Hein S. Zelisse

Performance Assessment of a Deep Learning–based Algorithm for Ovarian Cancer Histotyping in an Independent Data Set

Artificial intelligence diagnostic tools show promise for improving histotype classification in epithelial ovarian cancer but face challenges due to slide variability across institutions. To address this domain shift, the adversarial Fourier-based domain adaptation (AIDA) model was developed. This retrospective study evaluates AIDA’s performance in classifying the 5 major ovarian cancer subtypes using an independent cohort. Surgically treated patients diagnosed with clear cell (CCC), endometrioid (EC), high-grade serous (HGSC), low-grade serous (LGSC), or mucinous (MC) ovarian cancer at Amsterdam University Medical Center (1985-2022) were included in the study. The deep learning method AIDA, trained on data from Vancouver General Hospital, was applied to all cases. Final histotype predictions were made through majority voting across 15 independently trained models. For misclassified cases, up to 3 additional slides were scanned, and the AIDA model was retrained. Classification was then assessed using single-slide and majority voting approaches. The AIDA algorithm achieved an overall balanced accuracy of 79.7% across all histotypes. Accuracy was highest for CCC (90.9%) and LGSC (89.8%), and lowest for EC (62.4%). Common misclassifications included MC as EC and EC as HGSC or LGSC. Retraining with additional slides improved balanced accuracy to 85.8% based on single-slide voting and 82.6% based on majority voting. This study highlights the future potential of the AIDA model in classifying epithelial ovarian cancer histotypes. With further refinement to improve performance on more challenging cases, the model could enhance diagnostic accuracy in clinical practice.

Improving cell-free DNA detection in advanced-stage high-grade serous ovarian cancer using combined TP53 mutational status and copy number changes

Circulating tumor DNA (ctDNA) is a promising biomarker in patients with high-grade serous ovarian cancer (HGSOC). However, the detection rate of TP53 mutations in ctDNA of HGSOC patients has previously been shown to be inadequate. Given the prevalence of copy number aberrations (CNAs) in HGSOC, this study aimed to improve ctDNA detection by combining TP53 sequencing with shallow whole-genome sequencing (sWGS), and to evaluate the correlation with clinicopathological features and survival outcomes. This exploratory, retrospective cohort study included 53 advanced-stage HGSOC patients, comprising 18 treatment-naive patients and 35 patients treated with two neoadjuvant chemotherapy cycles. TP53 targeted sequencing was integrated with sWGS (<5x coverage) for CNA estimation using ichor copy number aberration tumor fraction (ichorCNA TF). TP53 mutations were detected in 28 patients (52.8 %), and 17 patients (32.1 %) showed positive ichorCNA TF. Combining TP53 mutation detection with ichorCNA TF identified 62.3 % (n = 33) of patients as ctDNA-positive, showing a trend towards improved detection compared to TP53 mutation alone (p = .063). Treatment-naive patients exhibited higher TP53 mutation (72.2 % vs. 42.9 %, p = .043) and ichorCNA TF (66.7 % vs. 14.3 %, p < .001) detection rates compared to chemotherapy-treated patients. No correlations between ctDNA metrics and clinicopathological characteristics or survival outcomes were found. In conclusion, the integration of ichorCNA TF with TP53 mutation analysis showed a trend towards improved ctDNA detection in advanced-stage HGSOC patients. Future studies should further explore ctDNA detection rates by ichorCNA TF and its potential clinical implications in HGSOC.

Improving histotyping precision: The impact of immunohistochemical algorithms on epithelial ovarian cancer classification

To improve the precision of epithelial ovarian cancer histotyping, Köbel et al. (2016) developed immunohistochemical decision-tree algorithms. These included a six- and four-split algorithm, and separate six-split algorithms for early- and advanced stage disease. In this study, we evaluated the efficacy of these algorithms. A gynecological pathologist determined the hematoxylin and eosin (H&E)-based histotypes of 230 patients. Subsequently, the final histotypes were established by re-evaluating the H&E-stained sections and immunohistochemistry outcomes. For histotype prediction using the algorithms, the immunohistochemical markers Napsin A, p16, p53, progesterone receptor (PR), trefoil factor 3 (TFF3), and Wilms' tumor 1 (WT1) were scored. The algorithmic predictions were compared with the final histotypes to assess their precision, for which the early- and advanced stage algorithms were assessed together as six-split-stages algorithm. The six-split algorithm demonstrated 96.1% precision, whereas the six-split-stages and four-split algorithms showed 93.5% precision. Of the 230 cases, 16 (7%) showed discordant original and final diagnoses; the algorithms concurred with the final diagnosis in 14/16 cases (87.5%). In 12.4%-13.3% of cases, the H&E-based histotype changed based on the algorithmic outcome. The six-split stages algorithm had a lower sensitivity for low-grade serous carcinoma (80% versus 100%), while the four-split stages algorithm showed reduced sensitivity for endometrioid carcinoma (78% versus 92.7-97.6%). Considering the higher sensitivity of the six-split algorithm for endometrioid and low-grade serous carcinoma compared with the four-split and six-split-stages algorithms, respectively, we recommend the adoption of the six-split algorithm for histotyping epithelial ovarian cancer in clinical practice.

The Information Technology (IT) Infrastructure of the Multicenter Archipelago of Ovarian Cancer Research Biobank: A Potential Blueprint for Other Biobanks

Objective: Biobanks play a crucial role in fundamental and translational research by storing valuable biomaterials and data for future analyses. However, the design of their information technology (IT) infrastructures is often customized to specific requirements, thereby lacking the ability to be used for biobanks comprising other (types of) diseases. This results in substantial costs, time, and efforts for each new biobank project. The Dutch multicenter Archipelago of Ovarian Cancer Research (AOCR) biobank has developed an innovative, reusable IT infrastructure capable of adaptation to various biobanks, thereby enabling cost-effective and efficient implementation and management of biobank IT systems. Methods and Results: The AOCR IT infrastructure incorporates preexisting biobank software, mainly managed by Health-RI. The web-based registration tool Ldot is used for secure storage and pseudonymization of patient data. Clinicopathological data are retrieved from the Netherlands Cancer Registry and the Dutch nationwide pathology databank (Palga), both established repositories, reducing administrative workload and ensuring high data quality. Metadata of collected biomaterials are stored in the OpenSpecimen system. For digital pathology research, a hematoxylin and eosin-stained slide from each patient’s tumor is digitized and uploaded to Slide Score. Furthermore, adhering to the Findable, Accessible, Interoperable, and Reusable (FAIR) principles, genomic data derived from the AOCR samples are stored in cBioPortal. Conclusion: The IT infrastructure of the AOCR biobank represents a new standard for biobanks, offering flexibility to handle diverse diseases and types of biomaterials. This infrastructure bypasses the need for disease-specific, custom-built software, thereby being cost- and time-effective while ensuring data quality and legislative compliance. The adaptability of this infrastructure highlights its potential to serve as a blueprint for the development of IT infrastructures in both new and existing biobanks.

Evaluation of the prognostic potential of histopathological subtyping in high-grade serous ovarian carcinoma

Abstract High-grade serous ovarian carcinoma (HGSOC) can be categorized into four gene expression-based subtypes, with supposedly distinct prognoses and treatment responses. Murakami et al. translated these gene expression-based subtypes into the histopathological mesenchymal, immunoreactive, solid and proliferative, and papilloglandular subtypes, showing differences in survival outcomes. Miyagawa et al. refined these criteria to improve the interobserver concordance. The current retrospective study evaluated the interobserver variability and the prognostic differences between the histopathologic subtypes using the criteria of both Murakami et al. and Miyagawa et al. in 208 HGSOC cases. The mesenchymal subtype was considered first, followed by the immunoreactive subtype. Non-conforming cases were categorized as solid and proliferative or papilloglandular. The mesenchymal subtype was identified in 122 patients (58.7%) for both criteria. Using the criteria of Murakami et al., 10 cases (4.8%) were immunoreactive, 26 (12.5%) solid and proliferative, and 50 (24%) papilloglandular, with a concordance rate of 62.5% (κ = 0.34, p &lt; .001). Using the Miyagawa et al. criteria, 23 cases (11%) were immunoreactive, 20 (9.6%) solid and proliferative, and 43 (20.7%) papilloglandular. No survival differences were observed between the subtypes. The fair reproducibility of the histopathological subtype classification of HGSOC and the lack of survival differences among these subtypes indicate the need for further refinement of the criteria and exploration of their correlation with overall survival outcomes before clinical application.