Digital pathology involves the conversion of tissue samples from glass slides to digital images, improving diagnostic efficiency while providing an infrastructure for computational pathology, which is based on quantitative measurement, mathematical modeling, and the development of algorithms for machine learning. This, in turn, can enhance the interpretation of disease processes and allow for the integration of genetic and clinical information with the morphometric analysis, offering a higher level of understanding of tumor pathology. Implementing a fully digital workflow will allow this enhancement in diagnosis through computer-augmented diagnostic algorithms and decision-support systems.
The Warren Alpert Center was established in 2017, but the groundwork started with the purchase of the department’s first whole-slide imaging scanner in 2008, based on the innovative foresight of pathologist Victor Reuter and his early interest in novel technologies.
Establishing a Fully Digital Pathology Workflow
Under the leadership of Dr. Reuter and Surgical Pathology Service Chief Meera Hameed, the Department of Pathology implemented prospective digital scanning for archiving in 2015, having successfully established a secure image management workflow to facilitate access to the digital slides via the department’s laboratory information system. The department currently has a digital archive of more than 300,000 slides from approximately 100,000 cases. Soon the team will be scanning 40,000 slides per month as well as beginning to scan MSK’s glass slide archive.
As MSK’s clinical operations have expanded beyond the hospital’s main campus, the Department of Pathology explored telepathology technology in order to provide the necessary pathology support to the various locations. Cytology Service Chief Oscar Lin and Director of Pathology Informatics S. Joseph Sirintrapun led the charge in incorporating telecytology for remote adequacy assessments to support MSK’s regional sites. The opening of the Josie Robertson Surgery Center in 2016 allowed the department to validate and use telepathology for intraoperative frozen section consultations. The experience gained from viewing digital slides for the review of prior material made it easier for MSK pathologists to adapt to a telepathology workflow. The digital archive also helped provide immediate access to prior pathology material, minimizing the need to transport glass slides between sites.
Novel Imaging Technologies for Digital Pathology
The Warren Alpert Center’s Digital Imaging Laboratory at the Josie Robertson Surgery Center provides an incubator to evaluate new technologies that advance digital pathology in a robust clinical setting. Through active engagement with industry partners, MSK is helping improve the technology and guide clinical applications. Interdepartmental collaborations with medical physics, informatics, clinical services (e.g., surgery), and radiology enhance cooperation and create opportunities for multidisciplinary applications.
Yukako Yagi is The Warren Alpert Center’s Director of Pathology Digital Imaging. With 20 years in the field of digital pathology, Dr. Yagi has extensive first-hand experience and in-depth knowledge of the technology and its applications.
Imaging technician Alexei Teplov loads a specimen for a micro–computed tomography (micro-CT) scan in Dr. Yagi’s laboratory.
Recent advances, such as super-resolution imaging with the ability to resolve anatomic structures in unprecedented detail, have been fostered by the development of optical techniques to allow in situ visualization of molecular events, making micro-level versions of MRI, CT, SPECT, and PET feasible. Similar to 3-D reconstruction of radiographic images, the application of 3-D imaging techniques in histopathology enables a unique new way to visualize and quantify tissue. One of Dr. Yagi’s major efforts involves high-resolution 3-D imaging to microscopically examine portions of tissue without the need to prepare tissue sections. Dr. Yagi is developing 3-D imaging capabilities at the single-cell level and establishing correlations with gross images, radiology, and clinical information. Dr. Yagi was the first to coin the term “whole-block imaging” and is leading efforts to develop a data management system for all modalities used in evaluating tissue samples.
Tissue and cell-level visualization by these modalities — as well as others being developed for use in pathology, such as Raman and MALDI imaging — represent the next frontier for collaboration between pathology and radiology. Long-term efforts are geared toward the development of technologies for direct microscopic visualization of tissue without the need to remove it, section it, and create glass slides. The applications of such technology include in vivo microscopy, or “virtual biopsy.”
High-Performance Compute Cluster for Pathology Image and Data Management
A fully digital pathology workflow will yield millions of digital images over the next few years, resulting in petabytes of image data. The Warren Alpert Center will support the extension of MSK’s compute cluster and HPC infrastructure for pathology data and image management. This HPC infrastructure enables computational pathology at scale and provides the ability to digitize a vast number of clinically annotated glass slides, which can then be used for deep machine learning and validating disease-specific algorithms. Modern machine-learning algorithms, such as deep neural networks and Bayesian models, are often trained for weeks on hundreds of compute nodes to build clinically relevant decision-support systems.
MSK as the Leader in Computational Pathology
The practical use of digital pathology for primary diagnosis requires the development of useful analytic algorithms to compensate for the reduced efficiency of slide review that is inherent in the current digital-slide-interface applications. MSK is uniquely positioned to develop these algorithms while addressing the user experience. Examples are nuclear detection, nuclear classification, tissue segmentation, staining estimation, and quantifying morphology. These features form the basis of the higher-level tasks of cancer detection, grading, staging, micrometastasis detection, and quantifying immune cell infiltration. Computational pathology research has shown promising results that will further enrich an overall knowledge of disease. By integrating computational pathology data with other specimen-related data (genomics, proteomics, radiographic imaging, etc.), the team can bring an unprecedented breadth and depth of information to each individual case, yielding a comprehensive, multidimensional analysis that would otherwise be impossible.
One of the major goals of The Warren Alpert Center is to enable basic research in this area and develop deep-learning models that can either predict genetic mutations based on tissue morphology or find new diagnostic patterns by correlating patient outcome with tissue morphology.
Computational pathology enables pathologists to be faster, more efficient, and more accurate by supplanting subjective criteria with objective criteria. Decision-support tools can help identify regions of interest, match unusual morphologic features among similar cases, and rapidly calculate microscopic measurements, cell counts, and other quantitative metrics. This work facilitates large-scale, quantitative correlations between tumor characteristics, protein expression, and genetic panels, such as MSK-IMPACT™.