ABSTRACT
Background
Whole‐slide imaging (WSI) is a promising tool in pathology. However, the use of WSI in cytopathology has lagged behind that in histology. We aimed to evaluate the utility of WSI for the intraoperative touch imprint cytological diagnosis of axillary sentinel lymph nodes (SLNs) in breast cancer patients.
Methods
Glass slides from touch imprint cytology of 480 axillary SLNs were scanned using two different WSI scanners. The intra‐ and interobserver concordance, accuracy, possible reasons for misdiagnosis, scanning time, and review time for three cytopathologists were compared between WSI and light microscopy (LM).
Results
A total of 4320 diagnoses were obtained. There was substantial to strong intraobserver concordance when comparing reads among paired LM slides and WSI digital slides (κ coefficient ranged from 0.63 to 0.88, and concordance rates ranged from 94.58% to 98.33%). Substantial to strong interobserver agreement was also observed among the three cytopathologists (κ coefficient ranged from 0.67 to 0.85, and concordance rates ranged from 95.42% to 97.92%). The accuracy of LM was slightly higher (average of 98.06%) than that of WSI (averages of 96.81% and 97.78%). The majority of misdiagnoses were false negative diagnoses due to the following top three causes: few cancer cells, confusing cancer cells with histiocytes, and confusing cancer cells with lymphocytes.
Conclusions
This study is the first to address the feasibility of WSI in touch imprint cytology. The use of WSI for intraoperative touch imprint cytological diagnosis of SLNs is a practical option when experienced staff are not available on‐site.
Keywords: breast cancer, concordance, sentinel lymph nodes, touch imprint cytology, whole‐slide imaging
1. Introduction
In recent decades, whole‐slide imaging (WSI), a newer digital imaging technology, has become popular and has developed rapidly. WSI technology is the digitization (scanning) of a glass slide with a whole‐slide scanner, which is then used to generate a digital slide that can be viewed with imaging software that simulates light microscopy (LM) [1]. WSI technology has several advantages over LM, such as portability, ease of sharing and retrieval of archival images, and the ability to make use of computer‐aided diagnostic tools [2]. The United States Food and Drug Administration approved the use of a WSI system in primary surgical pathology diagnosis in recent years [3]. Therefore, WSI is a promising tool to be used in pathology [4, 5, 6, 7, 8, 9]. However, the use of WSI in cytopathology has lagged behind that in histology. Histological sections have a relatively uniform and even surface, while cytological glass slides are more difficult to digitize due to areas of variable thickness and the presence of 3‐dimensional groups of cells and possible obscuring material (e.g., blood, mucus, neutrophils, or ultrasound gel) [10, 11]. Despite the studies that are now beginning to accrue regarding cytopathology as a use case for WSI, the evidence is considered immature relative to that for surgical pathology [12].
The study of the application of WSI in cytopathological specimen assessment is limited [10]. Most efforts and considerations to date have been concerned mainly with the use of WSI combined with artificial intelligence to assess cervical specimens [10, 13, 14, 15]. Recently, some studies have been concerned about the use of WSI for non‐gynecological cytology [10, 16, 17]. The practice of rapid on‐site evaluation (ROSE) for fine‐needle aspiration biopsy samples, imprint cytology, and brush cytology using digital pathology has been reported [18, 19, 20]. Perez et al. in 2021 investigated the utilization of dynamic telecytopathology for touch imprint cytology of needle core biopsies [21]. However, the use of WSI in touch imprint cytology of axillary sentinel lymph nodes (SLNs) has not been reported. In this study, we aimed to evaluate the utility of WSI for the intraoperative touch imprint cytological diagnosis of axillary SLNs in breast cancer patients. The advantage of this kind of cytological specimen is that the diagnosis is relatively easy because only the status of SLN metastasis needs to be determined and no obscuring material is present (e.g., blood, mucus, neutrophils, or ultrasound gel). Additionally, a reviewer's own subjective variability regarding the diagnosis is minimal. We compared the concordance between the WSI and LM results using two different whole‐slide scanners and three cytopathologists with different levels of seniority. The viability of WSI assessment in intraoperative scenarios was evaluated. We also analyzed the possible reasons for misdiagnosed cases and issues with attention when viewing WSI digital slides.
2. Materials and Methods
2.1. Case Selection
Four hundred and eighty glass slides of touch imprint cytology of axillary SLNs were selected from 358 patients with breast cancer who underwent surgical resection at Fudan University Shanghai Cancer Center, a large tertiary cancer center in China in 2022. We set two groups with 240 cases each in this study. The first 240 cases were called Group A and were selected randomly from the overall database of SLNs. Group A represented the spectrum and proportion of diagnoses to be encountered in practical application [12] and was used to evaluate concordance and accuracy. The other 240 cases, called Group B, one‐half of which were selected from the negative database, and the other half from the positive database based on the original diagnosis. Group B, with an artificially increased positive rate of 50%, was set for misdiagnosis analysis to find more potential problems and reasons for misdiagnoses during WSI usage because the majority of misdiagnoses were false negative diagnoses of positive SLNs. Clinical and pathological data, including age, sex, treatment, histopathological diagnosis of breast cancer, and histopathological diagnosis of SLNs, were obtained from medical records. The detailed process of touch imprint cytology of SLNs was described previously [22]. The study was approved by the ethics committee of the hospital.
2.2. WSI
All glass slides were scanned using two different scanners, HAMAMATSU Nano Zoomer S360 and MoticEasyScan Pro 6‐FS, at 20× resolution in a single z‐plane to form digital files. Before scanning, the ink marks on the glass slides were erased. The scanner operators checked the integrity and clarity of the digital images during the scanning phase; then, the three pathologists rechecked the digital images during the reviewing phase. The slides were rescanned if the digital image was blurred, the images were incomplete, or for other reasons that might affect diagnosis. The scanning time and frequency of rescanning were recorded.
2.3. Viewing Glass and Digital Slides
A total of 480 glass slides and 960 paired digital slides were reviewed by three pathologists (the P1, P2, and P3) engaged in cytology, namely, two senior pathologists (P1 and P2) and one junior pathologist (P3). All cytopathologists had experience using WSI digital slides and were blinded to the original reported results and other clinical and pathological data. The glass slides were reviewed via LM first. Then, WSI digital slides scanned by Scanner 1 (S1) or Scanner 2 (S2) were reviewed after a washout period of at least 2 weeks [12], followed by reviewing WSI digital slides with the other scanner after at least 4 weeks. All digital imaging files were copied to the computers of the three cytopathologists and reviewed offline. The diagnosis and time spent reviewing were recorded.
Therefore, every case had 10 diagnoses, including 4 LM diagnoses (original LM, P1 LM, P2 LM, and P3 LM) and 6 WSI diagnoses (P1S1, P2S1, P3S1, P1S2, P2S2, and P3S2). The original LM diagnosis was made by a variety of cytopathologists during real‐world operation. The cytological staff knew the clinical and other pathological data in advance and were permitted to discuss them during diagnosis. Except for the original LM diagnosis, other diagnoses were made independently and blindly.
2.4. Final Cytological Diagnosis
The final cytological diagnosis of “positive” or “negative” was made according to the histological diagnosis if it was consistent with the cytological diagnosis. The histological diagnosis of the SLNs was evaluated according to the 8th edition of the American Joint Committee on Cancer TNM staging system for breast cancer. If the cytological diagnosis was inconsistent with the histological diagnosis, the cases were discussed by at least three cytopathologists to make the final cytological diagnosis based on consensus. The SLNs with a final cytological diagnosis of “atypia” indicated difficult cases for which it was challenging to reach a consensus after discussion. In this situation, the cytopathologists insisted that some cells were problematic, even though the histological diagnoses were negative, which might have been caused by the removal of atypical cells when sections were cut from formalin‐fixed paraffin‐embedded (FFPE) tissues.
2.5. Concordance and Accuracy
Concordance and accuracy were calculated based on Group A data. Intraobserver and interobserver concordance were calculated and reported as a κ coefficient and a percentage of concordance. The accuracy of each diagnosis was calculated based on the final cytological diagnosis. The misdiagnosed cases were reviewed by the three cytopathologists, and the reasons for misdiagnosis were analyzed based on Group B data.
2.6. Statistical Analysis
The SPSS 20.0 package (IBM, Chicago, IL, USA) and Microsoft Excel (Microsoft Cooperation, Redmond, Washington, USA) were used for statistical analyses and scientific graphing. Statistical significance between different groups was calculated using paired t‐tests for the analysis of time spent reviewing. The κ coefficient was calculated for concordance of diagnosis. A κ value of 0.00–0.20 indicated slight agreement, 0.21–0.40 fair agreement, 0.41–0.60 moderate agreement, 0.61–0.80 substantial agreement, and 0.81–1.00 strong agreement [23]. Statistical significance was defined as a p < 0.05.
3. Results
3.1. Clinicopathological Features
In this study, 480 glass slides and 960 matched digital slides of touch imprint cytology of axillary SLNs from 358 patients with breast cancer were evaluated. All of the patients were female and between 24 and 87 years of age, with a median age of 52 years. Among them, 10 patients received neoadjuvant therapy. Invasive carcinoma of no specific type accounted for the majority of the primary breast cancer diagnoses. The histopathological diagnoses of breast cancer and SLNs are summarized in Table 1. The median diameter of the SLNs was 0.85 cm, ranging from 0.1 to 3.5 cm. The final cytological diagnoses of SLNs were 221 (92.08%) negative and 19 (7.92%) positive in Group A and 119 (49.58%) negative, 116 (48.33%) positive, and 5 (2.08%) atypia in Group B.
TABLE 1.
Histopathological diagnosis of breast cancer and SLNs.
| Histopathological diagnosis of breast cancer | Number of Group A patients (N = 228) (%) | Number of Group B patients (N = 130) (%) |
|---|---|---|
| Invasive carcinoma of no specific type | 179 (78.51%) | 107 (82.31%) |
| Microinvasive carcinoma | 10 (4.39%) | 8 (6.15%) |
| Ductal carcinoma in situ | 12 (5.26%) | 6 (4.62%) |
| Encapsulated papillary carcinoma | 0 (0%) | 2 (1.54%) |
| Invasive solid papillary carcinoma | 7 (3.07%) | 2 (1.54%) |
| Solid papillary carcinoma in situ | 6 (2.63%) | 1 (0.77%) |
| Invasive lobular carcinoma | 6 (2.63%) | 1 (0.77%) |
| Mucinous carcinoma | 4 (1.75%) | 3 (2.31%) |
| Invasive micropapillary carcinoma | 3 (1.32%) | 0 (0%) |
| Carcinoma with apocrine differentiation | 1 (0.44%) | 0 (0%) |
| Histological diagnosis of SLNs a | Number of SLNs of Group A (N = 240) (%) | Number of SLNs of Group B (N = 240) (%) |
|---|---|---|
| Negative | 223 (92.92%) | 131 (54.58%) |
| Macrometastases b | 10 (4.17%) | 87 (36.25%) |
| Micrometastases c | 7 (2.92%) | 21 (8.75%) |
| Isolated tumor cells d | 0 (%) | 1 (0.42%) |
Abbreviation: SLN, sentinel lymph node.
The histological diagnosis of the SLNs was evaluated according to the 8th edition of the American Joint Committee on Cancer TNM staging system for breast cancer.
Metastasis greater than 2.0 mm.
Metastasis greater than 0.2 mm and/or more than 200 cells, but none greater than 2.0 mm.
Isolated tumor cells are defined as small clusters of cells not greater than 0.2 mm, single tumor cells, or a cluster of fewer than 200 cells in a single histologic cross section.
3.2. Scanning and Consumed Time
All 480 SLNs (Group A and Group B) were included in this part of the analysis. The frequency of rescanning for S1 was 7.09% and that for S2 was 2.29%. The scanning time of S1 was 2.93 min/slide at 20× resolution, whereas that of S2 was 5.24 min/slide at the same resolution. The length of time spent reviewing glass slides and WSI digital slides by the three cytopathologists were compared (Figure 1). The average time spent reviewing glass slides and the S1 and S2 WSI digital slides was 1.26 min/slide, 2.02 min/slide, and 2.50 min/slide, respectively.
FIGURE 1.
Time (min/slide) spent reading glass slides and S1 and S2 WSI digital slides. *p < 0.05. LM, light microscope; P1, Pathologist 1; P2, Pathologist 2; P3, Pathologist 3; S1, Scanner 1; S2, Scanner 2. [Color figure can be viewed at wileyonlinelibrary.com]
3.3. Intraobserver Concordance
All three cytopathologists reviewed all the slides associated with the 240 SLNs from Group A. The κ value of each pathologist between LM versus S1 WSI, LM versus S2 WSI, and S1 versus S2 WSI was over 0.6 (range, 0.63–0.88) (Figure 2A). The intraobserver analysis showed substantial to strong concordance for LM and WSI for both scanners. Intraobserver concordance rates were over 90% (range, 94.58%–98.33%) in all comparisons (Figure 2B).
FIGURE 2.
Intra‐ and interobserver concordance of Group A. κ coefficient (A) and concordance rate (B) of intraobserver concordance. κ coefficient (C) and concordance rate (D) of interobserver concordance. LM, light microscope; P1, Pathologist 1; P2, Pathologist 2; P3, Pathologist 3; S1, Scanner 1; S2, Scanner 2. [Color figure can be viewed at wileyonlinelibrary.com]
3.4. Interobserver Concordance
In this part, the results of Group A were analyzed. The interobserver κ coefficients for LM and WSI using the two scanners between P1 versus P2, P1 versus P3, and P2 versus P3 were over 0.6 (range, 0.67–0.85), indicating substantial to strong agreement (Figure 2C). Similarly, the interobserver concordance rate among P1, P2, and P3 was over 95% (range, 95.42%–97.92%) (Figure 2D).
3.5. Accuracy
All accuracy rates were over 95% and are summarized in Figure 3. The accuracy of LM was the highest (average of 98.06%, range from 97.08% to 99.17%), followed by S2 (average of 97.78%, range from 97.08% to 99.17%), and S1 (average of 96.81%, range from 96.25% to 97.50%). The accuracies of P1 (senior) and P2 (senior) were slightly higher than that of P3 (junior).
FIGURE 3.
Accuracy rates of the diagnoses from Group A. LM, light microscope; P1, Pathologist 1; P2, Pathologist 2; P3, Pathologist 3; S1, Scanner 1; S2, Scanner 2. [Color figure can be viewed at wileyonlinelibrary.com]
3.6. Misdiagnosis Analysis
The misdiagnosed SLN cases were analyzed based on Group B. Every SLN case had 10 diagnoses: original LM, P1LM, P2LM, P3LM, P1S1, P2S1, P3S1, P1S2, P2S2, and P3S2. A total of 2400 diagnoses were obtained using both glass slides and WSI digital slides. Compared with the final cytological diagnosis, there were 181 misdiagnoses (181/2400, 7.54%) from 49 SLNs in total. The distribution of misdiagnoses is shown in Figure 4 and Table 2. The majority of them were false negative diagnoses (87.85%).
FIGURE 4.
Distribution of 49 misdiagnosed SLNs of Group B. LM, light microscope; O, original diagnosis; P1, Pathologist 1; P2, Pathologist 2; P3, Pathologist 3; S1, Scanner 1; S2, Scanner 2. [Color figure can be viewed at wileyonlinelibrary.com]
TABLE 2.
Summary of the misdiagnoses in Group B.
| Misdiagnosis situation | Number of misdiagnoses N = 181 | Percentage of misdiagnosis % | |
|---|---|---|---|
| Diagnosis a | Final cytological diagnosis | ||
| Negative | Positive | 88 | 48.62% |
| Atypia | Positive | 50 | 27.62% |
| Negative | Atypia | 21 | 11.60% |
| Positive | Atypia | 11 | 6.08% |
| Atypia | Negative | 9 | 4.97% |
| Positive | Negative | 2 | 1.10% |
| Total false negative diagnoses | 159 | 87.85% | |
| Total false positive diagnoses | 22 | 12.15% | |
Diagnoses including original LM, P1LM, P2LM, P3LM, P1S1, P2S1, P3S1, P1S2, P2S2, and P3S2.
The misdiagnosed cases were reviewed by the three cytopathologists, and the reasons for the misdiagnosis were analyzed. One misdiagnosed case might have several possible reasons for misdiagnosis. The cytopathologists identified 14 possible reasons for misdiagnosis, and the percentage of misdiagnoses of each reason is shown in Table 3 and Figure 5. The top three reasons were few cancer cells, confusing cancer cells with histiocytes, and confusing cancer cells with lymphocytes (e.g., centrocytes, centroblasts, and immunoblasts). “Few cancer cells” means that the diameter of the cancer cell groups was less than 2 mm. Compared to LM, WSI was more likely to result in a misdiagnosis due to few cancer cells and cancer cells near the edge of the slide or image. The percentage of misdiagnoses by using WSI was slightly higher than that by using LM in cases with the following characteristics: crowded and overlapping cells and small cancer cells. The color deviation of staining affected LM and WSI viewing differently.
TABLE 3.
Percentages of possible reasons for misdiagnoses in Group B.
| Possible reasons | Overall misdiagnoses (n = 49) | Misdiagnoses using LM (n = 35) | Misdiagnoses using WSI (n = 44) |
|---|---|---|---|
| Few cancer cells | 65.31% | 65.71% | 70.45% |
| Confusing cancer cells with histiocytes | 51.02% | 54.29% | 54.55% |
| Confusing cancer cells with lymphocytes | 36.73% | 40.00% | 36.36% |
| Small cancer cells | 26.53% | 28.57% | 29.55% |
| Crowded and overlapping cells | 24.49% | 25.71% | 27.27% |
| Scattered cancer cells | 16.33% | 17.14% | 15.91% |
| Degeneration of cancer cells | 14.29% | 20.00% | 15.91% |
| Light staining | 14.29% | 17.14% | 13.64% |
| Subjective misinterpretation | 10.20% | 11.43% | 6.82% |
| Atypical and controversial cases | 10.20% | 11.43% | 11.36% |
| Cancer cells near the edge of the slide or image | 8.16% | 5.71% | 9.09% |
| Cancer cells in a region of cellular sparseness | 4.08% | 5.71% | 4.55% |
| Staining gray | 2.04% | 0.00% | 2.27% |
| Staining too red | 2.04% | 0.00% | 2.27% |
FIGURE 5.
False negative diagnosed cases. (A) Scattered cancer cells confused with histiocytes. (B) Cancer cells confused with germinal center lymphocytes. (C) Small cancer cells. (D) Crowded and overlapping cells, which are difficult to diagnose even with LM (hematoxylin and eosin, original magnification ×200). [Color figure can be viewed at wileyonlinelibrary.com]
4. Discussion
Many studies have reported on the concordance of pathological diagnoses rendered by WSI compared with LM [10, 24]. For example, Goacher et al. [24] summarized the diagnostic intraobserver concordance (including 38 studies, almost all were histological specimens) ranging from 63% to 100% (κ coefficient range, 0.48–0.87) and the diagnostic interobserver concordance ranging from 84% to 100%. The weighted mean diagnostic concordance of WSI and LM was 92.4%, and the weighted mean κ coefficient was 0.75, signifying substantial agreement. In the guidelines of the College of American Pathologists, cytology cases are excluded [12]. In 2020, Girolami et al [10] showed that the overall concordance (including 19 cytology studies) between the diagnosis based on WSI and the original diagnosis was 84.1%. The intraobserver concordance was reported to range from 77.5% to 100% (mean, 92.5%), and the κ coefficient ranged from 0.44 to 0.93 (mean, 0.66). The interobserver concordance with virtual slides varied, with κ ranging from 0.57 to 0.82 (mean, 0.69). The concordance showed a wide range in different studies.
We compared the concordance between LM and WSI results using two different whole‐slide scanners. The intraobserver and interobserver analysis showed substantial to strong concordance for LM and digital images among the three pathologists, with κ coefficients ranging from 0.63 to 0.88 and concordance rates ranging from 94.58% to 98.33%. The concordance of LM versus S2 had slightly higher agreement than LM versus S1 for all three pathologists. The accuracy analysis showed that both P1 and P3 had the highest accuracy with LM and S2 equally. The accuracy rates for S2 were slightly higher than those for S1, differing by 1%–2%. Therefore, the accuracy of the WSI, especially S2, was very close to that of LM.
False negative diagnosis accounted for 87.85% of misdiagnoses. We analyzed, via reviewing misdiagnosis slides, several possible reasons for the misdiagnoses: few cancer cells, confusing cancer cells with histiocytes, confusing cancer cells with lymphocytes, and so on (for details, see Table 3). Some cases had more than one reason, making diagnosis more difficult. In the literature regarding LM diagnoses, similar reasons are also mentioned. Low‐grade carcinoma, lobular carcinoma (characterized by small and dispersed cells), carcinoma with only micrometastases or isolated tumor cells, misinterpretation of histiocytes, endothelial cells and large lymphoid cells, and specimen sampling were common causes [25, 26, 27, 28].
In our study, we calculated these reasons by frequency according to LM or WSI separately. The conditions of few cancer cells and cancer cells near the edge of the slide or image were more likely to cause misdiagnosis by using WSI compared to LM, which indicates that some viewing fields were likely missed in WSI viewing. Therefore, it was recommended to ensure that all cells on the glass slide were completely captured by scanners before reading and to check the reading track on the software to avoid overlooking diagnostic regions in the WSI before submitting a diagnosis.
Many studies have shown that 3‐dimensional groups of cells affect the utility of WSI in cytology [10, 11, 29]. To our surprise, crowded and overlapping cells resulted in minor differences between WSI and LM with misdiagnosis percentages of 27.27% and 25.71% in our study, respectively. One possible reason is that in addition to crowded and overlapping cells, diagnostic cells could be seen in other regions of the image, so the WSI slides could still yield a correct diagnosis. Second, 3‐dimensional groups affect slide reading by both WSI and LM. Some 3‐dimensional groups were too thick and difficult to see clearly, even using LM (see Figure 5D).
The color deviation of staining affected LM and WSI viewing differently in our study. Previous research has demonstrated that surgical pathologists do not rely primarily on color to render accurate diagnoses of breast biopsy cases but rather use architectural features of tissue and cellular morphology to reach a diagnostic conclusion [30]. However, in cytological specimens, the architectural features are not complete or are even missing. The staining color displayed by LM or WSI was more important for diagnosis in cytological specimens than in histological specimens. The color normalization in WSI in various color appearances is expected to help decrease discordance and misdiagnosis [31].
Scanning time and viewing time were other parameters analyzed in this study. The average time for scanning was 2.93 min/slide and 5.24 min/slide at 20× resolution using two different scanners. Scanning at 40× resolution or Z‐stack mode required more time, which may not be suitable in intraoperative scenarios. The average time spent viewing WSI digital slides was slightly longer than that spent viewing glass slides. S1 digital slides needed shorter scanning and viewing times than S2 slides. The pathologists mentioned that the software of S1 was easier to operate than that of S2. After the first slide finished being scanned, the scanning and viewing slides could be processed simultaneously, which would save time in real‐world scenarios. Therefore, the time necessary to scan and view slides by WSI was acceptable for intraoperative standards.
The current study has some limitations. First, we used at least a 2‐week washout period between LM and WSI reading, but we could not radically remove recall bias. Furthermore, WSI by offline mode cannot simulate real intraoperative conditions. The turnaround time may be affected by computer and network performance. Last but not least, acceptance of digital pathology among pathologists and scanner costs are barriers to its implementation.
5. Conclusions
This study is the first to address the feasibility of WSI in touch imprint cytology. The study included three reviewers and two scanners to assess the difference between WSI and LM. According to our results, LM outperformed WSI in terms of diagnostic accuracy and cost, particularly the time to review slides. Use of WSI for intraoperative imprint cytological diagnosis of SLNs is a practical option in locations where experienced staff are not available on‐site. More studies are needed on cytological specimens to improve WSI guidelines and standards for the subspecialty of cytology.
Author Contributions
Fei Ren and Huange Li: investigation, data acquisition and analysis, writing – original draft, and writing – review and editing. Wentao Yang: scanning, data acquisition and analysis, writing – original draft, and writing – review and editing. Ying Chen, Yuwei Zheng, Hao Zhang, Bo Ping, Xiaochun Wan: reviewing controversial cases, data analysis, and writing – review and editing. Shuling Zhou: reviewing the histopathology, and writing – review and editing. Peng Shi: data analysis, statistical analysis, and writing – review and editing. Yanli Wang: conceptualization, methodology, data acquisition and analysis, validation, supervision, writing – original draft, and writing – review and editing.
Conflicts of Interest
The authors declare no conflicts of interest.
Acknowledgments
This work was supported in part by Shanghai Municipal Key Clinical Specialty (shslczdzk01301) and Shanghai Science and Technology Development Fund (19MC1911000).
Funding: This work was supported in part by Shanghai Municipal Key Clinical Specialty (shslczdzk01301) and Shanghai Science and Technology Development Fund (19MC1911000).
Fei Ren, Huange Li, and Wentao Yang contributed equally to this work.
Data Availability Statement
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.
References
- 1. Dietz R. L., Hartman D. J., Zheng L., Wiley C., and Pantanowitz L., “Review of the Use of Telepathology for Intraoperative Consultation,” Expert Review of Medical Devices 15 (2018): 883–890. [DOI] [PubMed] [Google Scholar]
- 2. Pantanowitz L., Sinard J. H., Henricks W. H., et al., “Validating Whole Slide Imaging for Diagnostic Purposes in Pathology: Guideline From the College of American Pathologists Pathology and Laboratory Quality Center,” Archives of Pathology & Laboratory Medicine 137 (2013): 1710–1722. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. U.S. Food and Drug Administration , “FDA News Release: FDA Allows Marketing of First Whole Slide Imaging System for Digital Pathology,” 2017, https://www.fda.gov/news‐events/press‐announcements/fda‐allows‐marketing‐first‐whole‐slide‐imaging‐system‐digital‐pathology.
- 4. Araujo A. L. D., Arboleda L. P. A., Palmier N. R., et al., “The Performance of Digital Microscopy for Primary Diagnosis in Human Pathology: A Systematic Review,” Virchows Archiv 474 (2019): 269–287. [DOI] [PubMed] [Google Scholar]
- 5. Campbell W. S., Hinrichs S. H., Lele S. M., et al., “Whole Slide Imaging Diagnostic Concordance With Light Microscopy for Breast Needle Biopsies,” Human Pathology 45 (2014): 1713–1721. [DOI] [PubMed] [Google Scholar]
- 6. Saco A., Diaz A., Hernandez M., et al., “Validation of Whole‐Slide Imaging in the Primary Diagnosis of Liver Biopsies in a University Hospital,” Digestive and Liver Disease 49 (2017): 1240–1246. [DOI] [PubMed] [Google Scholar]
- 7. Al‐Janabi S., Huisman A., Nap M., Clarijs R., and van Diest P. J., “Whole Slide Images as a Platform for Initial Diagnostics in Histopathology in a Medium‐Sized Routine Laboratory,” Journal of Clinical Pathology 65 (2012): 1107–1111. [DOI] [PubMed] [Google Scholar]
- 8. Villa I., Mathieu M. C., Bosq J., et al., “Daily Biopsy Diagnosis in Surgical Pathology: Concordance Between Light Microscopy and Whole‐Slide Imaging in Real‐Life Conditions,” American Journal of Clinical Pathology 149 (2018): 344–351. [DOI] [PubMed] [Google Scholar]
- 9. Ibrahim A., Lashen A. G., Katayama A., et al., “Defining the Area of Mitoses Counting in Invasive Breast Cancer Using Whole Slide Image,” Modern Pathology 35 (2022): 739–748. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Girolami I., Pantanowitz L., Marletta S., et al., “Diagnostic Concordance Between Whole Slide Imaging and Conventional Light Microscopy in Cytopathology: A Systematic Review,” Cancer Cytopathology 128 (2020): 17–28. [DOI] [PubMed] [Google Scholar]
- 11. Capitanio A., Dina R. E., and Treanor D., “Digital Cytology: A Short Review of Technical and Methodological Approaches and Applications,” Cytopathology 29 (2018): 317–325. [DOI] [PubMed] [Google Scholar]
- 12. Evans A. J., Brown R. W., Bui M. M., et al., “Validating Whole Slide Imaging Systems for Diagnostic Purposes in Pathology,” Archives of Pathology & Laboratory Medicine 146 (2022): 440–450. [DOI] [PubMed] [Google Scholar]
- 13. Bongaerts O., Clevers C., Debets M., et al., “Conventional Microscopical Versus Digital Whole‐Slide Imaging‐Based Diagnosis of Thin‐Layer Cervical Specimens: A Validation Study,” Journal of Pathology Informatics 9 (2018): 29. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Cheng S., Liu S., Yu J., et al., “Robust Whole Slide Image Analysis for Cervical Cancer Screening Using Deep Learning,” Nature Communications 12 (2021): 5639. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Alsalatie M., Alquran H., Mustafa W. A., Mohd Y. Y., and Ali A. A., “Analysis of Cytology Pap Smear Images Based on Ensemble Deep Learning Approach,” Diagnostics 12 (2022): 2756. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Ross J., Greaves J., Earls P., Shulruf B., and Van Es S. L., “Digital vs Traditional: Are Diagnostic Accuracy Rates Similar for Glass Slides vs Whole Slide Images in a Non‐Gynaecological External Quality Assurance Setting?” Cytopathology 29 (2018): 326–334. [DOI] [PubMed] [Google Scholar]
- 17. Kim D., Burkhardt R., Alperstein S. A., et al., “Evaluating the Role of Z‐Stack to Improve the Morphologic Evaluation of Urine Cytology Whole Slide Images for High‐Grade Urothelial Carcinoma: Results and Review of a Pilot Study,” Cancer Cytopathology 130 (2022): 630–639. [DOI] [PubMed] [Google Scholar]
- 18. Greaves J., Pagliuso J., Ross J., and Badrick T., “Rapid On‐Site Evaluation Technical Exercise Using Digital Pathology: Peer Comparison in an External Quality Assurance Setting,” Acta Cytologica 66, no. 3 (2022): 235–243. [DOI] [PubMed] [Google Scholar]
- 19. Elmas H., Önal B., Steurer S., et al., “Rapid Remote Online Evaluation in Endoscopic Diagnostics: An Analysis of Biopsy‐Proven Respiratory Cytopathology,” Diagnostics 13, no. 21 (2023): 3329. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Lin R., Sheng L. P., Han C. Q., et al., “Application of Artificial Intelligence to Digital‐Rapid On‐Site Cytopathology Evaluation During Endoscopic Ultrasound‐Guided Fine Needle Aspiration: A Proof‐of‐Concept Study,” Journal of Gastroenterology and Hepatology 38, no. 6 (2023): 883–887. [DOI] [PubMed] [Google Scholar]
- 21. Perez D., Stemmer M. N., and Khurana K. K., “Utilization of Dynamic Telecytopathology for Rapid Onsite Evaluation of Touch Imprint Cytology of Needle Core Biopsy: Diagnostic Accuracy and Pitfalls,” Telemedicine Journal and E‐Health 27 (2021): 525–531. [DOI] [PubMed] [Google Scholar]
- 22. Ren F., Ping B., Chen Y., et al., “Application Value of Rapid Immunocytochemistry for Intraoperative Touch Imprint Cytology Diagnosis of Axillary Sentinel Lymph Nodes in Breast Cancer,” Zhonghua Bing Li Xue Za Zhi 51 (2022): 444–446. [DOI] [PubMed] [Google Scholar]
- 23. Landis J. R. and Koch G. G., “The Measurement of Observer Agreement for Categorical Data,” Biometrics 33 (1977): 159–174. [PubMed] [Google Scholar]
- 24. Goacher E., Randell R., Williams B., and Treanor D., “The Diagnostic Concordance of Whole Slide Imaging and Light Microscopy: A Systematic Review,” Archives of Pathology & Laboratory Medicine 141 (2017): 151–161. [DOI] [PubMed] [Google Scholar]
- 25. Chang Y. C. and Tzen C. Y., “Intraoperative Sentinel Lymph Node Imprint Cytology Diagnosis in Breast Cancer Patients by General Surgical Pathologists: A Single‐Institution Experience of 4327 Cases,” Journal of Cytology 39 (2022): 20–25. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Aldoheyan T. and Klein J., “Quality Assurance Review: Intra‐Operative Evaluation of Sentinel Lymph Nodes in Breast Cancer,” Cancer Medicine 10 (2021): 7213–7221. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Komenaka I. K., Torabi R., Nair G., et al., “Intraoperative Touch Imprint and Frozen Section Analysis of Sentinel Lymph Nodes After Neoadjuvant Chemotherapy for Breast Cancer,” Annals of Surgery 251 (2010): 319–322. [DOI] [PubMed] [Google Scholar]
- 28. Khanna R., Bhadani S., Khanna S., Pandey M., and Kumar M., “Touch Imprint Cytology Evaluation of Sentinel Lymph Node in Breast Cancer,” World Journal of Surgery 35 (2011): 1254–1259. [DOI] [PubMed] [Google Scholar]
- 29. Eccher A. and Girolami I., “Current State of Whole Slide Imaging Use in Cytopathology: Pros and Pitfalls,” Cytopathology 31 (2020): 372–378. [DOI] [PubMed] [Google Scholar]
- 30. Campbell W. S., Talmon G. A., Foster K. W., Lele S. M., Kozel J. A., and West W. W., “Sixty‐Five Thousand Shades of Gray: Importance of Color in Surgical Pathology Diagnoses,” Human Pathology 46 (2015): 1945–1950. [DOI] [PubMed] [Google Scholar]
- 31. Zheng Y., Jiang Z., Zhang H., Xie F., Shi J., and Xue C., “Adaptive Color Deconvolution for Histological WSI Normalization,” Computer Methods and Programs in Biomedicine 170 (2019): 107–112. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.