Background: Machine learning models provide significant opportunities for improvement in health care, but their “black-box” nature poses many risks. Methods: We built a custom Python module as part of a framework for generating artifacts that are meant to be tunable and describable to allow for future testing needs. We conducted an analysis of a previously published digital pathology classification model and an internally developed kidney tissue segmentation model, utilizing a variety of generated artifacts including testing their effects. The artifacts simulated were bubbles, tissue folds, uneven illumination, marker lines, uneven sectioning, altered staining, and tissue tears. Results: We found that there is some performance degradation on the tiles with artifacts, particularly with altered stains but also with marker lines, tissue folds, and uneven sectioning. We also found that the response of deep learning models to artifacts could be nonlinear. Conclusions: Generated artifacts can provide a useful tool for testing and building trust in machine learning models by understanding where these models might fail.

Background: With the emergence of whole slide imaging (WSI) and widespread access to high-speed Internet, pathology labs are now poised to implement digital pathology as a way to access diagnostic pathology expertise. This paper describes a collaborative partnership between a high-volume reference diagnostic laboratory (Labcorp) and an academic pathology department (Mount Sinai Hospital) in the transition from a traditional glass slide service to a digital platform. Using the standard framework of implementation science, we evaluate the consistency and quality of the Philips IntelliSite Pathology Solution (PIPS) in delivering save and efficient diagnostic services. Materials and Methods: Digital and glass slide diagnoses of all consult cases were documented over a 12-month period. The Proctor guideline was used to quantitatively and qualitatively measure (e.g., focus group studies, field notes, and administrative data) implementation success. Lean techniques (e.g., value stream mapping) were applied to measure changes in efficiency with the transition to a digital platform. Results: Our study supports the acceptability, high adoption, appropriateness, feasibility, fidelity, and sustainability of the digital pathology platform. The digital portal also improved the quality of patient care by increasing efficiency, effectiveness, safety, and timeliness. The intraobserver concordance rate was 100%. The digital transition resulted in a reduction in turnaround time from 86 h to an average 35 min and a 20-fold increase in efficiency of the consultation process. Conclusion: As the pathology community contemplates digital pathology as a transformational tool in providing broad access to diagnostic expertise across time and space, our study provides an implementation strategy along with evidence that the digital platform is safe, effective, and efficient.

Multiparametric fluorescence imaging through CODEX allows the simultaneous imaging of many biomarkers in a single tissue section. While the digital fluorescence data thus obtained can provide highly specific characterizations of individual cells and microenvironments, the images obtained are different from those usually interpreted by pathologists (i.e., hematoxylin and eosin [H&E] slides and 3,3′-diaminobenzidine-stained immunohistochemistry slides). Having the fluorescence data plus coregistered H&E or similar data could facilitate the adoption of multiparametric imaging into regular workflows, as well as facilitate the transfer of algorithms and machine learning previously developed around H&E slides. Since commercial CODEX instruments do not produce H&E-like images by themselves, we developed a staining protocol and associated image processing to make “virtual H&E” images that can be incorporated into the CODEX workflow. While there are many ways to achieve virtual H&E images, including the use of a fluorescent nuclear stain and tissue autofluorescence to simulate eosin staining, we opted to combine fluorescent nuclear staining (through 4′,6-diamidino-2-phenylindole) with actual eosin staining. We also output images derived from fluorescent nuclear staining and autofluorescence images for additional evaluation.

Introduction: Manual gating of flow cytometry (FCM) data for marrow cell analysis is a standard approach in current practice, although it is time- and labor-consuming. Recent advances in cytometry technology have led to significant efforts in developing partially or fully automated analysis methods. Although multiple supervised and unsupervised FCM data analysis algorithms have been developed, they have not been widely adopted by the clinical and research laboratories. In this study, we evaluated flowDensity, an open source freely available algorithm, as an automated analysis tool for classification of lymphocyte subsets in the bone marrow biopsy specimens. Materials and Methods: FlowDensity-based gating was applied to 102 normal bone marrow samples and compared with the manual analysis. Independent expression of each cell marker was assessed for comprehensive expression analysis and visualization. Results: Our findings showed a correlation between the manual and flowDensity-based gating in the lymphocyte subsets. However, flowDensity-based gating in the populations with a small number of cells in each cluster showed a low degree of correlation. Comprehensive expression analysis successfully identified and visualized the lymphocyte subsets. Discussion: Our study found that although flowDensity might be a promising method for FCM data analysis, more optimization is required before implementing this algorithm into day-to-day workflow.

Background: Scanners are the main tool in digital pathology. The technical abilities of scanners determine the workflow logic in the pathology laboratory. Its performance can be restricted by the divergence between the scanning time presented by the manufacturer and the actual scanning time. This could lead to critical deviations from the established business processes in a 24/7 laboratory. Aim: Our investigation is focused in exploring the performance of three main models of high-performance scanners available on the Russian market: 3DHistech, Hamamatsu и Leica. Objectives: We compared the performance of the scanners on the samples of a given size with the manufacturer's stated specifications and evaluated the speed of the scanners on the reference and routine laboratory material. Subjects and Methods: We examined 3DHistech Pannoramic 1000, Hamamatsu NanoZoomer s360 and Leica AT2 with default settings and automatic mode. Two sets of glasses were used (glass slide): Group 1 included 120 slides with 15 mm × 15 mm slices, Group 2 included 120 workflow slides. Results: The average slide scan times in Groups 1 and 2 for the C13220 (156 ± 1.25 s and 117 ± 4.17 s) and Pannoramic 1000 (210 ± 1.64 s and 183 ± 3.78 s) differ statistically significantly (P < 0.0001). Total scanning time including rack reloading was shorter for the workflow slide set group for the modern C13220 and Pannoramic 1000 scanner models. Conclusions: The scanner specifications provided by manufacturers are not sufficient to evaluate the performance. The guidelines and regulations concerning scanner selection should be consented by the digital pathology community. We suggest discussing criteria for evaluating scanner performance.

Background: Plasma cell neoplasm and/or plasma cell myeloma (PCM) is a mature B-cell lymphoproliferative neoplasm of plasma cells that secrete a single homogeneous immunoglobulin called paraprotein or M-protein. Plasma cells accumulate in the bone marrow (BM) leading to bone destruction and BM failure. Diagnosis of PCM is based on clinical, radiologic, and pathological characteristics. The percent of plasma cells by manual differential (bone marrow morphology), the white blood cell (WBC) count, cytogenetics, fluorescence in situ hybridization (FISH), microarray, and next-generation sequencing of BM are used in the risk stratification of newly diagnosed PCM patients. The genetics of PCM is highly complex and heterogeneous with several genetic subtypes that have different clinical outcomes. National Comprehensive Cancer Network guidelines recommend targeted FISH analysis of plasma cells with specific DNA probes to detect genetic abnormalities for the staging of PCM (4.2021). Recognition of risk categories through training software for classification of high-risk PCM and a novel way of addressing the current approaches through bioinformatics will be a significant step toward automation of PCM analysis. Methods: A new artificial neural network (ANN) classification model was developed and tested in Python programming language with a first data set of 301 cases and a second data set of 176 cases for a total of 477 cases of PCM at diagnosis. Classification model was also developed with support vector machines (SVM) algorithm in R studio and interactive data visuals using Tableau. Results: The resulting ANN algorithm had 94% accuracy for the first and second data sets with a classification summary of precision (PPV): 0.97, recall (sensitivity): 0.76, f1 score: 0.83, and accuracy of logistic regression of 1.0. SVM of plasma cells versus TP53 revealed a 95% accuracy level. Conclusion: A novel classification model based only on specific morphological and genetic variables was developed using a machine learning algorithm, the ANN. ANN identified an association of WBC and BM plasma cell percentage with two of the high-risk genetic categories in the diagnostic cases of PCM. With further training and testing of additional data sets that include morphologic and additional genetic rearrangements, the newly developed ANN model has the potential to develop an accurate classification of high-risk categories of PCM.

Background: Frozen section (FS) diagnosis is one of the promising applications of digital pathology (DP). However, the implementation of an appropriate and economically viable DP solution for FS in routine practice is challenging. The objective of this study was to establish the non-inferiority of whole-slide imaging (WSI) versus optical microscopy (OM) for FS diagnosis using a low cost and portable DP system. Materials and Methods: A validation study to investigate the technical performance and diagnostic accuracy of WSI versus OM for FS diagnosis was performed using 60 FS cases[120 slides i.e, 60 hematoxylin and eosin (H & E) and 60 toluidine blue (TOLB)]. The diagnostic concordance, inter- and intra-observer agreement for FS diagnosis by WSI versus OM were recorded. Results: The first time successful scanning rate was 89.1% (107/120). Mean scanning time per slide for H and E and TOLB slide was 1:47 min (range; 0:22–3: 21 min) and 1:46 min (range; 0:21–3: 20 min), respectively. Mean storage space per slide for H and E and TOLB slide was 0.83 GB (range: 0.12–1.73 GB) and 0.71 GB (range: 0.11–1.66 GB), respectively. Considering major discrepancies, the overall diagnostic concordance for OM and WSI, when compared with the reference standard, was 95.42% and 95.83%, respectively. There was almost perfect intra as well as inter-observer agreement (k ≥ 0.8) among 4 pathologists between WSI and OM for FS diagnosis. Mean turnaround time (TAT) of 14:58 min was observed using WSI for FS diagnosis, which was within the College of American Pathologists recommended range for FS reporting. The image quality was average to best quality in most of the cases. Conclusion: WSI was noninferior to OM for FS diagnosis across various specimen types. This portable WSI system can be safely adopted for routine FS diagnosis and provides an economically viable alternative to high-end scanners.

Background: Web-based digital slide viewers for pathology commonly use OpenSlide and OpenSeadragon (OSD) to access, visualize, and navigate whole-slide images (WSI). Their standard settings represent WSI as deep zoom images (DZI), a generic image pyramid structure that differs from the proprietary pyramid structure in the WSI files. The transformation from WSI to DZI is an additional, time-consuming step when rendering digital slides in the viewer, and inefficiency of digital slide viewers is a major criticism for digital pathology. Aims: To increase efficiency of digital slide visualization by serving tiles directly from the native WSI pyramid, making the transformation from WSI to DZI obsolete. Methods: We implemented a new flexible tile source for OSD that accepts arbitrary native pyramid structures instead of DZI levels. We measured its performance on a data set of 8104 WSI reviewed by 207 pathologists over 40 days in a web-based digital slide viewer used for routine diagnostics. Results: The new FlexTileSource accelerates the display of a field of view in general by 67 ms and even by 117 ms if the block size of the WSI and the tile size of the viewer is increased to 1024 px. We provide the code of our open-source library freely on https://github.com/schuefflerlab/openseadragon. Conclusions:This is the first study to quantify visualization performance on a web-based slide viewer at scale, taking block size and tile size of digital slides into account. Quantifying performance will enable to compare and improve web-based viewers and therewith facilitate the adoption of digital pathology.

Background: Artificial intelligence has an emerging progress in diagnostic pathology. A large number of studies of applying deep learning models to histopathological images have been published in recent years. While many studies claim high accuracies, they may fall into the pitfalls of overfitting and lack of generalization due to the high variability of the histopathological images. Aims and Objects: Use the model training of osteosarcoma as an example to illustrate the pitfalls of overfitting and how the addition of model input variability can help improve model performance. Materials and Methods: We use the publicly available osteosarcoma dataset to retrain a previously published classification model for osteosarcoma. We partition the same set of images into the training and testing datasets differently than the original study: the test dataset consists of images from one patient while the training dataset consists images of all other patients. We also show the influence of training data variability on model performance by collecting a minimal dataset of 10 osteosarcoma subtypes as well as benign tissues and benign bone tumors of differentiation. Results: The performance of the re-trained model on the test set using the new partition schema declines dramatically, indicating a lack of model generalization and overfitting. We show the additions of more and moresubtypes into the training data step by step under the same model schema yield a series of coherent models with increasing performances. Conclusions: In conclusion, we bring forward data preprocessing and collection tactics for histopathological images of high variability to avoid the pitfalls of overfitting and build deep learning models of higher generalization abilities.

Background: Cervical intraepithelial neoplasia (CIN) is regarded as a potential precancerous state of the uterine cervix. Timely and appropriate early treatment of CIN can help reduce cervical cancer mortality. Accurate estimation of CIN grade correlated with human papillomavirus type, which is the primary cause of the disease, helps determine the patient's risk for developing the disease. Colposcopy is used to select women for biopsy. Expert pathologists examine the biopsied cervical epithelial tissue under a microscope. The examination can take a long time and is prone to error and often results in high inter-and intra-observer variability in outcomes. Methodology: We propose a novel image analysis toolbox that can automate CIN diagnosis using whole slide image (digitized biopsies) of cervical tissue samples. The toolbox is built as a four-step deep learning model that detects the epithelium regions, segments the detected epithelial portions, analyzes local vertical segment regions, and finally classifies each epithelium block with localized attention. We propose an epithelium detection network in this study and make use of our earlier research on epithelium segmentation and CIN classification to complete the design of the end-to-end CIN diagnosis toolbox. Results: The results show that automated epithelium detection and segmentation for CIN classification yields comparable results to manually segmented epithelium CIN classification. Conclusion: This highlights the potential as a tool for automated digitized histology slide image analysis to assist expert pathologists.

Background: Despite increasing interest in whole-slide imaging (WSI) over optical microscopy (OM), limited information on comparative assessment of various digital pathology systems (DPSs) is available. Materials and Methods: A comprehensive evaluation was undertaken to investigate the technical performance–assessment and diagnostic accuracy of four DPSs with an objective to establish the noninferiority of WSI over OM and find out the best possible DPS for clinical workflow. Results: A total of 2376 digital images, 15,775 image reads (OM - 3171 + WSI - 12,404), and 6100 diagnostic reads (OM - 1245, WSI - 4855) were generated across four DPSs (coded as DPS: 1, 2, 3, and 4) using a total 240 cases (604 slides). Onsite technical evaluation revealed successful scan rate: DPS3 < DPS2 < DPS4 < DPS1; mean scanning time: DPS4 < DPS1 < DPS2 < DPS3; and average storage space: DPS3 < DPS2 < DPS1 < DPS4. Overall diagnostic accuracy, when compared with the reference standard for OM and WSI, was 95.44% (including 2.48% minor and 2.08% major discordances) and 93.32% (including 4.28% minor and 2.4% major discordances), respectively. The difference between the clinically significant discordances by WSI versus OM was 0.32%. Major discordances were observed mostly using DPS4 and least in DPS1; however, the difference was statistically insignificant. Almost perfect (κ ≥ 0.8)/substantial (κ = 0.6–0.8) inter/intra-observer agreement between WSI and OM was observed for all specimen types, except cytology. Overall image quality was best for DPS1 followed by DPS4. Mean digital artifact rate was 6.8% (163/2376 digital images) and maximum artifacts were noted in DPS2 (n = 77) followed by DPS3 (n = 36). Most pathologists preferred viewing software of DPS1 and DPS2. Conclusion: WSI was noninferior to OM for all specimen types, except for cytology. Each DPS has its own pros and cons; however, DPS1 closely emulated the real-world clinical environment. This evaluation is intended to provide a roadmap to pathologists for the selection of the appropriate DPSs while adopting WSI.

Objectives: Bone marrow evaluation plays a critical role in the diagnosis, staging, and monitoring of many diseases. Although there are standardized guidelines for assessing bone marrow specimen quality, there is a lack of evidence-based tools to perform such assessments. The objective was to monitor bone marrow sample quality in real time by standardizing the basic components of a synoptic report and incorporating it into a bone marrow report template. Materials and Methods: A relational database of bone marrow quality parameters was developed and incorporated into our laboratory information system bone marrow report template, with data entry completed during specimen sign out. Data from multiple reports created within a date range were extracted by Structured Query Language query, and summarized in tabular form. Reports generated from these data were utilized in quality improvement efforts. Results: The synoptic reporting system was routinely used to record the quality of bone marrow specimens from adult patients. Data from 3189 bone marrow aspirates, 3302 biopsies, and 3183 biopsy touch imprints identified hemodilution as the principal issue affecting bone marrow aspirate quality, whereas aspiration artifact and fragmentation affected bone marrow biopsy quality. Conclusions: The bone marrow synoptic reporting process was easy to use, readily adaptable, and has proved a useful component of the overall quality assurance process to optimize bone marrow quality.

Three-dimensional (3D) imaging is increasingly being incorporated into a variety of medical specialties: surgery and radiology being but two prominent examples. Image-intensive disciplines, such as anatomic pathology (AP), represent excellent potential candidates for further exploration of this innovative technology. Multiple potential use cases exist within AP, involving patient care, education, and research. These use cases broadly include direct utilization of the 3D digital assets for viewing on a 2D screen, populating 3D extended reality platforms (virtual reality, augmented reality, and mixed reality) as well as generation of 3D printed photorealistic specimen models. Herein, these use cases are explored with specific regard to our experiences and yet unrealized potential. Future directions and considerations are also discussed.

We believe the switch to a digital pathology (DP) workflow is imminent and it is essential to understand the economic implications of conversion. Many aspects of the adoption of DP will be disruptive and have a direct financial impact, both in short term costs, such as investment in equipment and personnel, and long term revenue potential, such as improved productivity and novel tests. The focus of this whitepaper is to educate pathologists, laboratorians and other stakeholders about the business and monetary considerations of converting to a digital pathology workflow. The components of a DP business plan will be thoroughly summarized, and guidance will be provided on how to build a case for adoption and implementation as well as a roadmap for transitioning from an analog to a digital pathology workflow in various laboratory settings. It is important to clarify that this publication is not intended to list prices although some financials will be mentioned as examples. The authors encourage readers who are evaluating conversion to a DP workflow to use this paper as a foundational guide for conducting a thorough and complete assessment while incorporating in current market pricing. Contributors to this paper analyzed peer-reviewed literature and data collected from various institutions, some of which are mentioned. Digital pathology will change the way we practice through facilitating patient access to expert pathology services and enabling image analysis tools and assays to aid in diagnosis, prognosis, risk stratification and therapeutic selection. Together, they will result in the delivery of valuable information from which to make better decisions and improve the health of patients.

Background: Observer studies in pathology often utilize a limited number of representative slides per case, selected and reported in a nonstandardized manner. Reference diagnoses are commonly assumed to be generalizable to all slides of a case. We examined these issues in the context of pathologist concordance for histologic subtype classification of ovarian carcinomas (OCs). Materials and Methods: A cohort of 114 OCs consisting of 72 cases with a single representative slide (Group 1) and 42 cases with multiple representative slides (148 slides, 2-“6 sections per case, Group 2) was independently reviewed by three experts in gynecologic pathology (case-based review). In a follow-up study, each individual slide was independently reviewed in a randomized order by the same pathologists (section-based review). Results: Average interobserver concordance varied from 100% for Group 1 to 64.3% for Group 2 (86.8% across all cases). Across Group 2, 19 cases (45.2%) had at least one slide classified as a different subtype than the subtype assigned from case-based review, demonstrating the impact of intratumoral heterogeneity. Section-based concordance across individual sections from Group 2 was comparable to case-based concordance for those cases indicating diagnostic challenges at the individual section level. Findings demonstrate the increased diagnostic complexity of heterogeneous tumors that require multiple section sampling and its impact on pathologist performance. Conclusions: The proportion of cases with multiple representative slides in cohorts used in validation studies, such as those conducted to evaluate artificial intelligence/machine learning tools, can influence diagnostic performance, and if not accounted for, can cause disparities between research and real-world observations and between research studies. Case selection in validation studies should account for tumor heterogeneity to create balanced datasets in terms of diagnostic complexity.

Background: The development of artificial intelligence (AI) in pathology frequently relies on digitally annotated whole slide images (WSI). The creation of these annotations – manually drawn by pathologists in digital slide viewers – is time consuming and expensive. At the same time, pathologists routinely annotate glass slides with a pen to outline cancerous regions, for example, for molecular assessment of the tissue. These pen annotations are currently considered artifacts and excluded from computational modeling. Methods: We propose a novel method to segment and fill hand-drawn pen annotations and convert them into a digital format to make them accessible for computational models. Our method is implemented in Python as an open source, publicly available software tool. Results: Our method is able to extract pen annotations from WSI and save them as annotation masks. On a data set of 319 WSI with pen markers, we validate our algorithm segmenting the annotations with an overall Dice metric of 0.942, Precision of 0.955, and Recall of 0.943. Processing all images takes 15 min in contrast to 5 h manual digital annotation time. Further, the approach is robust against text annotations. Conclusions: We envision that our method can take advantage of already pen-annotated slides in scenarios in which the annotations would be helpful for training computational models. We conclude that, considering the large archives of many pathology departments that are currently being digitized, our method will help to collect large numbers of training samples from those data.

Background: Recently, research data are increasingly shared through social media and other digital platforms. Traditionally, the influence of a scientific article has been assessed by the publishing journal's impact factor (IF) and its citation count. The Altmetric scoring system, a new bibliometric that integrates research “mentions” over digital media platforms, has emerged as a metric of online research distribution. The aim of this study was to explore the relationship of the Altmetric Score with IF and citation number within the pathology literature. Methods: Citation count and Altmetric scores were obtained from the top 10 most-cited articles from the 15 pathology journals with the highest IF for 2013 and 2016. These variables were analyzed and correlated with each other, as well as the age of the publishing journal's Twitter account. Results: Three hundred articles were examined from the two cohorts. The total citation count of the articles decreased from 21,043 (2013) to 14,679 (2016), while the total Altmetric score increased from 830 (2013) to 4066 (2016). In 2013, Altmetric score weakly correlated with citation number (r = 0.284, P < 0.001) but not with journal IF (r = 0.024, P = 0.771). In 2016, there was strong correlation between citation count and Altmetric Score (r = 0.714, P < 0.0001) but not the IF (r = 0.0442, P = 0.591). Twitter was the single most important contributor to the Altmetric score; however, the age of the Twitter account was not associated with citation number nor Altmetric score. Conclusions: In the pathology literature studied, the Altmetric score correlates with article citation count, suggesting that the Altmetric score and conventional bibliometrics can be treated as complementary metrics. Given the trend towards increasing use of social media, additional investigation is warranted to evaluate the evolving role of social media metrics to assess the dissemination and impact of scientific findings in the field of pathology.

Introduction: Hologic is developing a digital cytology platform. An educational website was launched for users to review these digitized Pap test cases. The aim of this study was to analyze data captured from this website. Materials and Methods: ThinPrep^® Pap test slides were scanned at ×40 using a volumetric (14 focal plane) technique. Website cases consisted of an image gallery and whole slide image (WSI). Over a 13 month period data were recorded including diagnoses, time participants spent online, and number of clicks on the gallery and WSI. Results: 51,289 cases were reviewed by 918 reviewers. Cytotechnologists spent less time (M [Median] = 65.0 s) than pathologists (M = 82.2 s) reviewing cases (P < 0.001). Longer times were associated with incorrect diagnoses and cases with organisms. Cytotechnologists matched the reference diagnoses in 85% of cases compared to pathologists who matched in 79.8%. While in 62% of cases reviewers only examined the gallery, they attained the correct diagnosis 92.7% of the time. Pathologists made more clicks on the gallery and WSI than cytotechnologists (P < 0.001). Diagnostic accuracy decreased with increasing clicks. Conclusions: Website participation provided feedback about how cytologists interact with a digital platform when reviewing cases. These data suggest that digital Pap test review when comprised of an image gallery displaying diagnostically relevant objects is quick and easy to interpret. The high diagnostic concordance of digital Pap tests with reference diagnoses can be attributed to high image quality with volumetric scanning, image gallery format, and ability for users to freely navigate the entire digital slide.

Background: The tumor microenvironment is highly heterogeneous, and it is understood to affect tumor progression and patient outcome. A number of studies have reported the prognostic significance of tumor-infiltrating lymphocytes and tumor budding in colorectal cancer (CRC). However, the significance of the intratumoral heterogeneity present in the spatial distribution of these features within the tumor immune microenvironment (TIME) has not been previously reported. Evaluating this intratumoral heterogeneity may aid the understanding of the TIME's effect on patient prognosis as well as identify novel aggressive phenotypes which can be further investigated as potential targets for new treatment. Methods: In this study, we propose and apply two spatial statistical methodologies for the evaluation of the intratumor heterogeneity present in the distribution of CD3 ⁺ and CD8 ⁺ lymphocytes and tumor buds (TB) in 232 Stage II CRC cases. Getis-Ord hotspot analysis was applied to quantify the cold and hotspots, defined as regions with a significantly low or high number of each feature of interest, respectively. A novel spatial heatmap methodology for the quantification of the cold and hotspots of each feature of interest, which took into account both the interpatient heterogeneity and the intratumor heterogeneity, was further developed. Results: Resultant data from each analysis, characterizing the spatial intratumor heterogeneity of lymphocytes and TBs were used for the development of two new highly prognostic risk models. Conclusions: Our results highlight the value of applying spatial statistics for the assessment of the intratumor heterogeneity. Both Getis-Ord hotspot and our proposed spatial heatmap analysis are broadly applicable across other tissue types as well as other features of interest. Availability: The code underpinning this publication can be accessed at https://doi.org/10.17630/c2306fe9-66e2-4442-ad89-f986220053e2.

Aims: Histology, the microscopic study of normal tissues, is a crucial element of most medical curricula. Learning tools focused on histology are very important to learners who seek diagnostic competency within this important diagnostic arena. Recent developments in machine learning (ML) suggest that certain ML tools may be able to benefit this histology learning platform. Here, we aim to explore how one such tool based on a convolutional neural network, can be used to build a generalizable multi-classification model capable of classifying microscopic images of human tissue samples with the ultimate goal of providing a differential diagnosis (a list of look-alikes) for each entity. Methods: We obtained three institutional training datasets and one generalizability test dataset, each containing images of histologic tissues in 38 categories. Models were trained on data from single institutions, low quantity combinations of multiple institutions, and high quantity combinations of multiple institutions. Models were tested against withheld validation data, external institutional data, and generalizability test images obtained from Google image search. Performance was measured with macro and micro accuracy, sensitivity, specificity, and f1-score. Results: In this study, we were able to show that such a model's generalizability is dependent on both the training data source variety and the total number of training images used. Models which were trained on 760 images from only a single institution performed well on withheld internal data but poorly on external data (lower generalizability). Increasing data source diversity improved generalizability, even when decreasing data quantity: models trained on 684 images, but from three sources improved generalization accuracy between 4.05' and 18.59'. Maintaining this diversity and increasing the quantity of training images to 2280 further improved generalization accuracy between 16.51' and 32.79'. Conclusions: This pilot study highlights the significance of data diversity within such studies. As expected, optimal models are those that incorporate both diversity and quantity into their platforms.s

Background: The COVID-19 pandemic accelerated the widespread adoption of digital pathology (DP) for primary diagnosis in surgical pathology. This paradigm shift is likely to influence how we function routinely in the postpandemic era. We present learnings from early adoption of DP for a live digital sign-out from home in a risk-mitigated environment. Materials and Methods: We aimed to validate DP for remote reporting from home in a real-time environment and evaluate the parameters influencing the efficiency of a digital workflow. Eighteen pathologists prospectively validated DP for remote use on 567 biopsy cases including 616 individual parts from 7 subspecialties over a duration from March 21, 2020, to June 30, 2020. The slides were digitized using Roche Ventana DP200 whole-slide scanner and reported from respective homes in a risk-mitigated environment. Results: Following re-review of glass slides, there was no major discordance and 1.2% (n = 7/567) minor discordance. The deferral rate was 4.5%. All pathologists reported from their respective homes from laptops with an average network speed of 20 megabits per second. Conclusion: We successfully validated and adopted a digital workflow for remote reporting with available resources and were able to provide our patients, an undisrupted access to subspecialty expertise during these unprecedented times.