Built to Last? Reproducibility and Reusability of Deep Learning Algorithms in Computational Pathology.

Recent progress in computational pathology has been driven by deep learning. While code and data availability are essential to reproduce findings from preceding publications, ensuring a deep learning model's reusability is more challenging. For that, the codebase should be well-documented and easy to integrate into existing workflows and models should be robust toward noise and generalizable toward data from different sources. Strikingly, only a few computational pathology algorithms have been reused by other researchers so far, let alone employed in a clinical setting. To assess the current state of reproducibility and reusability of computational pathology algorithms, we evaluated peer-reviewed articles available in PubMed, published between January 2019 and March 2021, in 5 use cases: stain normalization; tissue type segmentation; evaluation of cell-level features; genetic alteration prediction; and inference of grading, staging, and prognostic information. We compiled criteria for data and code availability and statistical result analysis and assessed them in 160 publications. We found that only one-quarter (41 of 160 publications) made code publicly available. Among these 41 studies, three-quarters (30 of 41) analyzed their results statistically, half of them (20 of 41) released their trained model weights, and approximately a third (16 of 41) used an independent cohort for evaluation. Our review is intended for both pathologists interested in deep learning and researchers applying algorithms to computational pathology challenges. We provide a detailed overview of publications with published code in the field, list reusable data handling tools, and provide criteria for reproducibility and reusability.

Tumour budding-based grading as independent prognostic biomarker in HPV-positive and HPV-negative head and neck cancer.

BACKGROUND: The prognostic significance of tumour budding (TB) and minimal cell nest size (MCNS) was shown in human papillomavirus (HPV)-negative head and neck squamous cell carcinomas (HNSCC). However, the optimisation of cutpoints, the prognostic impact in HPV-positive HNSCC, and the comparison with other histopathological grading systems are insufficiently investigated. METHODS: TB and MCNS were analysed digitally in 1 and 10 high-power fields (HPF) of 331 HPV-positive and HPV-negative cases from TCGA. Optimising the cutpoints a new cellular dissociation grading (CDG) system was defined and compared to the WHO grading and the Brandwein-Gensler (BG) risk model. RESULTS: The two-tiered CDG system based solely on TB yielded optimal prognostic stratification with shortened overall survival for CDG-high cases. Optimal cut-offs were two buds (1 HPF) and six buds (10 HPF), respectively. Analysing MCNS did not add prognostic significance to quantifying TB. CDG was a significant prognostic marker in HPV-negative and HPV-positive tumours and prognostically superior to the WHO and BG systems. High CDG was associated with clinically occult lymph-node metastases. CONCLUSIONS: The most comprehensive study of TB in HNSCC so far confirmed its prognostic impact in HPV-negative tumours and for the first time in HPV-positive tumours. Further studies are warranted to evaluate its applicability for therapy guidance in HNSCC.

Comparison of PD-L1 expression in squamous cell cancer of unknown primary and oropharyngeal squamous cell carcinoma.

PURPOSE: The tumorigenesis of squamous cell cancer of unknown primary (SCCUP) in the head and neck area has not been decoded so far, while poor survival rates and limited therapeutic options pose a serious challenge. The aim of this project was to investigate immunological characteristics of SCCUPs and compare them to oropharyngeal squamous cell carcinoma (OPSCC). METHODS: PD-L1 expression (TC) was examined by immunohistochemistry in 50 lymph node metastases of SCCUP and 47 primaries of OPSCC. CD3 + and CD8 + lymphocytic infiltration was measured in 5 high power fields. Expression of p16 and HPV ISH were assessed. RESULTS: SCCUP demonstrated a significantly higher expression of PD-L1 than OPSCC. In p16-negative SCCUPs PD-L1 proved to be an independent prognostic factor to prioritize high-risk patients. CONCLUSIONS: Immunologic differences between SCCUP and OPSCC were detected. A higher PD-L1 expression in SCCUP could potentially facilitate further evaluation of checkpoint inhibitor therapy.

Loss of CDX2 in colorectal cancer is associated with histopathologic subtypes and microsatellite instability but is prognostically inferior to hematoxylin-eosin-based morphologic parameters from the WHO classification.

BACKGROUND: Immunohistochemical loss of CDX2 has been proposed as a biomarker of dismal survival in colorectal carcinoma (CRC), especially in UICC Stage II/III. However, it remains unclear, how CDX2 expression is related to central hematoxylin-eosin (HE)-based morphologic parameters defined by 2019 WHO classification and how its prognostic relevance is compared to these parameters. METHODS: We evaluated CDX2 expression in 1003 CRCs and explored its prognostic relevance compared to CRC subtypes, tumour budding and WHO grade in the overall cohort and in specific subgroups. RESULTS: CDX2-low/absent CRCs were enriched in specific morphologic subtypes, right-sided and microsatellite-instable (MSI-H) CRCs (P < 0.001) and showed worse survival characteristics in the overall cohort/UICC Stage II/III (e.g. DFS: P = 0.005) and in microsatellite stable and left-sided CRCs, but not in MSI-H or right-sided CRCs. Compared with CDX2, all HE-based markers showed a significantly better prognostic discrimination in all scenarios. In multivariate analyses including all morphologic parameters, CDX2 was not an independent prognostic factor. CONCLUSION: CDX2 loss has some prognostic impact in univariate analyses, but its prognostic relevance is considerably lower compared to central HE-based morphologic parameters defined by the WHO classification and vanishes in multivariate analyses incorporating these factors.

The immunologic tumor microenvironment in endometrioid endometrial cancer in the morphomolecular context: mutual correlations and prognostic impact depending on molecular alterations.

OBJECTIVE: POLE-mutant, microsatellite-instable (MSI), p53-mutant and non-specific molecular profile (NSMP) are TCGA-defined molecular subgroups of endometrial cancer (EC). Hypothesizing that morphology and tumor immunology might differ depending on molecular background concerning composition and prognostic impact, we aimed to comprehensively interconnect morphologic, immunologic and molecular data. METHODS: TCGA-defined molecular groups were determined by immunohistochemistry and sequencing in n = 142 endometrioid EC. WHO-defined histopathological grading was performed. The immunologic microenvironment (iTME) was characterised by the quantification of intraepithelial and stromal populations of tumor-infiltrating lymphocytes (TIL: overall T-cells; T-Killer cells; regulatory T-cells (Treg)). Immunologic parameters were correlated with WHO-grading, TCGA-subgroups and prognosis. RESULTS: High density TIL were significantly more frequent in high-grade (G3) compared to low-grade (G1/2) EC in the whole cohort and in the subgroup of POLE-wildtype-/microsatellite-stable-EC. MSI was associated with high-level TIL-infiltration when taking into account the type of mismatch repair defect (MLH1/PMS2; MSH2/MSH6). Prognostic impact of biomarkers depended on molecular subgroups: In p53-mutant EC, Treg were independently prognostic, in NSMP, the unique independently prognostic biomarker was WHO-grading. CONCLUSIONS: EC morphology and immunology differ depending on genetics. Our study delineated two molecularly distinct subgroups of immunogenic EC characterized by high-density TIL-infiltration: MSI EC and high-grade POLE-wildtype/microsatellite-stable-EC. Prognostic impact of TIL-populations relied on TCGA-subgroups indicating specific roles for TIL depending on molecular background. In NSMP, histopathological grading was the only prognostic biomarker demonstrating the relevance of WHO-grading in an era of molecular subtyping.

Composition and Clinical Impact of the Immunologic Tumor Microenvironment in Oral Squamous Cell Carcinoma.

Immunotherapy shows promising results and revolutionizes treatment of oral squamous cell carcinoma (OSCC). The immunologic microenvironment might have prognostic/predictive implications. Morphologic immunologic parameters (inflammatory infiltrate, stromal content, and budding activity [BA] [potentially indicating epithelial-mesenchymal transition]) were evaluated in 66 human primary therapy-naive OSCCs. Intraepithelial/stromal tumor-infiltrating lymphocytes (TILs; CD3+/CD4+/CD8+/CD4+FOXP3+/IL-17A+) were quantified, and ratios were calculated. HLA class I in tumor cells was evaluated immunohistochemically. mRNA in situ hybridization to detect IFN-γ was performed. Analysis was performed within invasive front (IF) and tumor center (TCe). Decreased HLA expression was associated with low TIL density, pronounced stromal content, and high BA; IFN-γ in TILs was correlated with high-density TILs; and IFN-γ in tumor cells was correlated with absence of BA (p < 0.05). Heterogeneity of parameters (TCe/IF) was rare. Low density of stromal CD4+FOXP3+ TILs within TCe and IF was identified as an independent prognostic factor for poor overall, disease-specific, and disease-free survival (p ≤ 0.011). Refining prognostication in OSCC with high-density CD4+FOXP3+ infiltrate within TCe and/or IF, high FOXP3:CD4 ratio was significantly correlated with favorable outcome in this subgroup. Furthermore, high-stromal CD8:CD4 ratio was found to be an independent favorable prognostic factor. In summary, immunologic parameters were closely intertwined. Morphologic correlates of epithelial-mesenchymal transition were associated with downregulation of HLA and decreased inflammation. Heterogeneity was infrequent. Low-density stromal CD4+FOXP3+ infiltrate within TCe and IF was an independent poor prognostic factor. Stratification of cases with high-density CD4+FOXP3+ TILs by FOXP3:CD4 ratio enables refinement of prognostication of this subgroup. CD8:CD4 ratio was identified as an independent prognostic factor.

TP53 germline mutations in the context of families with hereditary breast and ovarian cancer: a clinical challenge.

PURPOSE: TP53germline (g) mutations, associated with the Li-Fraumeni syndrome (LFS), have rarely been reported in the context of hereditary breast and ovarian cancer (HBOC). The prevalence and cancer risks in this target group are unknown and counseling remains challenging. Notably an extensive high-risk surveillance program is implemented, which evokes substantial psychological discomfort. Emphasizing the lack of consensus about clinical implications, we aim to further characterize TP53g mutations in HBOC families. METHODS: Next-generation sequencing was conducted on 1876 breast cancer (BC) patients who fulfilled the inclusion criteria for HBOC. RESULTS: (Likely) pathogenic variants in TP53 gene were present in 0.6% of the BC cohort with higher occurrence in early onset BC < 36 years. (1.1%) and bilateral vs. unilateral BC (1.1% vs. 0.3%). Two out of eleven patients with a (likely) pathogenic TP53g variant (c.542G > A; c.375G > A) did not comply with classic LFS/Chompret criteria. Albeit located in the DNA-binding domain of the p53-protein and therefore revealing no difference to LFS-related variants, they only displayed a medium transactivity reduction constituting a retainment of wildtype-like anti-proliferative functionality. CONCLUSION: Among our cohort of HBOC families, we were able to describe a clinical subgroup, which is distinct from the classic LFS-families. Strikingly, two families did not adhere to the LFS criteria, and functional analysis revealed a reduced impact on TP53 activity, which may suit to the attenuated phenotype. This is an approach that could be useful in developing individualized screening efforts for TP53g mutation carrier in HBOC families. Due to the low incidence, national/international cooperation is necessary to further explore clinical implications. This might allow providing directions for clinical recommendations in the future.

Pre-operative cellular dissociation grading in biopsies is highly predictive of post-operative tumour stage and patient outcome in head and neck squamous cell carcinoma.

BACKGROUND: Pre-operative treatment planning in head and neck squamous cell carcinoma (HNSCC) is mainly dictated by clinical staging, which has major shortcomings. Histologic grading is irrelevant due to its lack of prognostic impact. Recently, a novel grading termed Cellular Dissociation Grade (CDG) based on Tumour Budding and Cell Nest Size was shown to be highly prognostic for resected HNSCC. We aimed to probe the predictive and prognostic impact of CDG in the pre-operative biopsies of HNSCC. METHODS: We evaluated CDG in n = 160 pre-therapeutic biopsies from patients who received standardised treatment following German guidelines, and correlated the results with pre- and post-therapeutic staging data and clinical outcome. RESULTS: Pre-operative CDG was highly predictive of post-operative tumour stage, including the prediction of occult lymph node metastasis. Uni- and multivariate analysis revealed CDG to be an independent prognosticator of overall, disease-specific and disease-free survival (p < 0.001). Hazard ratio for disease-specific survival was 6.1 (11.1) for nG2 (nG3) compared with nG1 tumours. CONCLUSIONS: CDG is a strong outcome predictor in the pre-treatment scenario of HNSCC and identifies patients with nodal-negative disease. CDG is a purely histology-based prognosticator in the pre-therapeutic setting that supplements clinical staging and may aide therapeutic stratification of HNSCC patients.

Systemic antitumor effect by regional hyperthermia combined with low-dose chemotherapy and immunologic correlates in an adolescent patient with rhabdomyosarcoma - a case report.

Introduction: An abscopal effect is a clinical observation whereby a local treatment is associated with regression of metastatic cancer at a site distant from the primary location of treatment. Here, we describe the clinical systemic effect induced by regional hyperthermia combined with low-dose chemotherapy and provide immunologic correlates.Case presentation: A 15-year-old patient had been diagnosed with alveolar rhabdomyosarcoma (ARMS). All previous treatment options failed in the patient including haploidentical stem cell transplantation and donor lymphocyte infusion. The patient presented with local and metastatic disease, and upon admission, underwent regional hyperthermia combined with low-dose chemotherapy. Immediately following therapy severe skin reactions were observed. Skin biopsies revealed an intraepithelial lymphocytic infiltration dominated by CD3+/CD8+ T cells with a regular network of dendritic cells. Clinical images compared before and during sequential treatment cycles showed complete metabolic response of the local tumor for more than 10 months of therapy. In addition, metastases completely regressed although they were not direct targets of regional hyperthermia. The systemic effect was associated with enhanced frequency of NK cells and T cells expressing the lectin-like natural-killer group 2 D activating receptor (NKG2D), an increase of the CD56bright subset of NK cells, as well as an increase of effector/memory and effector CD8+ and CD4+ T cells in the blood while the percentage of CD25+FOXP3+ regulatory T cells declined.Conclusions: Regional hyperthermia combined with low-dose chemotherapy had the potential to create a systemic effect which was associated with activation of NK cells and T cells.

SOX9 Knockout Induces Polyploidy and Changes Sensitivity to Tumor Treatment Strategies in a Chondrosarcoma Cell Line.

As most chemotherapeutic drugs are ineffective in the treatment of chondrosarcoma, we studied the expression pattern and function of SOX9, the master transcription factor for chondrogenesis, in chondrosarcoma, to understand the basic molecular principles needed for engineering new targeted therapies. Our study shows an increase in SOX9 expression in chondrosarcoma compared to normal cartilage, but a decrease when the tumors are finally defined as dedifferentiated chondrosarcoma (DDCS). In DDCS, SOX9 is almost completely absent in the non-chondroid, dedifferentiated compartments. CRISPR/Cas9-mediated knockout of SOX9 in a human chondrosarcoma cell line (HTB94) results in reduced proliferation, clonogenicity and migration, accompanied by an inability to activate MMP13. In contrast, adhesion, apoptosis and polyploidy formation are favored after SOX9 deletion, probably involving BCL2 and survivin. The siRNA-mediated SOX9 knockdown partially confirmed these results, suggesting the need for a certain SOX9 threshold for particular cancer-related events. To increase the efficacy of chondrosarcoma therapies, potential therapeutic approaches were analyzed in SOX9 knockout cells. Here, we found an increased impact of doxorubicin, but a reduced sensitivity for oncolytic virus treatment. Our observations present novel insight into the role of SOX9 in chondrosarcoma biology and could thereby help to overcome the obstacle of drug resistance and limited therapy options.