[Artificial intelligence for pathology-how, where, and why?]

Artificial intelligence promises many innovations and simplifications in pathology, but also raises just as many questions and uncertainties. In this article, we provide a brief overview of the current status, the goals already achieved by existing algorithms, and the remaining challenges.

Postmortem Minimally Invasive Autopsy in Critically Ill COVID-19 Patients at the Bedside: A Proof-of-Concept Study at the ICU

BACKGROUND

Economic restrictions and workforce cuts have continually challenged conventional autopsies. Recently, the COVID-19 pandemic has added tissue quality and safety requirements to the investigation of this disease, thereby launching efforts to upgrade autopsy strategies.

METHODS

In this proof-of-concept study, we performed bedside ultrasound-guided minimally invasive autopsy (US-MIA) in the ICU of critically ill COVID-19 patients using a structured protocol to obtain non-autolyzed tissue. Biopsies were assessed for their quality (vitality) and length of biopsy (mm) and for diagnosis. The efficiency of the procedure was monitored in five cases by recording the time of each step and safety issues by swabbing personal protective equipment and devices for viral contamination.

FINDINGS

Ultrasound examination and tissue procurement required a mean time period of 13 min and 54 min, respectively. A total of 318 multiorgan biopsies were obtained from five patients. Quality and vitality standards were fulfilled, which not only allowed for specific histopathological diagnosis but also the reliable detection of SARS-CoV-2 virions in unexpected organs using electronic microscopy and RNA-expressing techniques.

INTERPRETATION

Bedside multidisciplinary US-MIA allows for the fast and efficient acquisition of autolytic-free tissue and offers unappreciated potential to overcome the limitations of research in postmortem studies.

How to use AI in pathology

AI plays an important role in pathology, both in clinical practice supporting pathologists in their daily work, and in research discovering novel biomarkers for improved patient care. Still, AI is in its starting phase, and many pathology labs still need to transition to a digital workflow to be able to enjoy the benefits of AI. In this perspective, we explain the major benefits of AI in pathology, highlight key requirements that need to be met and example how to use it in a typical workflow.

Pan-tumor T-lymphocyte detection using deep neural networks: Recommendations for transfer learning in immunohistochemistry

The success of immuno-oncology treatments promises long-term cancer remission for an increasing number of patients. The response to checkpoint inhibitor drugs has shown a correlation with the presence of immune cells in the tumor and tumor microenvironment. An in-depth understanding of the spatial localization of immune cells is therefore critical for understanding the tumor's immune landscape and predicting drug response. Computer-aided systems are well suited for efficiently quantifying immune cells in their spatial context. Conventional image analysis approaches are often based on color features and therefore require a high level of manual interaction. More robust image analysis methods based on deep learning are expected to decrease this reliance on human interaction and improve the reproducibility of immune cell scoring. However, these methods require sufficient training data and previous work has reported low robustness of these algorithms when they are tested on out-of-distribution data from different pathology labs or samples from different organs. In this work, we used a new image analysis pipeline to explicitly evaluate the robustness of marker-labeled lymphocyte quantification algorithms depending on the number of training samples before and after being transferred to a new tumor indication. For these experiments, we adapted the RetinaNet architecture for the task of T-lymphocyte detection and employed transfer learning to bridge the domain gap between tumor indications and reduce the annotation costs for unseen domains. On our test set, we achieved human-level performance for almost all tumor indications with an average precision of 0.74 in-domain and 0.72-0.74 cross-domain. From our results, we derive recommendations for model development regarding annotation extent, training sample selection, and label extraction for the development of robust algorithms for immune cell scoring. By extending the task of marker-labeled lymphocyte quantification to a multi-class detection task, the pre-requisite for subsequent analyses, e.g., distinguishing lymphocytes in the tumor stroma from tumor-infiltrating lymphocytes, is met.

Use of MS-GUIDE for identification of protein biomarkers for risk stratification of patients with prostate cancer

Background

Non-invasive liquid biopsies could complement current pathological nomograms for risk stratification of prostate cancer patients. Development and testing of potential liquid biopsy markers is time, resource, and cost-intensive. For most protein targets, no antibodies or ELISAs for efficient clinical cohort pre-evaluation are currently available. We reasoned that mass spectrometry-based prescreening would enable the cost-effective and rational preselection of candidates for subsequent clinical-grade ELISA development.

Methods

Using Mass Spectrometry-GUided Immunoassay DEvelopment (MS-GUIDE), we screened 48 literature-derived biomarker candidates for their potential utility in risk stratification scoring of prostate cancer patients. Parallel reaction monitoring was used to evaluate these 48 potential protein markers in a highly multiplexed fashion in a medium-sized patient cohort of 78 patients with ground-truth prostatectomy and clinical follow-up information. Clinical-grade ELISAs were then developed for two of these candidate proteins and used for significance testing in a larger, independent patient cohort of 263 patients.

Results

Machine learning-based analysis of the parallel reaction monitoring data of the liquid biopsies prequalified fibronectin and vitronectin as candidate biomarkers. We evaluated their predictive value for prostate cancer biochemical recurrence scoring in an independent validation cohort of 263 prostate cancer patients using clinical-grade ELISAs. The results of our prostate cancer risk stratification test were statistically significantly 10% better than results of the current gold standards PSA alone, PSA plus prostatectomy biopsy Gleason score, or the National Comprehensive Cancer Network score in prediction of recurrence.

Conclusion

Using MS-GUIDE we identified fibronectin and vitronectin as candidate biomarkers for prostate cancer risk stratification.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12014-022-09349-x.

Quantifying Heterogeneity in Tumors: Proposing a New Method Utilizing Convolutional Neuronal Networks

Heterogeneity is a hallmark of glioblastoma (GBM), the most common malignant brain tumor, and a key reason for the poor survival rate of patients. However, establishing a clinically applicable, cost-efficient tool to measure and quantify heterogeneity is challenging. We present a novel method in an ongoing study utilizing two convolutional neuronal networks (CNN). After digitizing tumor samples, the first CNN delimitates GBM from normal tissue, the second quantifies heterogeneity within the tumor. Since neuronal networks can detect and interpret underlying and hidden information within images and have the ability to incorporate different information sets (i.e. clinical data and mutational status), this approach might venture towards a next level of integrated diagnosis. It may be applicable to other tumors as well and lead to a more precision-based medicine.

(Re) Defining the High-Power Field for Digital Pathology

BACKGROUND

The microscope high-power field (HPF) is the cornerstone for histopathology diagnostic evaluation such as the quantification of mitotic figures, lymphocytes, and tumor grading. With traditional light microscopy, HPFs are typically evaluated by quantifying histologic events in 10 fields of view at × 400 magnification. In the era of digital pathology, new variables are introduced that may affect HPF evaluation. The aim of this study was to determine the parameters that influence HPF in whole slide images (WSIs).

MATERIALS AND METHODS

Glass slides scanned on various devices (Leica's Aperio GT450, AT2, and ScanScope XT; Philips UltraFast Scanner; Hamamatsu's Nanozoomer 2.0HT; and 3DHistech's P1000) were compared to acquired digital slides reviewed on each vendor's respective WSI viewer software (e.g., Aperio ImageScope, ImageScope DX, Philips IMS, 3DHistech CaseViewer, and Hamamatsu NDP.view) and an in-house developed vendor-agnostic viewer. WSIs were reviewed at "×40" equivalent HPF on different sized monitors with varying display resolutions (1900 × 1080-4500 × 3000) and aspect ratios (e.g., Food and Drug Administration [FDA]-cleared 27" Philips PS27QHDCR, FDA-cleared 24" Dell MR2416, 24" Hewlett Packard Z24n G2, and 28" Microsoft Surface Studio). Digital and microscopic HPF areas were calculated and compared.

RESULTS

A significant variation of HPF area occurred between differing monitor size and display resolutions with minor differences between WSI viewers. No differences were identified by scanner or WSIs scanned at different resolutions (e.g., 0.5, 0.25, 0.24, and 0.12 μm/pixel).

CONCLUSION

Glass slide HPF at × 400 magnification with conventional light microscopy was not equivalent to "×40" digital HPF areas. Digital HPF quantification may vary due to differences in the tissue area displayed by monitor sizes, display resolutions, and WSI viewers but not by scanner or scanning resolution. These findings will need to be further clinically validated with potentially new digital metrics for evaluation.

Epstein-Barr virus infection and altered control of apoptotic pathways in posttransplant lymphoproliferative disorders

Posttransplant lymphoproliferative disorders (PTLD) represent a spectrum of lymphoid diseases complicating the clinical course of transplant recipients. Most PTLD are Epstein-Barr virus (EBV) associated with viral latency type III. Several in vitro studies have revealed an interaction between EBV latency proteins and molecules of the apoptosis pathway. Data on human PTLD regarding an association between Bcl-2 family proteins and EBV are scarce. We analyzed 60 primary PTLD for expression of 8 anti- (Bcl-2, Bcl-XL, and Mcl-1) and proapoptotic proteins (Bak and Bax), the so-called BH3-only proteins (Bad, Bid, Bim, and Puma), as well as the apoptosis effector cleaved PARP by immunohistochemistry. Bim and cleaved PARP were both significantly (p = 0.001 and p = 5.251e-6) downregulated in EBV-positive compared to EBV-negative PTLD [Bim: 6/40 (15%), cleaved PARP: 10/43 (23%), vs. Bim: 13/16 (81%), cleaved PARP: 12/17 (71%)]. Additionally, we observed a tendency toward increased Bcl-2 protein expression (p = 0.24) in EBV-positive PTLD. Hence, we provide evidence of a distinct regulation of Bcl-2 family proteins in EBV-positive versus negative PTLD. The low-expression pattern of the proapoptotic proteins Bim and cleaved PARP together with the high-expression pattern of the antiapoptotic protein Bcl-2 by trend in EBV-positive tumor cells suggests disruption of the apoptotic pathway by EBV in PTLD, promoting survival signals in the host cells.

MethMarker: user-friendly design and optimization of gene-specific DNA methylation assays

A software workflow to translate known differentially methylated regions into clinical biomarkers

DNA methylation is a key mechanism of epigenetic regulation that is frequently altered in diseases such as cancer. To confirm the biological or clinical relevance of such changes, gene-specific DNA methylation changes need to be validated in multiple samples. We have developed the MethMarker http://methmarker.mpi-inf.mpg.de/ software to help design robust and cost-efficient DNA methylation assays for six widely used methods. Furthermore, MethMarker implements a bioinformatic workflow for transforming disease-specific differentially methylated genomic regions into robust clinical biomarkers.

Fifteen years of env C2V3C3 evolution in six individuals infected clonally with human immunodeficiency virus type 1

The study of the evolution of human immunodeficiency virus type 1 (HIV-1) requires blood samples collected longitudinally and data on the approximate time point of infection. Although these requirements were fulfilled in several previous studies, the infectious sources were either unknown or heterogeneous genetically. In the present study, HIV-1 env C2V3C3 (nt 7029-7315) evolution was examined retrospectively in a cohort of hemophiliacs. Compared to other cohorts, the area of interest here was the infection of six hemophiliacs by the same virus strain, that is, the infecting viruses shared an identical genome. As expected, divergence from the founder sequence as well as interpatient divergence of the predominant virus strains increased significantly over time. Based on the V3 nucleotide sequences, CCR5 usage was predicted exclusively throughout the whole period of infection in all patients. Interestingly, common patterns of viral evolution were detected in the patients of the cohort. Four amino acid substitutions within the V3 loop emerged and persisted subsequently in five (positions 305 and 308 of the HXB2 gp120 reference sequence) and six patients (positions 325 and 328 in HXB2 gp120), respectively. These common changes within the V3 loop are likely to be enforced by HIV-1 specific immune response.