Publications

Insight, Publicly Sharable

Background: The Visiopharm automated estrogen receptor (ER) digital imaging analysis (DIA) algorithm assesses digitized ER immunohistochemistry (IHC) by segmenting tumor nuclei and detecting stained nuclei automatically. We aimed to integrate and validate this algorithm in a digital pathology workflow for clinical use. Design: The study cohort consisted of a serial collection of 97 invasive breast carcinoma specimens including 73 biopsies and 24 resections. ER IHC slides were scanned into Philips Image Management System (IMS) during our routine digital workflow and digital images were directly streamed into Visiopharm platform and analyzed using automated ER algorithm to obtain the positively stained tumor nuclei and staining intensity. ER DIA scores were compared with pathologists’ manual scores. Results: The overall concordance between pathologists’ reads and DIA reads was excellent (91/97, 93.8%). Pearson Correlation Coefficient of the percentage of ER positive nuclei between the original reads and VIS reads was 0.72. Six cases (3 ER-negative and 3 ER-positive) had discordant results. All 3 false negative cases had very weak ER staining and no more than 10% positivity. The causes for false positive DIA were mainly pre-analytic/pre-imaging and included intermixed benign glands in tumor area, ductal carcinoma in-situ (DCIS) components, and tissue folding. Conclusions: Automated ER DIA demonstrates excellent concordance with pathologists’ scores and accurately discriminates ER positive from negative cases. Furthermore, integrating automated biomarker DIA into a busy clinical digital workflow is feasible and may save time and labor for pathologists.

Saba Shafi, David A. Kellough, Giovanni Lujan, Swati Satturwar, Anil V. Parwani, Zaibo Li

90159, Insight, Oncotopix Clinical, Publicly Sharable

Abstract

Identifying lymph node (LN) metastasis in invasive breast carcinoma (IBC) can be tedious and time consuming. We investigated an artificial intelligence (AI) algorithm to detect LN metastasis by screening H&E slides in a clinical digital workflow. The study included two sentinel LN (SLN) cohorts (validation cohort with 234 SLNs and consensus cohort with 102 SLNs) and one non-sentinel LN (NSLN) cohort (258 LNs enriched with lobular carcinoma and post-neoadjuvant therapy cases). All H&E slides were scanned into whole slide images (WSI) in clinical digital workflow and WSIs were automatically batch analyzed using Visiopharm (VIS) metastasis AI algorithm. For SLN validation cohort, VIS metastasis AI algorithm detected all 46 metastases including 19 macrometastases, 26 micrometastases, 1 with isolated tumor cells (ITC) with a sensitivity of 100%, specificity of 41.5%, positive predictive value (PPV) of 29.5% and negative predictive value (NPV) of 100%. The false positivity was caused by histiocytes (52.7%), crushed lymphocytes (18.2%), and others, which were readily recognized during pathologists' reviews. For SLN consensus cohort, three pathologists examined all VIS AI annotated H&E slides and cytokeratin IHC slides with similar average concordance rates (99% for both modalities). However, the average time consumed by pathologists using VIS AI annotated slides was significantly less than the time using IHC slides (0.6 vs 1.0 minute, p=0.0377). For NSLN cohort, AI algorithm detected all 81 metastases, including 23 from lobular carcinoma and 31 from post-neoadjuvant chemotherapy cases with sensitivity of 100%, specificity of 78.5%, PPV of 68.1%, and NPV of 100%. VIS AI algorithm showed perfect sensitivity and NPV in detecting LN metastasis and less time consumed, suggesting its potential utility as a screening modality in routine clinical digital pathology workflow to improve efficiency.

Bindu Challa, Maryam Tahir, Yan Hu, David Kellough, Giovani Lujan, Shaoli Sun, Anil V. Parwani, Zaibo Li

Insight, Oncotopix Clinical, Publicly Sharable

Aim: Manual detection and scoring of Ki67 hotspots is difficult and prone to variability, limiting its clinical utility. Automated hotspot detection and scoring by digital image analysis (DIA) could improve the assessment of the Ki67 hotspot proliferation index (PI). This study compared the clinical performance of Ki67 hotspot detection and scoring DIA algorithms based on virtual dual staining (VDS) and deep learning (DL) with manual Ki67 hotspot PI assessment. Methods: Tissue sections of 135 consecutive invasive breast carcinomas were immunohistochemically stained for Ki67. Two DIA algorithms, based on VDS and DL, automatically determined the Ki67 hotspot PI. For manual assessment; two independent observers detected hotspots and calculated scores using a validated scoring protocol. Results: Automated hotspot detection and assessment by VDS and DL could be performed in 73% and 100% of the cases, respectively. Automated hotspot detection by VDS and DL led to higher Ki67 hotspot PIs (mean 39.6% and 38.3%, respectively) compared to manual consensus Ki67 PIs (mean 28.8%). Comparing manual consensus Ki67 PIs with VDS Ki67 PIs revealed substantial correlation (r = 0.90), while manual consensus versus DL Ki67 PIs demonstrated high correlation (r = 0.95). Conclusion: Automated Ki67 hotspot detection and analysis correlated strongly with manual Ki67 assessment and provided higher PIs compared to manual assessment. The DL-based algorithm outperformed the VDS-based algorithm in clinical applicability, because it did not depend on virtual alignment of slides and correlated stronger with manual scores. Use of a DL-based algorithm may allow clearer Ki67 PI cutoff values, thereby improving the clinical usability of Ki67.

Mieke C. Zwager, Shibo Yu, Henk J. Buikema, Geertruida H. de Bock, Thomas W. Ramsing, Jeppe Thagaard, Timco Koopman, Bert van der Vegt

90159, Insight, Oncotopix Clinical, Publicly Sharable

Pathologists’ assessment of sentinel lymph nodes (SNs) for breast cancer (BC) metastases is a treatment-guiding yet labor-intensive and costly task because of the performance of immunohistochemistry (IHC) in morphologically negative cases. This non-randomized, single-center clinical trial (International Standard Randomized Controlled Trial Number:14323711) assessed the efficacy of an artificial intelligence (AI)-assisted workflow for detecting BC metastases in SNs while maintaining diagnostic safety standards. From September 2022 to May 2023, 190 SN specimens were consecutively enrolled and allocated biweekly to the intervention arm (n = 100) or control arm (n = 90). In both arms, digital whole-slide images of hematoxylin–eosin sections of SN specimens were assessed by an expert pathologist, who was assisted by the ‘Metastasis Detection’ app (Visiopharm) in the intervention arm. Our primary endpoint showed a significantly reduced adjusted relative risk of IHC use (0.680, 95% confidence interval: 0.347–0.878) for AI-assisted pathologists, with subsequent cost savings of ~3,000 €. Secondary endpoints showed significant time reductions and up to 30% improved sensitivity for AI-assisted pathologists. This trial demonstrates the safety and potential for cost and time savings of AI assistance. Van Dooijeweert et al. conducted a prospective study on the clinical implementation of artificial-intelligence-assisted detection of sentinel lymph node metastasis in persons with breast cancer and report on its effects, including on time and cost.

C. van Dooijeweert, R. N. Flach, N. D. ter Hoeve, C. P. H. Vreuls, R. Goldschmeding, J. E. Freund, P. Pham, T. Q. Nguyen, E. van der Wall, G. W. J. Frederix, N. Stathonikos, P. J. van Diest

Insight, Lymphnode Metastasis, Oncotopix Clinical, Publicly Sharable

The rapid introduction of digital pathology has greatly facilitated development of artificial intelligence (AI) models in pathology that have shown great promise in assisting morphological diagnostics and quantitation of therapeutic targets. We are now at a tipping point where companies have started to bring algorithms to the market, and questions arise whether the pathology community is ready to implement AI in routine workflow. However, concerns also arise about the use of AI in pathology. This article reviews the pros and cons of introducing AI in diagnostic pathology.

Paul J. van Diest, Rachel N. Flach, Carmen van Dooijeweert, Seher Makineli, Gerben E. Breimer, Nikolas Stathonikos, Paul Pham, Tri Q. Nguyen, Mitko Veta

Insight, Oncotopix Clinical, Publicly Sharable, Qu-Path

Quantification of Ki67 in breast cancer is a well-established prognostic and predictive marker, but inter-laboratory variability has hampered its clinical usefulness. This study compares the prognostic value and reproducibility of Ki67 scoring using four automated, digital image analysis (DIA) methods and two manual methods. The study cohort consisted of 367 patients diagnosed between 1990 and 2004, with hormone receptor positive, HER2 negative, lymph node negative breast cancer. Manual scoring of Ki67 was performed using predefined criteria. DIA Ki67 scoring was performed using QuPath and Visiopharm® platforms. Reproducibility was assessed by the intraclass correlation coefficient (ICC). ROC curve survival analysis identified optimal cutoff values in addition to recommendations by the International Ki67 Working Group and Norwegian Guidelines. Kaplan–Meier curves, log-rank test and Cox regression analysis assessed the association between Ki67 scoring and distant metastasis (DM) free survival. The manual hotspot and global scoring methods showed good agreement when compared to their counterpart DIA methods (ICC > 0.780), and good to excellent agreement between different DIA hotspot scoring platforms (ICC 0.781–0.906). Different Ki67 cutoffs demonstrate significant DM-free survival (p < 0.05). DIA scoring had greater prognostic value for DM-free survival using a 14% cutoff (HR 3.054–4.077) than manual scoring (HR 2.012–2.056). The use of a single cutoff for all scoring methods affected the distribution of prediction outcomes (e.g. false positives and negatives). This study demonstrates that DIA scoring of Ki67 is superior to manual methods, but further study is required to standardize automated, DIA scoring and definition of a clinical cut-off.

Emma Rewcastle, Ivar Skaland, Einar Gudlaugsson, Silja Kavlie Fykse, Jan P. A. Baak, Emiel A. M. Janssen

Insight, Oncotopix Clinical, Publicly Sharable

his review discusses the profound impact of artificial intelligence (AI) on breast cancer (BC) diagnosis and management within the field of pathology. It examines the various applications of AI across diverse aspects of BC pathology, highlighting key findings from multiple studies. Integrating AI into routine pathology practice stands to improve diagnostic accuracy, thereby contributing to reducing avoidable errors. Additionally, AI has excelled in identifying invasive breast tumors and lymph node metastasis through its capacity to process large whole-slide images adeptly. Adaptive sampling techniques and powerful convolutional neural networks mark these achievements. The evaluation of hormonal status, which is imperative for BC treatment choices, has also been enhanced by AI quantitative analysis, aiding interobserver concordance and reliability. Breast cancer grading and mitotic count evaluation also benefit from AI intervention. AI-based frameworks effectively classify breast carcinomas, even for moderately graded cases that traditional methods struggle with. Moreover, AI-assisted mitotic figures quantification surpasses manual counting in precision and sensitivity, fostering improved prognosis. The assessment of tumor-infiltrating lymphocytes in triple-negative breast cancer using AI yields insights into patient survival prognosis. Furthermore, AI-powered predictions of neoadjuvant chemotherapy response demonstrate potential for streamlining treatment strategies. Addressing limitations, such as preanalytical variables, annotation demands, and differentiation challenges, is pivotal for realizing AI’s full potential in BC pathology. Despite the existing hurdles, AI’s multifaceted contributions to BC pathology hold great promise, providing enhanced accuracy, efficiency, and standardization. Continued research and innovation are crucial for overcoming obstacles and fully harnessing AI’s transformative capabilities in breast cancer diagnosis and assessment. Graphical Abstract: (Figure presented.) HER2 ER lymph node metastasis

Amr Soliman, Zaibo Li, Anil V. Parwani

Insight, Publicly Sharable

Abstract

Trastuzumab-deruxtecan (T-DXd), an antibody-drug conjugate targeting human epidermal growth factor receptor 2 (HER2), improves overall survival in patients with breast cancer showing low or ultra-low HER2 expression. Differences in HER2 test results have been reported between various HER2 assays and between primary tumors and metastases, but an objective comparison with incorporation of the new HER2-ultralow cut-off values is needed. This study aimed to assess the performance of four routine clinical-grade HER2 assays across and between primary tumors and their metastases using digital image analysis (DIA).

Primary tumors and metastases from 193 patients with breast cancer who participated in the IMPACT-MBC trial were incorporated into six tissue microarrays. Samples were stained by four laboratories using their routine HER2 immunohistochemistry protocols: 4B5 ultraView, 4B5 OptiView, SP3, and HercepTest (DG44). HER2 scores were determined using DIA. The four HER2 assays showed significant differences in HER2 status in both primary tumors and metastases.

Eighty-five matched primary tumors and metastases were analyzed to investigate concordance in HER2 status. While no significant differences were found in HER2 scores between primary tumors and metastases for SP3 and both 4B5 assays, DG44 showed significantly higher HER2 scores in the metastasis (p = 0.004). Concordance between primary tumors and metastases was highest for 4B5 ultraView (69.4%), followed by SP3 (61.2%) and 4B5 OptiView (51.8%). DG44 showed the most variability, with only 36.5% of matched samples receiving the same HER2 category.

DG44 identified a significantly higher proportion of HER2-(ultra)low cases and showed the most variability in HER2 status between matched primary tumors and metastases compared to 4B5 and SP3. The choice of HER2 assay can lead to discrepancies in HER2 status assessment, which could directly influence patient eligibility for T-DXd treatment.

Maaike Anna Hempenius, Mieke C. Zwager, Jeppe Thagaard, Lorian Slagter-Menkema, Henk J. Buikema, Ellis Barbé, Michael A. den Bakker, Marjolein G.J. Heerema, Elisabeth G.E. de Vries, Carolina P. Schröder, Nils A. ’t Hart, Bert van der Vegt

Insight, Oncotopix Clinical, Publicly Sharable

Inter‐observer concordance data for the HER2 category as assessed by a group of 16 specialist breast pathologists on 50 diagnostic core biopsies was compared with that produced by digital image analysis (DIA) using the HER2 APP, CE2797 (VP APP; Visiopharm, Hoersholm, Denmark). Comparing pathologists' consensus scores and DIA scores, 36 cases (73.5%) agreed. Fleiss' kappa statistic was 0.433 (indicative of moderate agreement). Cohen's weighted kappa was used to compare the scores of individual raters to consensus scores; for all 50 cases the kappa scores had a range between 0.412 and 0.854; the VP APP was ranked 12th of 17 raters (kappa score 0.638 indicating substantial agreement). Results for HER2‐low cases ( N  = 44) showed a kappa score range of 0.295 to 0.823; the VP APP ranked 12th of 17 (score 0.535 indicating moderate agreement). For high agreement cases the kappa score range was 0.664 to 1.000 for all HER2 scores ( N  = 24) and the VP APP scored 0.916 (indicating almost perfect agreement). For the HER2‐low scores ( N  = 20), the kappa score range was 0.506–1.000 and the VP APP scored 0.860 (almost perfect agreement). DIA of the proportions of tumour cells showing expression within each of the HER2 categories demonstrated that the majority of cases showing a low level of agreement between pathologists showed heterogeneity and/or a level of expression close to a cut‐point for decision making. This study demonstrates that the VP APP produces results that are extremely well‐aligned to those of expert pathologists in cases with good overall agreement, and in difficult cases its reproducibility will outperform that of the visual scorer. The results also suggest that use of the VP APP has the potential to reduce the proportion of cases referred for gene amplification testing by reducing the number of cases incorrectly classified as HER2 2+.

Suzanne Parry, Lila Zabaglo, Abeer M Shaaban, Andrew Dodson

Insight, Publicly Sharable

David J. Dabbs, Emina Torlakovic, Søren Nielsen, Suzanne C. Parry, Jing Yu, Catherine Stoos, Beth Clark, Henrik Høeg, Jeppe Thagaard, Seshi R. Sompuram, Stephen P. Naber, Yukako Yagi, James Sayre, Kodela Vani, Mélissande Cossutta, Francoise Soussaline, Alexandre Papine, Nils A. t'Hart, Matthias J. Szabolcs, Bharat Jasani, Mary Kinloch, Luis Chiriboga, Keith Miller, Steve Bogen

Insight, Publicly Sharable

Implementation of a standard such as Digital Imaging and Communications in Medicine (DICOM) is key to all aspects of interoperability in whole-slide imaging. But the devil is in the details, so practical testing and demonstration of specific features are needed to show the path forward for practical clinical deployment, select the appropriate profile of features, and identify gaps and opportunities for improvement. The most recent Connectathon was conducted for this purpose, as a virtual Internet-mediated event. Thirty-two (32) implementers in the role of Acquisition Manager (AP-LIS) (2), Acquisition Modality (scanner) (9), Image Manager/Archive (PACS, IMS, or VNA) (7), Image Display (viewer) (15), and Evidence Creator (annotation source) (8) participated. The AP-LIS provided HL7 V2 metadata identifying and describing patients and specimens in response to a slide barcode-based query, which was then incorporated by the scanner into standard DICOM whole-slide microscopy images, which were encoded as tiled pyramids using the TILED_FULL pattern. These images were transferred to the archive using the standard DICOM protocol, and made available for virtual microscopy viewing using the standard DICOMweb mechanisms for query and retrieval of metadata and selected frames. Human and algorithm generated annotations were produced and stored in the standard DICOM annotation format, and transferred and retrieved for display, also using the standard DICOMweb mechanisms.

David A. Clunie, Brian Napora, John Groth, Mustafa Yousif, Gitesh Shelar, Tamás Sárga, Sean Borman, Charles Cheng, Philipp Plewa, Luís Bastião Silva, Ryan Birmingham, Kenneth Philbrick, Matti Pellikka, Richard Preston, Lucas Tata, Drew Anderson, Frank Yang, Andrey Fedorov, Saurav Patel, JiHae Kwon, Michael Knapp, Ty Usrey, Francisco José Carrasco Tena, Emilio Madrigal, Dennis Wang, Chung-Yueh Lien, Louise Collins, Arjan Somers, Aditya Saxena, Eric Martin, Mohannad Hussain, C.Y. Lien, Paul Cram, Johan Doré, Filipe Carreira

Insight, Publicly Sharable

Increasing workload combined with the shortage of pathologists is the leading cause of diagnostic errors and delays. Nonetheless, in clinical practice, pathologists often spend hours on tedious tasks such as counting mitoses and searching for lymph node micro‐metastasis, which may yield unreliable results. The advent of digital pathology and the development of artificial intelligence (AI) applications (app) for image analysis have opened new possibilities for improving the efficiency and accuracy of pathologists. However, the perceived black box nature of AI has led to skepticism among many pathologists about its diagnostic capabilities, resulting in a lack of trust in AI. In addition, it is a common belief that AI applications should be limited to the areas they were trained in, which has significantly limited their generalizability. Given the homogeneous cell population of lymph nodes and overlapping of tumor morphology across different organs, we hypothesized that a lymph node metastasis detection application trained on a few organs could potentially recognize metastasis from multiple organs. We used the commercially available Visiopharm app (AI tool), initially trained on lymph node metastases from breast and colon cancer, to detect metastasis of 12 distinct types of cancer from 15 organ systems based on the analysis of 172 slides (all with corresponding immunohistochemical staining confirmation). Furthermore, by using the annotation map generated by the app as a guide, pathologists were also able to reduce the time spent searching for metastasis substantially (from 54.7 to 42.1 s per slide on average) without compromising diagnostic accuracy. With pathologists serving as the trusted gatekeepers and the development of more sophisticated image analysis applications, the use of AI can help to address the shortage of pathologists, enhance their performance and eventually improve patient care.

Jennifer Vazzano, Bindu Challa, Vidya Arole, Konstantin Shilo, Sarah Reuss, Peter Kobalka, Swati Satturwar, Juan Xie, Dongjun Chung, Saba Shafi, David Kellough, Erin Palermini, Zaibo Li, Wei Chen, Anil Parwani, Shaoli Sun