Article Text

Download PDFPDF

Virtual microscopy for histology quality assurance of screen-detected polyps
  1. Mauro Risio1,
  2. Gianni Bussolati2,
  3. Carlo Senore3,
  4. Stefano Vigna2,
  5. Elena Frangipane1,
  6. Nereo Segnan3,
  7. Paola Cassoni2
  1. 1Unit of Pathology, IRCC, Candiolo-Torino, Italy
  2. 2Department of Biomedical Sciences and Human Oncology, University of Torino, Torino, Italy
  3. 3AOU S Giovanni Battista, SCDO Epidemiologia, CPO Piemonte, Torino, Italy
  1. Correspondence to Professor Paola Cassoni, Department of Biomedical Sciences and Human Oncology, University of Torino, Via Santena 7, Torino 10126, Italy; paola.cassoni{at}unito.it

Abstract

Aim Histology quality assurance is crucial for screening programmes and can be performed by circulating glass slides, which has certain intrinsic disadvantages. The present study aimed to assess the accuracy of virtual microscopy in terms of reliability and diagnostic reproducibility in colorectal cancer screening programmes.

Methods 457 consecutive lesions detected in people undergoing colonoscopy were examined histologically in two pathology units, using both traditional optical microscopy and virtual microscopy (6–12 months later). Intra- and inter-observer agreement using the two approaches was determined using κ statistics.

Results Intra- and inter-observer agreements were substantially unmodified by the use of the virtual microscopy approach compared with traditional optical microscopy; moreover, for some histological features critical for patient management in colorectal cancer screening programmes (such as the presence of a villous component within the adenoma), virtual microscopy increased interobserver agreement (κ statistics 0.66 versus 0.52).

Conclusions This study shows that virtual microscopy can be an effective tool for diagnostic quality assurance in colorectal cancer screening programmes, and its accuracy is equivalent to or higher than that of optical microscopy in the validation of histological criteria (eg, advanced adenoma) crucial for patient management in screening programmes.

  • Colorectal cancer screening
  • quality assurance
  • virtual microscopy
  • diagnostic agreement
  • diagnostic screening

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Introduction

Quality assurance (QA) and quality control of a programme are among the principles of population screening and are also thought to increase the quality of management of symptomatic patients.1 External QA programmes for pathology, together with standards and quality indicators, have been implemented in the NHS National Bowel Cancer Screening Programme, using glass slide circulation or electronic images on the web (http://www.virtualpathology.leeds.ac.uk). Moreover, the introduction of QA programmes has also been recommended in the recent European guidelines for colorectal cancer (CRC) screening.2 They focus on verifying the diagnostic concordance of the pathological features that are crucial for patient management: distinguishing adenomatous from non-adenomatous large bowel polyps; defining advanced adenomas (size, ‘villousness’, grade of dysplasia); identifying cancerised polyps.3

Telepathology systems could be used to circulate microscopic images in quality control efforts in screening, and also to obtain a second opinion on selected cases, given their cost-effectiveness advantages over preparation, distribution and circulation of glass slides. Improvements in image digitisation technology have led to significant progress in telepathology. However, the main disadvantage of static telepathology—only being able to have one image at a time for any given magnification—means that the observer cannot obtain a complete overview of the tissue specimen. This, in turn, may hamper the circulation of static telepathology in daily practice, and may bias interobserver agreement with respect to glass slide diagnosis.4 Conversely, digital virtual microscopy organises the acquisition process, by scanning the whole histology slide at selected resolutions, and provides the observer, through real-time image compression, with either the overview image or a series of microscopic images derived by zooming in on well-defined histological sectors.5 6 The system should also lessen the user's workload. This study aimed to assess the accuracy of virtual microscopy in terms of reliability and diagnostic reproducibility in large-scale CRC screening programmes.

Materials and methods

Patients and histology

People undergoing CRC screening in the context of a population-based screening programme, performed January to December 2007, were considered for the study. All residents of the Piedmont Region are invited at age 58 for sigmoidoscopy screening offered once in a lifetime. We included in the analysis 457 consecutive patients examined in this regional programme in two endoscopy units in Turin. All biopsy and polypectomy specimens were examined histologically in two pathology units, identified as A and B (IRCC and San Giovanni Battista Hospital, respectively).

Histological classification of polyps and cancers was in accordance with WHO criteria.7 Cancer was defined as the invasion of malignant cells beyond the muscularis mucosa. Advanced adenoma was defined as an adenoma with any of the following features: size ≥10 mm, high-grade dysplasia, villous component ≥20%. Polyps with intramucosal carcinoma or carcinoma in situ were included in the group of advanced adenomas. Serrated adenomas were diagnosed when nuclear abnormalities featuring low-grade or high-grade dysplasia were seen within the serrated epithelium, whereas serrated polyps without dysplasia were categorised as hyperplastic polyps/sessile serrated lesions; this group included, besides typical hyperplastic polyps, serrated lesions displaying patchy or diffuse architectural distorsions mainly in the deeper sectors of the mucosa.2

The study was approved by the local ethics committees of the institutions involved.

Optical and virtual microscopy

In both units, an initial diagnosis was performed on all samples using traditional optical microscopy on H&E-stained slides. After 6–12 months, the same pathologists re-evaluated the cases using a virtual microscopy approach. Briefly, histology images from all slides were acquired using the dedicated software Olyvia (http://www.olympus4u.com/product/images/slide/sub/olyvia.GIF). An archive of all the lesions was then created on a PC to store the images. Each case could then be analysed on screen and directly magnified during examination, as necessary, using a joystick.

All pathologists were blinded to the initial diagnosis as well as to the diagnoses made in the other histology unit using both optical and virtual microscopy.

Diagnostic agreement

To determine whether virtual microscopy is a reliable alternative to optical microscopy, the diagnoses performed using the two approaches were compared in order to validate intraobserver agreement, comparing optical and virtual microscopy-based diagnoses for each pathologist. In addition, interobserver agreement was determined by comparing the concordance achieved using (in both units) optical or virtual approaches. In both units, all cases were first diagnosed using optical microscopy and then reviewed by the same pathologist 6–12 months later using the digital approach.

Statistical analysis

Statistical analysis was performed using κ statistics, which reflects the agreement between two measurements (two observers or a single observer over separate measures of the same sample), after removal of chance agreement, as a measure of reliability; 95% CI was calculated using the method proposed by Fleiss.8 9

As suggested by Landis and Koch,8 κ values <0.2 were reconsidered to indicate poor agreement, 0.21–0.40 fair, 0.40–0.60 moderate, 0.61–0.80 good, and >0.80 very good agreement.

Results

Intraobserver and interobserver agreement

Advanced adenomas

The total number of screen-detected lesions was 457. The distribution of all the diagnoses by centre and by diagnostic approach is presented in table 1 for centre A and table 2 for centre B, crossing optical and virtual microscopy data.

Table 1

Distribution of the reported diagnoses on 457 screen-detected polyps in the pathology unit A

Table 2

Distribution of the reported diagnoses on 457 screen-detected polyps in the pathology unit B

By optical microscopy, 105/457 polyps (23%) were diagnosed as advanced adenomas in centre A and 112 (24.5%) in centre B. By virtual microscopy, the number of advanced adenomas was 108 (23.6%) in centre A and 117 (25.6%) in centre B. Therefore a slight increase in the number of advanced adenomas was observed in both centres using the digital approach.

The intraobserver agreement for advanced adenoma diagnosis (optical approach first and virtual reading used for re-viewing) was good in both units: 0.76 (95% CI 0.69 to 0.83) and 0.81 (95% CI 0.74 to 0.87), respectively, in centre A and B (table 3).

Table 3

Intraobserver agreement (first diagnosis using optical microscopy, second using virtual microscopy) in 457 screen-detected polyps

The interobserver agreement comparing diagnoses was similar using either optical (κ statistics: 0.63; 95% CI 0.55 to 0.70) or virtual (κ statistics: 0.65; 95% CI 0.57 to 0.73) microscopy (table 4).

Table 4

Interobserver agreement using either optical or virtual microscopy in 457 screen-detected polyps

High-grade dysplasia

By optical microscopy, high-grade dysplasia was reported in 40/457 (8.7%) cases in centre A and 99/457 (22%) cases in centre B. By virtual microscopy, high-grade dysplasia diagnoses were increased up to 54/457 (11.8%) cases in centre A and remained substantially unmodified (101/457, 22.1%) in centre B.

The κ statistics value for intraobserver agreement was 0.70 (95% CI 0.59 to 0.81) and 0.78 (95% CI 0.71 to 0.85) for the A and B centre, respectively (table 3).

The interobserver agreement was similar using optical (κ statistics: 0.42; 95% CI 0.32 to 0.53) or virtual (κ statistics: 0.39; 95% CI 0.28 to 0.49) microscopy (table 4).

Villous component (tubulovillous and villous adenomas)

In the A centre, the presence of a villous component was outlined in 95 (20.8%) and 81 (17.7%) cases, respectively, using either optical or virtual microscopy. In centre B, the cases with a villous component were 57 (12.5%) using optical microscopy and 62 (13.6%) using virtual microscopy. The κ statistics value for intracentre agreement was 0.78 (95% CI 0.70 to 0.85) and 0.89 (95% CI 0.83 to 0.96) for the A and B centre, respectively (table 3).

The interobserver agreement was improved from moderate to good by the use of virtual microscopy (κ statistics: 0.66; 95% CI 0.57 to 0.76) compared with the optical approach (κ statistics: 0.52; 95% CI 0.41 to 0.62) (table 4).

Serrated polyps without dysplasia (hyperplastic polyps/sessile serrated lesions)

A diagnosis of serrated polyp without dysplasia was reached using optical microscopy in 167/457 (36.5%) and 153/457 (33.4%) screen-detected lesions in centre A and B, respectively. The intraobserver agreement was very good in both centres, with κ statistics of 0.87 (95% CI 0.82 to 0.91) and 0.88 (95% CI 0.83 to 0.92) in centre A and B, respectively (table 3).

The interobserved agreement was also found to be good by both optical (κ statistics: 0.77; 95% CI 0.71 to 0.83) and virtual (κ statistics: 0.79; 95% CI 0.73 to 0.85) microscopy (table 4).

Discussion

Several retrospective and prospective studies have shown that patients with advanced adenoma (ie, >1 cm, with villous component and/or high-grade dysplasia) have a higher risk of developing synchronous/metachronous advanced adenomas or invasive CRC. Therefore these histological features determine surveillance colonoscopy of patients and drive the organisation guidelines of screening programmes.10 11 Histological examination of advanced adenoma is therefore required to be objective, consistent and reproducible.12 The results of this investigation indicate substantial agreement in the histological identification of advanced adenomas, which is consistent with previous studies in which finer histological categories were collapsed into broader diagnostic categories (ie, any villous component >20% versus tubular and low-grade versus high-grade dysplasia).13–16 In the same restricted diagnostic domain, virtual microscopy shows the same reproducibility, for both intraobserver agreement (κ=0.76–0.81) and interobserver agreement (κ=0.63–0.65), establishing it as equally reliable in screening programmes.

Grading of dysplasia is based on architectural changes (glandular crowding and budding, back-to-back glands) supported by nuclear and cytological features (nuclear enlargement, stratification, stainability, loss of cell polarity, atypical mitoses), therefore high-grade dysplasia is often suspected at low magnification.2 Although the diagnostic grade is determined from the most severely dysplastic area, no studies have dealt with extension of such an area within the adenomatous tissue: it is conceivable that at least two glands should be involved by high-grade dysplasia to make them identifiable at low magnification.2 7 The different importance assigned to nuclear dysplastic abnormalities (nuclear stratification, in particular) may explain the significantly different detection levels observed in this study between the two centres (40/457 and 54/457 cases versus 99/457 and 101/457 cases for optical and virtual microscopy, respectively). As previously demonstrated, when dysplasia grading is determined by architectural criteria, based on large areas, high grade is more easily and frequently identifiable at a lower magnification.14 Conversely, focusing on nuclear abnormalities in smaller areas and even barely identifiable epithelial tracts can fulfil the criteria for high-grade dysplasia, lowering the detection rate.3 This seems to be further biased by virtual microscopy, in that suboptimal fine tuning of motorised Z stage could generate images with different focus level, mimicking nuclear stratification with increased stainability and may explain the higher virtual detection of high-grade dysplasia with respect to optical microscopy in one of the two centres involved in the study (+35%, see table 3).

Detection of villous growth was found to be slightly more reproducible with virtual microscopy than with optical microscopy (κ=0.66 vs 0.52). The entire virtual slide can be viewed on screen, facilitating both identification of the villous profile and quantification of any villous component in comparison with the polyp outline. The opposite is true for optical microscopy, which allows only small radians of polyps to be separately examined, even at the lowest magnification. Under these circumstances, villous assessment is likely to be made by recalled pattern recognition, rather than heuristic procedures, similarly to the telepathology diagnosis of colonic polyp histotype.4 17 In accordance with Terry et al,15 and at variance with what we have observed in this study for dysplasia grading, the level of agreement for adenoma ‘villousness’ was linked to polyp size, being higher in the largest ones (≥1 cm) both under the microscope and in virtual images (data not shown). As evidence is lacking that villous growth in a small adenoma could be predictive of metachronous neoplasia risk, it would seem advisable to focus the pathological diagnosis of screen-detected polyps on dysplasia grade, an independently reproducible feature, rather than villous growth.12 18

Dysplasia can affect the serrated epithelium of hyperplastic polyps, featuring mainly right-sided colonic polyps, the sessile serrated lesions and serrated adenomas, in the former architectural abnormalities being uncoupled, with nuclear and cytological dysplastic changes found in the latter.19 Whereas sessile serrated lesions2 only display patchy or diffuse distortions of tissue organisation consistent with ‘architectural dysplasia’ in the deepest portion of the mucosa (sessile serrated adenomas, in accordance with Snover et al19), in traditional serrated adenomas, nuclear and cytological frankly dysplastic features are seen within the serrated epithelium, besides the architectural ones. Intermingling and/or intermediate features of sessile serrated lesion and hyperplastic polyp can often be seen in the same histological section, mainly in small polyps, impairing diagnostic reproducibility, and likewise in superficial or tangentially cut lesions, where correct assessment of the deepest portions of the mucosa is impossible.20 Moderate levels of concordance for hyperplastic and sessile serrated lesions have been previously reported, by either internet-based static telepathology or glass slide microscopy.21 22 It should be taken into account, however, that the prevalence of sessile serrated lesions in a patient undergoing colonoscopy is 1.9%, and that they represent 7–15% of serrated polyps.23 It has been estimated that only 8.3% of the polyps previously diagnosed as hyperplastic polyps would now be reclassified as sessile serrated lesions, and such a level of diagnostic accuracy is felt acceptable for clinical decision-making.22 The present investigation indicates that both intraobserver and interobserver agreement are good when hyperplastic polyps and sessile serrated lesions are collapsed into a single group, and virtual microscopy does not influence the values of diagnostic reproducibility. On the whole, given the absence of robust evidence demonstrating the risk of tumour progression associated with serrated architectural abnormalities, single categorisation of serrated polyps without nuclear dysplasia produces good levels of diagnostic reproducibility and seems advisable in screening programmes.

The small number of adenocarcinomas diagnosed in this investigation does not allow any inference to be drawn about the diagnostic reproducibility of such lesions. The underestimation may be ascribable to the poor sensitivity (22.2%) of the biopsy-based histopathology in identifying neoplastic invasion of the submucosa.24

This suggests that the biopsy-based treatment strategy for lesions unsuitable for endoscopic resection should include surgery when the biopsy samples are categorised as suspicious for submucosal invasion.25

In conclusion, the results of this study show that virtual microscopy can be an effective tool for diagnostic QA in CRC screening programmes, and its accuracy is equivalent to or higher than that of optical microscopy in the validation of histological criteria (ie, advanced adenoma) crucial for screening patient management. The time necessary to scan slides may limit the realistic applicability of virtual microscopy in high-volume series: however, an automated (operator-independent) slide-scanning system which can run overnight could solve the problem and result in crucial time saving, avoiding the need to restructure personnel resources.

Take-home messages

  • External quality assurance for pathological examination in large-scale screening programmes can be impaired by practical problems related to circulating of glass slides.

  • Virtual microscopy could be an effective, accurate tool for diagnostic quality assurance in colorectal cancer screening programmes, and can even improve interobserver agreement on certain histological features (such as adenoma ‘villousness’) that are crucial for screening patient management.

Acknowledgments

This work was supported by Regione Piemonte (Ricerca Sanitaria Finalizzata 2006).

References

Footnotes

  • Competing interests None.

  • Ethics approval This study was conducted with the approval of the CPO Piemonte.

  • Provenance and peer review Not commissioned; externally peer reviewed.