid difference of opinion between a GG2 tumor with 40% Gleason pattern 4 and a GG3 tumor with 60% Gleason pattern 4. Given the potentially high clinical importance of the percentage of pattern 4, it is good practice to keep a conservative approach when in doubt; for example, in a biopsy with a small focus of carcinoma that is predominantly composed of Gleason pattern 4, it is better to point to a diagnosis of 4 + 3 = 7 (GG3) rather than 4 + 4 = 8 (GG4), even if a few Gleason pattern 3 glands are present. Similarly, if a small focus of cancer shows a mixture of pattern 3 and pattern 4 glands, it may be appropriate to avoid rendering the percentage of pattern 4 since such a percentage could be skewed due to the limited amount of neoplastic glands.22 The distinction between Gleason pattern 3 and pattern 4 glands can be very challenging, especially when poorly formed glands are intermingled with well-formed glands. In such common and challenging situations, before classifying a tumor as Gleason pattern 4, it is essential to ensure that the diagnosis is not based on a few tangentially cut, poorly formed glands at the periphery of a tumor focus that is predominantly Gleason pattern 3. Avoiding this error can be achieved by examining sequential levels of the tumor and confirming that such a focus can be identified as well-formed glands. Furthermore, a focus should only be considered as a minimal amount of Gleason pattern 4 if there is a distinct group of poorly formed glands. In this regard, Zhou et al.23 proposed as a criterion the presence of more than ten poorly formed glands that are not immediately adjacent to well-formed glands to define a cancer focus as pattern 4. As a rule, if adjacent well-formed glands can be distinguished from one another (e.g., one can draw a mental circle around each gland), even an arrangement of back-to-back glands should still be considered Gleason pattern 3. One issue in distinguishing pattern 4 from pattern 3 is represented by the presence of mucinous fibroplasia, which can cause architectural distortion, thus complicating the diagnosis and leading to overdiagnosis of pattern 4.9

Acute and chronic inflammation may mimic carcinoma clinically, radiologically, and also histologically as it can trigger changes that mimic high-pattern carcinoma both cytologically and architecturally.

Another related issue is the presence of extracellular, extraluminal mucin. In this case, when faced with a mucinous tumor in the prostate, besides ruling out metastasis, the pathologist must concentrate on grading the tumor according to the architectural arrangement of the tumor glands, rather than being sidetracked by the presence of extracellular material.24 Among Gleason pattern 4, cribriform cancer represents an important subpattern characterized by adverse prognosis and merits a detailed discussion.

CRIBRIFORM CANCER

The term “cribriform”, derived from the Latin word “cribrum” meaning “sieve”, was first introduced in the context of prostate cancer by Gleason to describe the glands composed of sheets of tumor cells forming cohesive, rounded, or irregularly shaped trabeculae with multiple perforations or “punched out” lumina.25 Cribriform glands can be spherical or oblong and may occasionally have irregular edges. Among genitourinary pathologists, the cribriform pattern has the highest interobserver reproducibility compared to other Gleason 4 patterns, although the reproducibility ranges from 54% to 79%, indicating significant variability.4

The ISUP defines cribriform glands as a continuous sheet of connected malignant epithelial cells with multiple glandular lumina that can be easily observed at low power magnification (objective magnification 10×), as shown in Figure 1. These glands should not have any intervening stroma or mucin separating individual or fused glandular structures.26 Shah et al.27 identified certain features indicative of cribriform glands, including transluminal bridging, dense cellular proliferation (cell mass accounting for ≥50% of the glandular lumen), a clear luminal space along the gland’s periphery occupying <50% of the gland’s circumference, absence of intraglandular mucin, and lack of contact between the majority of intraglandular cells with stroma. Conversely, the presence of opposite morphologies and a glomeruloid pattern were identified as features that argue against the diagnosis of cribriform glands.28

Figure 1.

Cribriform pattern 4 prostate carcinoma. (a) Cribriform carcinoma characterized by a proliferation of tumor cells forming multiple lumens imparting a “sieve-like” architecture; (b) absence of basal cells (20×).

All forms of cribriform prostate cancer, irrespective of their size and shape, are currently classified as Gleason pattern 4.10 Increasing evidence suggests that cribriform cancer, found in both biopsies and radical prostatectomies, is linked to unfavorable clinical outcomes. This includes poorer biochemical recurrence-free survival, metastasis-free survival, and cancer-specific survival compared to cases without cribriform cancer.29,30,31

Three significant studies, all conducted by Kweldam et al.,1 have shown important clinical implications for cribriform architecture and/or intraductal carcinoma identified in core biopsy specimens and were summarized in a review published in 2019. One study found that the presence of invasive cribriform or intraductal growth in a biopsy specimen was a better predictor of biochemical recurrence after radical prostatectomy and/or radiation therapy in patients with GG2 prostate cancer than reporting percentage pattern 4.32 In another study, biopsy cases with cribriform/intraductal morphology were associated with worse disease-specific survival compared to those with similar Gleason scores but without these growth patterns.28 In a third study, the same group showed that men with biopsies showing GG2 prostate cancer on core biopsy without cribriform/intraductal carcinoma had similar biochemical recurrence-free survival after radical prostatectomy and radiation therapy as those with GG1 disease.33 These studies have certain limitations. One limitation is that the researchers grouped cribriform carcinoma and intraductal carcinoma together, which made it difficult to accurately assess whether the poor prognosis in cases with cribriform carcinoma was influenced by the simultaneous presence of intraductal carcinoma. In addition, the patient sample was drawn from the European Randomized Study of Screening for Prostate Cancer, which commenced in 1994. In this study, patients only underwent a six-core prostate biopsy, potentially leading to undersampling of high-grade prostate cancer.

The sensitivity to detect tumors with cribriform morphology by multiparametric magnetic resonance imaging (mpMRI) remains controversial; although mpMRI offers high sensitivity for the detection of cribriform morphologies at a lesion level, the cribriform component is frequently missed by targeted biopsies.34 It was shown that second-round targeted biopsy identified 25% more clinically significant cancer (ISUP Grade Group ≥2) in Prostate Imaging–Reporting and Data System (PI-RADS) 3 or higher lesions with initially negative or ISUP Grade Group 1 diagnosis.35 It is therefore conceivable that repeat biopsies should be taken into consideration in cases with suspicious mpMRI and negative or low-grade (GG1 or GG2 without cribriform pattern) lesions at biopsy. In this regard, future studies are needed to better map the heterogeneity within tumors as seen on mpMRI, corresponding to the variation in architecture, in order to refine the identification of important areas and improve the recognition of cribriform morphologies, as well as to define the minimum number of targeted biopsies needed to enhance the detection of such features.

INTRADUCTAL PROLIFERATIVE LESIONS

Cribriform cancer often poses a diagnostic challenge, as it can be difficult to distinguish from atypical intraductal proliferation (AIP) and intraductal carcinoma of the prostate (IDC-P). These entities, along with high-grade prostatic intraepithelial neoplasia (HG-PIN), represent a spectrum of intraductal proliferation characterized by increasing architectural and/or cytological abnormalities (Table 2). Low-grade prostatic intraepithelial neoplasia (LG-PIN) is no longer diagnosed due to low reproducibility among pathologists. HG-PIN is associated with a 20% cancer risk if unifocal and 30%–40% if multifocal in subsequent repeat biopsies. In the 2016 WHO classification, HG-PIN was identified in four main patterns: cribriform, micropapillary, flat, and tufted.36 However, the updated 2022 WHO classification removed the cribriform pattern,37 as most of these lesions are more accurately classified under the AIP category. This reclassification has significant implications, as AIP necessitates immediate repeat biopsy, while HG-PIN, particularly if it is a single focus, can be monitored conservatively.38 On the other hand, there is an increase in the risk up to 75% if more than three cores are involved by HG-PIN.39 Therefore, urologists should take into consideration a repeat biopsy with increased sampling of the area with HG-PIN for patients with three or more cores involved by HG-PIN, when no carcinoma has been detected.

Table 2.

Differential diagnosis between high-grade prostatic intraepithelial neoplasia, intraductal carcinoma of the prostate, and atypical intraductal proliferation

Feature	HG-PIN	AIP	IDC-P
Basal cell present	Yes	Yes	Yes
Focal coagulative necrosis	Absent	Absent	Present
Marked cytological atypia	No	No	Yes
Solid or dense cribriform (>50%)	No	No (loose)	Yes
Regular contour with lumen-spanning proliferation	No	Variable	Yes

HG-PIN: high-grade prostatic intraepithelial neoplasia; AIP: atypical intraductal proliferation; IDC-P: intraductal carcinoma of the prostate

AIP, previously referred to as “atypical cribriform lesion”, “atypical intraductal cribriform proliferation”, and “low-grade intraductal carcinoma”, is defined as an intraductal proliferation of prostatic secretory cells that may occasionally exhibit greater architectural complexity and/or cytological atypia than typical HG-PIN but falls short of the strict diagnostic threshold for IDC-P. A study by Shah et al.27 found that AIP identified in a biopsy without concurrent IDC-P was associated with one or more adverse pathology features (defined as ≥GG3, IDC-P, cribriform carcinoma, and pT3a/pT3b) in a subsequent biopsy or radical prostatectomy. Furthermore, when comparing the ETS-related gene (ERG) and phosphatase and tensin homolog (PTEN) status between AIP and IDC-P, there was a comparable frequency of ERG overexpression and PTEN loss among AIP, IDC-P, and invasive carcinoma, indicating that AIP also exhibits molecular characteristics of IDC-P,40 supporting the concept that AIP may represent the low-grade spectrum of IDC-P.

From a morphological perspective, over 90% of AIP cases exhibit loose cribriform architecture (meaning there are more luminal spaces relative to the epithelium) without intraluminal necrosis or severe nuclear atypia (Figure 2). Other characteristics include a solid or dense cribriform structure that does not completely span the glandular lumen.41 Generally, the term “AIP” encompasses any lesion, irrespective of its architecture, that shows significant nuclear atypia or pleomorphism beyond HG-PIN but does not meet the current diagnostic criteria for IDC-P.

Figure 2.

Atypical intraductal proliferation (AIP). (a) Glands with intraluminal proliferation exhibiting loose cribriform architecture with morphological features worse than HG-PIN but falling short of intraductal carcinoma; (b) preserved basal cells (20×). HG-PIN: high-grade prostatic intraepithelial neoplasia.

The most widely used definition of IDC-P, endorsed by the WHO classification, is based on the following architectural and/or cytological features: solid or dense cribriform architecture (>50% epithelial component relative to lumens), comedonecrosis, or marked atypical nuclei, as shown in Figure 3.37

Figure 3.

Intraductal carcinoma of prostate (IDC-P). (a) Expansile glands with dense cribriform proliferation with irregular contour; (b) preserved basal cells (20×).

There is now strong evidence that IDC-P is associated with poor outcomes and patients with prostate cancer that includes IDC-P are generally not suitable for active surveillance. When invasive cancer is present, IDC-P continues to be an independent predictor of various negative outcomes, including early biochemical recurrence, distant metastasis, and poorer disease-specific survival.42 While most instances of isolated IDC-P in needle biopsies are indicative of IDC-P with an unsampled aggressive invasive prostate cancer, there are rare cases that may represent precursor IDC-P with favorable outcomes. The treatment approach for patients whose prostate biopsies only show IDC-P without an associated invasive component is a subject of debate. Some specialists advocate for radical therapy, while others suggest a prompt repeat biopsy. In this regard, emerging evidence suggests that IDC-P can follow two distinct biological pathways, despite having morphologically similar characteristics.43 The majority of IDC-P cases are associated with high-grade, high-volume prostate cancer and are considered late events where preexisting prostate carcinoma spreads retrogradely into and colonizes benign nonneoplastic ducts and acini. Conversely, a small fraction of IDC-P cases are found either without prostate carcinoma or with low-grade carcinoma. These are believed to be in situ lesions, which evolve from HG-PIN precursors. Finally, from a molecular point of view, a growing body of evidence suggests that genomic instability may be one of the critical causal factors for IDC-P development, implicating that breast cancer gene 2 (BRCA2) mutations can promote the development of IDC-P.44,45 For these reasons, germline BRCA2 testing has been formally recommended when IDC-P is present by the Philadelphia Prostate Cancer Consensus Conference.46

GRADING AND REPORTING INTRADUCTAL CARCINOMA OF THE PROSTATE

As previously mentioned, the presence of IDC-P is an independent adverse prognostic factor typically associated with worse outcomes and, if an accompanying invasive lesion is present, with higher Gleason scores.28 Consequently, there is international consensus on the importance of reporting IDC-P in both biopsy and radical prostatectomy specimens. However, despite its well-established prognostic significance, the impact of IDC-P on the Gleason score of the lesion remains a subject of ongoing debate.

The only point of agreement among urological societies (ISUP and GUPS) is in the rare (<1%) instances where IDC-P is present alone without invasive cancer.20 In such cases, both ISUP and GUPS recommend performing IHC stains for basal markers (e.g., P63 and CK34BE12) to rule out the presence of an invasive component. If “pure” IDC-P is confirmed, both societies exclude this proliferation from lesion grading.19,20 Often, the presence of IDC-P alone in biopsies indicates unsampled invasive cancer, prompting the recommendation for repeat biopsies.

In cases where IDC-P coexists with invasive carcinoma, the 4^th WHO classification did not recommend including IDC-P in the grading of prostate carcinoma.36 However, urological societies have since developed differing opinions and approaches to grading IDC-P associated with invasive carcinoma (Table 3).47

Table 3.

Difference between Genitourinary Pathology Society (GUPS) and International Society of Uropathology (ISUP) in managing and reporting intraductal carcinoma of the prostate

Issue	GUPS	ISUP
Reporting the presence of IDC-P	Yes, always	Yes, always
IDC-P pure without invasive cancer	Not to be graded both in biopsy and radical prostatectomy	Not to be graded both in biopsy and radical prostatectomy
IDC-P admixed with invasive cancer	Only grade invasive component, do not consider IDC-P	Incorporate IDC-P in grading the entire lesion
Performing basal cell IHC to discriminate IDC-P (in case of presence of invasive cancer)	Necessary only in biopsy when the result would impact the GG	No, because IDC-P is graded anyway

IHC: immunohistochemistry; IDC-P: intraductal carcinoma of the prostate; GG: grade group

According to the ISUP guidelines,20 IDC-P associated with invasive carcinoma should be incorporated into the Gleason score, significantly influencing the resulting GG and clinical impact.48,49 This choice impacts cancer grading, particularly the quantification of Gleason pattern 4, resulting in different GG assignments (3 + 4 vs 4 + 3). Many studies support incorporating cribriform pattern and IDC-P into grading evaluations, as it appears to enhance outcome prediction, especially for disease-specific survival and metastasis-free survival.28,33,50 As a result of incorporating IDC-P into the Gleason score, ISUP does not recommend routine basal-cell immunostaining to distinguish IDC-P from invasive cancer, as it is graded regardless. Immunostaining should be limited to confirming “pure” IDC-P cases, where the lesion is not graded.

Conversely, GUPS recommends to not include IDC-P in the Gleason score, whether present alone or coexisting with invasive cancer.19 As already stated, IDC-P may represent either retrograde intraductal spread of invasive carcinoma or a precursor lesion. In the former case, the invasive portion should share features with the intraductal component; in the latter case, as a precursor lesion, it is inadequate for grading. Supporting this theory are studies showing discordant ERG and PTEN expression in IDC-P and low-grade invasive carcinoma in the same prostate,51 indicating different origins for these populations. A recent study demonstrated that including IDC-P in the Gleason score changed GG in only 1.6% of biopsies and 0.6% of radical prostatectomies.49,52 GUPS recommends performing IHC stains for basal markers to better discriminate the invasive nature of the lesion when the result would impact GG, especially for differential diagnosis between cribriform IDC-P and Gleason pattern 4 cancer.19

The 5^th WHO classification has not yet decided which of these two systems to adopt.37 Therefore, it recommends specifying in the report the Gleason grading variant used during the evaluation to avoid confusion.52,53

THE PROBLEM OF INTEROBSERVER VARIABILITY IN PROSTATE CANCER GRADING

One of the main problems of prostate cancer grading is the interobserver reproducibility of the Gleason score. Indeed, it is well demonstrated that there is a high interobserver variability both in the identification and in the quantification of different Gleason patterns. A recent multicenter study showed that there is only moderate agreement for primary Gleason pattern, total Gleason score, and the GGs system. Lower agreements were observed in mixed or general pathologists’ cohorts, demonstrating the necessity of long expertise to correctly recognize these lesions.54

Pattern 4 appears to be the most problematic pattern to evaluate, probably due to its variability. In one of the largest studies on this issue, the mean agreement on Gleason score among 337 members of “European Network of Uropathology (ENUP)” was 71.4% for Gleason score 3 + 3 = 6 and 56.4% for Gleason score 3 + 4 = 7.55 Moreover, the different morphologic types of pattern 4 (poorly formed glands, fused glands, glomeruloid, and cribriform) are associated with different levels of interobserver reproducibility.4

Although the percentages of concordance change in different studies, all the results suggest that glomeruloid and cribriform patterns are easier to identify than fused or poorly formed glands.56 In this regard, McKenney et al.57 demonstrated that the consistency in distinguishing between tangentially sectioned and poorly formed glands is only moderate. This can have important implications in clinically significant scenarios such as inclusion in active surveillance protocols.

Some criteria based on quantitative and topographic features have been proposed to standardize and increase the reproducibility of this pattern;23 however, in the end, it is a decision based on each pathologist’s criteria for ruling out the chance of tangential sectioning. It must be kept in mind that morphological evaluation in histopathology is hampered by the limit of human interpretation, despite standardization efforts. To this end, efforts to create novel initiatives for establishing international standard for the interpretation of pathology specimens like the one proposed by Egevad et al.58 are extremely important.

Besides the identification of pattern 4, it is also very important to estimate the extension and the percentage of it, given the clinical implications involved. Unsurprisingly, this visual method lacks precision, and when compared to digital measurement, it has shown a difference up to 2 mm in 76.6% of the cases.59 Moreover, pathologist visual estimation seems to over-estimates the pattern 4 proportion compared to the digital method, with a higher discordance in case of increased amount of pattern 4.

Overall, numerous research works have investigated how expertise affects the accuracy of histological reports by pathologists. When general pathologists and specialist pathologists evaluated the same biopsy samples, the specialists provided more precise and consistent results. Similarly, the quality of a biopsy is greatly influenced by the quality of the material collected. Future research should explore the reasons behind inconsistencies in sampling.60

TOWARD AUTOMATED PROSTATE CANCER GRADING: THE PROMISE OF ARTIFICIAL INTELLIGENCE (AI) AND COMPUTATIONAL PATHOLOGY

AI is a technology that employs learning models known as artificial neural networks, which are designed and trained to analyze the data similarly to the human mind. The most advanced AI technologies utilize machine/deep-learning algorithms capable of self-learning and improvement.

In recent years, numerous studies have developed neural network-based AI systems trained with prostate specimen scans previously diagnosed by the panels of international expert uropathologists. While more data are needed, the results of these studies suggest a wide range of potential applications in this field. These algorithms have been calibrated with high diagnostic sensitivity for cancer identification, resulting in many atypical glands being overdiagnosed and subsequently reviewed by pathologists, ensuring safe diagnostic screening of benign lesions.61,62

The first algorithm demonstrating good accuracy in identifying prostate cancer was developed in 2016 by Litjens et al.63 Despite the relatively limited data size (225 slides), prostate cancer was identified in all slides, and 30%–40% of slides with benign tissue were distinguished from suspicious ones. Later, in 2019, another group developed an algorithm trained on a large dataset (12 132 slides) capable of identifying various lesions (prostate cancer or breast metastasis) with a sensitivity of 97.7% and specificity of 99.3% for prostate carcinoma.64 In a recent review, Morozov et al.65 analyzed the diagnostic accuracy of AI in prostate cancer identification from 24 different studies, finding accuracy ranging from 83.7% to 98.3%, high sensitivity (87%–100%), and good specificity (68%–99%). As a result of these findings, the USA Food and Drug Administration (FDA) authorized the marketing of “Paige Prostate” in 2021, a software capable of detecting areas suspicious for cancer with a sensitivity of 97.7%, allowing pathologists to review highlighted areas. According to the FDA, AI-assisted pathologists improved cancer detection in digitized prostate slides by 7.3% compared to unassisted pathologists, reducing false negatives by 70% and false positives by 24%.66

Beyond cancer identification, another AI application is measuring lesion extension. Many algorithms trained for this purpose have shown good accuracy, with a level of agreement comparable to the interobserver reproducibility of expert uropathologists.67

Perhaps, the most promising application of AI in prostate cancer is grading. Many developed algorithms have shown promising results in both cancer detection and grading, sometimes comparable to international expert pathologists. The model trained by Ström et al.62 demonstrated good grading accuracy, with a concordance within the range of the experienced ISUP pathologists reference panel. Bulten et al.68 performed even better with an algorithm trained on 5759 biopsies, showing high agreement with the reference standard. During the the Prostate cANcer graDe Assessment (PANDA) challenge, 1290 developers used 10 616 digitized prostate biopsies to train the algorithm, achieving an agreement of 0.868 with a validation set of expert uropathologists.69 As these studies have shown, such tools could help address the major issue of prostate cancer grading reproducibility. Indeed, Bulten et al.70 demonstrated that AI-assisted pathologists performed better than both unassisted pathologists and AI alone in grading prostate carcinoma.

In the not-too-distant future, deep-learning systems could assist pathologists by screening specimens, improving reproducibility, recognizing and measuring neoplastic areas, providing second opinions, and evaluating other parameters (e.g., mitosis, stromal reaction, and biochemical recurrence).

CONCLUSIONS

Prostate cancer diagnosis hinges on the accuracy of histopathological grading, with the Gleason grading system at its core. This system’s emphasis on discerning pattern 4, especially the elusive ill-formed or fused glands, poses a significant challenge for pathologists. When these glands intermingle with pattern 3 glands, the differentiation becomes even more complex, leading to potential discrepancies in treatment recommendations. A more consistent marker in this grading landscape is the presence of cribriform structures and glomeruloid formations. The cribriform pattern, in particular, stands out for its association with the most severe prognosis among pattern 4 subtypes. The detection of IDC-P further underscores the importance of the issue, given its consistent link to high-grade tumors and poorer patient outcomes. As we navigate the complexities of prostate cancer grading, the implications of these differentiation challenges on clinical decisions become increasingly evident. However, the advent of computational pathology offers a promising solution. By leveraging technological advancements, we can aim for a future where prostate cancer grading is both precise and consistent, paving the way for optimal treatment strategies for the benefit of patients.

AUTHOR CONTRIBUTIONS

VA wrote the manuscript. EM wrote and reviewed the manuscript. Both authors read and approved the final manuscript.

COMPETING INTERESTS

Both authors declare no competing interests.