Introduction

Ex vivo fluorescence confocal microscopy (FCM) is a novel optical technology that allows the examination of freshly excised tissue to provide a quasi-histological view of the tissues in few minutes. FCM is based on the combination of two types of lasers to allow the examination of tissue specimens using the reflectance mode (785 nm) that depends on the difference of the refractive indices of various cellular structure, and the fluorescence mode (488 nm) that utilizes a contrast agent to allow the visualization of microstructures [20].

To date, FCM has been used in healthy skin [12] as well as in cutaneous lesions such as melanoma [10, 11], basal cell carcinoma, and squamous cell carcinoma [9, 17]. Furthermore, it was used for diagnosis of cutaneous inflammatory skin lesions [3]. Moreover, Ragazzi M et al. [26] examined the use of FCM on different surgical specimens such as breast, lymph node, thyroid, and colon, demonstrating that neoplastic tissues were easily distinguishable from normal structures.

The main advantage of the FCM is its ability to provide quasi-histological images that are similar to hematoxylin and eosin (H&E) images, which render the interpretation of images easier for pathologist novice to this technology.

Prostate cancer (PCa) is predicted to be the most commonly diagnosed male cancer in the USA in 2019 [29]. The gold standard diagnosis of PCa is based on the histopathological examination of prostatic tissues. Infiltrative growth pattern, absence of basal cell layer, and nuclear atypia (nuclear enlargement and prominent nuclei) represent the main pathological criteria for PCa diagnosis [19]. However, histopathological examination of tissues is a time-consuming procedure that may require up to 2 days in the highest quality pathological laboratories [1]. Currently, there is an increased interest in “real-time” pathological examination of prostatic tissues either during prostate biopsies or during radical prostatectomy [28, 31]. In this setting, several options were introduced for “real time” pathological examination of prostatic and periprostatic tissues including frozen section [28], light reflectance spectroscopy [21], multiphoton microscopy [30], confocal laser endomicroscopy [18], video-rate fluorescence-structured illumination microscopy [32], and FCM [24].

The aim of the current study is to create an atlas for the FCM images demonstrating its ability to identify different architectural structures in prostatic specimens (such as muscle, nerves, vessels, and adipose tissue) with differentiation of inflammatory changes from malignant transformation. Furthermore, among the important factors, determining the feasibility and superiority of any new imaging technology is the accurate and easy interpretation of its images. This can be demonstrated by assessing the learning curve of this new technology [15]. Thus, the learning curve for FCM images interpretation was evaluated as a secondary aim.

Material and methods

Patients

This study was approved by the Ethical Committee (protocol number 0018091/18) and written informed consent was obtained from all patients. Clinical data and pathological features were anonymously stored. We evaluated 80 prostatic biopsies of 18 gauge in thickness, obtained from radical prostatectomy (RP) specimens of 10 consecutive patients with clinically localized PCa undergoing robotic-assisted laparoscopic prostatectomies (RALP) using Da Vinci Surgical System (Intuitive Surgical, Inc., Sunnyvale, CA, USA).

Specimens and staining procedure

The fresh, no formalin fixed, biopsies were obtained within 10 min from the removal of the prostate, and were separately stained in a 0.6 mM solution of acridine orange dye (Sigma-Aldrich®) for 30 s then washed in physiological saline solution. All biopsies were placed between two glass slides sealed with silicon glue and then positioned onto the FCM stage for image acquisition. Following the acquisition, the samples were covered with a thin sponge, poned in biocassettes, and immersed in 10% neutral buffered formalin for 24-h fixation in order to maintain the proper orientation of the samples at the further histopathological examination.

Instrument

The FCM VivaScope® 2500 M-G4 (Mavig GmbH, Munich, Germany; Caliber I.D.; Rochester NY, USA) has two different lasers. In this study, both reflectance (785 nm) and fluorescence (488 nm) modes were used for each sample examination. Briefly, FCM has maximum examination depth of 200 μm, a vertical resolution of up to 4 μm, a magnification of × 550 and a maximum scan size of 25 × 25 mm [20]. However, the penetration depth can be increased by modifying the laser power and/or the incubation period of the specimen. Moreover, both sides of the glass slides can be examined by the FCM allowing deeper examination of the specimen. The reconstructed image is a collection of mosaic images (square-shaped images of 1024 × 1024 pixels). The laser filter has a × 38, 0.85 numerical aperture water immersion objective lens. The softwares VivaScan® (Version 11.0.1140 Mavig GmbH; Caliber I.D), VivaBlock®, and VivaStack® enable respectively the reconstruction of the images from the probes, the acquisition of multiple images in the X/Y directions within a single plane at a fixed depth, and a survey of multiple frames along the Z axis, visualizing deeper tissue. The grayscale fluorescence and reflectance mosaics were digitally stained with color and combined. Since pathologists interpret hematoxylin and eosin (H&E) staining, the purple and pink colors were used to improve association with cellular nuclei and non-cellular structures, respectively [8]. In this study, a self-written ImageJ software (https://imagej.nih.gov/ij/) was used to the standard digital image staining approach. Zoom capabilities enable an enhanced visualization of cell morphology details. The staining and imaging processes of each specimen were completed within 5 min.

FCM images’ evaluation

All FCM images were stored in a dedicated database. Two pathologists (general pathologist and an expert genitouro-pathologist) from two different centers (University of Modena and Reggio Emilia and University of Ancona) were integrated in the assessment of the FCM learning curve. The images were randomly displayed to each of the two pathologists while being blinded to the histopathologic diagnoses. After more than 90 days, they were asked to re-evaluate the specimens. All examinations were performed independently and included the evaluation of the tumor and non-neoplastic tissue comprised in the samples with particular regard to normal prostatic glands and soft tissue components (muscle bundles, nerve, adipocytes, vessels) as well as other non-neoplastic conditions such as inflammation and benign glandular hyperplasia.

Histopathological examination

From the biopsies included in paraffin and processed according to the standard protocol [27], we performed 3 consecutive sections of 3 μm thickness, starting from the most superficial portion without discarding material. The sections were H&E stained.

Diagnostic criteria for carcinoma included atypical acini arranged in one or more patterns of growth, combined with atypical cytologic findings including prominent nucleolus or pale enlarged and irregular nuclei, were useful criterions for cancer, particularly when small, irregular, abortive acini with primitive lumens and absence of basal layer were visible. The tumor grade groups were assessed according to International Society of Urological Pathology (ISUP) grading [6]. Immunohistochemical analysis included racemases/34betaE12 cytokeratin cocktail stain and were used for 8 specimens with suspicious small foci of cancer detected on H&E-stained slides.

H&E-stained slides were compared with the images obtained by FCM techniques in order to describe overlaps or discordant features.

Statistical analysis

The Cohen’s kappa (κ) statistic was used to measure the agreement between histopathologic diagnoses and the FCM diagnosis for each pathologist in each reading. We selected κ statistic as the measure of agreement as our variable of interest is binary. Kappa is a measure of this difference, standardized to lie on a − 1 to 1 scale, where 1 is perfect agreement, 0 is exactly what would be expected by chance, and negative values indicate agreement less than chance, i.e, potential systematic disagreement between the observers [5].

The Cohen’s kappa can be interpreted as following; (κ < 0) is less than chance agreement, (κ = 0.01 to 0.20) represents slight agreement, (κ = 0.21 to 0.40) is fair agreement, (κ = 0.41 to 0.60) is moderate agreement, (κ = 0.61 to 0.80) is substantial agreement, while, (κ = 0.81 to 0.99) is considered almost perfect agreement. The interpretation of reproducibility is marginal (κ = 0.00 to 0.40), good (κ = 0.40 to 0.75), and excellent (κ > 0.75) [7].

The process of deciding whether the numerical results quantifying hypothesized relationships between variables, are acceptable as a description of the data, is known as validation. We select k-Fold Cross-Validation, as it reduces the risk of errors induced by bias by utilizing the full-data set for training and for validation. In K Fold cross validation, the data is divided into k subsets. Now the holdout method is repeated k times, such that each time, one of the k subsets is used as the test set/validation set and the other k-1 subsets are put together to form a training set. The error estimation is averaged over all k trials to get total effectiveness of our model [13].

Results

FCM atlas

Samples from the biopsies were classified as non-neoplastic tissue (NNT), high-grade prostatic intraepithelial neoplasia (HG-PIN), and prostatic acinar adenocarcinoma (PAA).

Non-neoplastic prostatic and extraprostatic tissues

In some samples, probably taken from the periphery of the prostatic gland, a mix of connective and soft adipose tissue, crossed by thin muscle bundles, were detected as well as bundles of nerves (Fig. 1a–c). Chronic prostatitis typically shows an aggregation of lymphocytes and plasma cells within the stroma (Fig. 1d). The different components of prostatic and periprostatic soft tissue were easily recognized at the corresponding FCM images (Fig. 1e–h). NNT included prostatic glands lined by typical two cell layers represented by an outer layer of low cuboidal cells and an inner layer of tall columnar mucin-secreting epithelium, organized into non-infiltrating pattern of growth. The glands are surrounded by a thin connective tissue and, sometimes, contained corpora amylacea (Fig. 2a, b). As shown in Fig. 2d, e, in FCM mosaics, different cell layers of the prostatic glands as well as corpora amylacea were appreciable. Hyperplastic prostatic glands with taller and enlarged columnar secretory cells, with pale to clear cytoplasm and enlarged nuclei without atypia, are visible in the samples. The glands are evidently surrounded by basal cells (Fig. 2c, f).

Fig. 1
figure 1

Histological images of non-neoplastic extraprostatic muscle bundles (a), adipose tissue (b), nerve (c) and prostatic inflammatory infiltrate (d), and FCM correlates (e–h). (Scale bar, 100 μm)

Full size image
Fig. 2
figure 2

Histological images of prostatic glands (a), corpora amylacea (b) and hyperplastic prostatic glands (c), and FCM corresponding images (d–f). (Scale bar, 100 μm)

Full size image

High-grade prostatic intraepithelial neoplasia and prostatic acinar adenocarcinoma

HG-PIN included glands composed of crowded and irregularly spaced multi-layered epithelial cells, with enlarged, hyperchromatic and pleomorphic nuclei, with prominent nucleoli. Basal cells surrounded the glands (Fig. 3a). At FCM, we observed the prominent nucleolus and the stratified epithelium surrounded by evident basal cells (Fig. 3e).

Fig. 3
figure 3

Histological images and the corresponding FCM images of high-grade prostatic intraepithelial neoplasia (a–e) and prostatic acinar adenocarcinoma grade group 1 (b–f), grade group 2 (c–g), and grade group 3 (d–h). (Scale bar, 100 μm)

Full size image

PAA was composed of closely packed irregular glands variably in size and shape and with back-to-back distribution. Few or no interposing stroma was intermingled through the atypical glands. Enlarged round, hyperchromatic nuclei and prominent nucleolus are visible. Basal cells were absent. These tumors were graduated as grade group 1 [GG1] (Fig. 3b). Tumor composed by predominantly well-formed glands and exhibiting a lesser component of cribriform glands were classified as grade group 2 [GG2] (Fig. 3c); whereas the cribriform glands were predominant; the tumors were classified as grade group 3 [GG3]. When sheet of undifferentiated cell and poorly formed/fused glands were detected, we classified the tumors as grade group 4 [GG4] (Fig. 3d). The corresponding FCM images demonstrated the atypia of the nuclei, the irregular morphology of the glands, as well as the typical pattern of growth of the glands (Fig. 3f–h). Figure 4 a and b emphasize the infiltrating growth of PAA within a nerve and the atypia of the glands with particular regard among the prominent nucleolus and the absence of the basal cell layer. These features were well represented by the corresponding FCM images (Fig. 4c, d).

Fig. 4
figure 4

Histological images and FCM correlates of infiltrating growth of PAA within a nerve (a–c) and atypical glands with prominent nucleolus and the absence of the basal cell layer (b–d). (Scale bar, 50 μm)

Full size image

Learning curve assessment

Both pathologists achieved very encouraging levels of agreement between FCM evaluation and the gold standard histopathological diagnosis of prostatic biopsies (Table 1). The percentages of agreement become higher from the first evaluation 86% and 92% (κ = 0.68 and κ = 0.79 substantial agreement) to the second one, reaching 95% (κ = 0.87 almost perfect agreement) for both the pathologists. Furthermore, the reproducibility was excellent (κ > 0.75) for both raters in the second evaluation. Besides high accuracy, Table 1 also indicates a promising result in terms of specificity and sensitivity. Out of 59 negative biopsies, 50 and 57 have been correctly identified as negative by the first pathologist in the first (specificity 84.7%) and second evaluations (specificity 96.6%), respectively. For the second pathologist, 58 negative biopsies have been correctly identified both in the first and the second evaluations which correspond to 98.3% specificity. On the other hand, 19/21 positive biopsies have been correctly classified by the first pathologist in both evaluations which leads to a sensitivity value of 90.5%. The second pathologist correctly identified 16/21 positive samples during the first evaluation and 18/21 in the second evaluation (sensitivity 76.2% and 85.7% respectively). Moreover, the area under the receiver operating characteristic (ROC) curve, that describes the relationship between the sensitivity and specificity, becomes higher from the first to the second evaluation (from 0.87 to 0.92–0.93, respectively) for both raters.

Table 1 First and second FCM evaluation of prostatic biopsies compared to histopathological diagnoses, the percentage of correct diagnoses, κ value, the level of agreement, the sensitivity, the specificity, and ROC area for both raters
Full size table

Discussion

Surgeons show growing interest in the real-time pathological examination as it may potentially allow a more precise and patient-tailored surgery [25]. In PCa, real-time pathological examination plays an important role as it may be used for assessing the surgical margins (apical, bladder neck, and neurovascular bundle) during radical prostatectomy thus allowing complete resection of the tumor without compromising the functional outcomes [18, 22, 30]. It can be also used during prostate biopsy to decrease the need for additional biopsies [31]. Furthermore, it has been used in the histopathological screening for prostate cancer in organ donors, especially with the increasing age of organ donors [2]. The gold standard technique for real-time pathological examination is the frozen section, which, is debatable due to its associated drawbacks including the technical difficulty, the complexity of the procedure that requires multiple dedicated personnel, and being resource and time-consuming technique [22]. Other options include light reflectance spectroscopy [21], multiphoton microscopy [30], confocal laser endomicroscopy [18], and video-rate fluorescence structured illumination microscopy [32]. Furthermore, we have previously proved that FCM has a diagnostic accuracy of 91% compared to the standard histopathological examination with AUC, sensitivity, and specificity of 0.884, 83.33%, and 93.53% for the pathological examination of prostatic tissues [24], which render FCM as a promising option for “real-time” pathological examination of prostatic tissue.

In the current study, FCM was able to provide a rapid, and high-resolution quasi histological images of the freshly excised, non-formalin fixed prostatic and periprostatic specimens. The primary aim of our study was to create an atlas for the FCM images for prostatic and periprostatic tissue examination; however, throughout the process of collecting representative images for the atlas, FCM demonstrated a promising results in identifying cellular and nuclear characters of prostatic and periprostatic tissues with almost perfect agreement (κ = 0.87) compared to the gold standard histopathological examination after a short learning curve. Furthermore, it was able to accurately differentiate benign tissues and inflammatory changes from malignant transformation in the prostatic specimens.

However, we did not assess the potentiality of FCM to assess the specimens grade group; FCM was able to identify the tumor growth pattern of specimens used for creating this atlas. Currently, we have an ongoing study that will assess the ability of FCM to identify the group grade.

The learning curve is among the most important limitations of any new technology. It is important to examine the learning curve of image interpretation before any new cellular imaging technology is introduced in the clinical practice [14, 16]. In these setting, we evaluated the learning curve of FCM images. We intentionally integrated a general pathologist and an expert genitourinary pathologist in the evaluation of the learning curve to provide a more realistic assessment. We hypothesized that the great similarity between the FCM images and the gold standard H&E histopathological examination will render them easily interpretable by pathologists novice to FCM. Our results supported this hypothesis, since the agreement between the FCM reading and the histopathological diagnosis increased from 86 to 92% in the first reading for the first and the second pathologists, respectively, to reach 95% for both pathologists in the second reading. Noteworthy, 8 samples (10%) required immunohistochemistry for diagnosis, which, may be responsible for the persistent 5% disagreement obtained by both pathologists on the second reading. Interestingly, Bertoni et al. [4] demonstrated that FCM is capable of providing immunofluorescent images of special markers using fluorochrome-conjugated antibodies, which may enhance the diagnostic performance of FCM of doubtful cases; however, in their study, they used paraffin-fixed sections and not freshly excised specimens. Panarello D et al. [23] reported that the learning curve for the confocal laser endomicroscopy images interpretation seems to be short; however, this was not assessed in their study.

FCM provides several advantages including the H&E-like images which facilitate the interpretation of images, the fast preparation of specimens and rapid image acquisition (less than 5 min per specimen in our experience), and the preservation of the specimens’ integrity not only for subsequent histopathological examination but also for ancillary studies like immunohistochemistry [25]. Furthermore, FCM represents a step forward towards digitalized pathology, as the specimen preparation is simple and can be performed by the treating surgeon, and it provides a digital image that can be sent online to a remote pathologist for interpretation. Despite the promising level of agreement between FCM evaluation and the histopathological examination of prostatic and periprostatic tissues, there were some pitfalls (Supplementary Fig. 1); one of the specimens with prostatic acinar adenocarcinoma on the H&E was misdiagnosed as HG-PIN due to the presence of mild nuclear atypia associated with crowded nuclei in the FCM images. Moreover, adenocarcinoma cells were misdiagnosed as a result of the presence of technical artefact that masqueraded mild atypia in another specimen. Finally, in one specimen, a focus of atypical grade group 4 single cells were misdiagnosed as inflammatory cells.

The main limitation of the current study is the small sample size. On the other hand, our study has many strengths as it is the first study (to our knowledge) to provide a comprehensive description of the FCM images of the prostatic and periprostatic tissue and to assess the learning curve of this technology in the field of PCa.

Conclusion

FCM is an optical technology that can be used for “real-time” pathological examination of prostatic tissue. This atlas can serve as the base for urologists and pathologists for learning and interpreting FCM images of prostatic and periprostatic tissues. Further studies are required to define the potential applications of FCM in PCa.