Digital pathology in the medical laboratory

Nov. 21, 2017

Artificial Intelligence is on its way to your digital toolbox

Digital pathology (DP) is accepted technology for the use of whole slide images (WSI) in education, research, and consultation. With the first FDA approval earlier this year for use of WSIs in primary diagnosis, technologies supporting DP will progress to include additional use cases to improve the pathologist’s workflow. Rather than becoming a “replacement” for the microscope, digital pathology will offer analytic tools to improve diagnostic and prognostic accuracy.

Artificial Intelligence (AI) will be deployed to produce results both practical and far-reaching. Leveraging AI within digital images will produce productivity gains with automated identification and classification of key features within the images. The automated classification of features will range from Normal/Abnormal to more complex classifications over time. Pathologists will be able to let the computer support their efforts, minimizing mundane and repetitive tasks of case review.

Digital analysis of immunohistochemistry stains can also provide rapid, reproducible quantification of Her2, PD-L1, Ki67, and other biomarkers. Serial section images stained for different cellular components can be compared and evaluated side by side at the pathologist’s workstation. Pathologist-to-pathologist variation is eliminated, and pathologists can then use this information to refine their diagnoses to provide the most accurate result possible.

The potential applications for AI are limitless. DP and the use of AI will not only improve diagnostic accuracy for pathology specimens; they will provide a wealth of information to
improve patient prognoses and outcomes.

— Robin Weisburger, MS, HTL(ASCP)
Manager, Client Support Services

Utilizing digital pathology for immunotherapy

Digital pathology (DP) is currently a timely intervention in the field of oncology, and it can act as a checkpoint for immune therapy. Current digitization, archival and image analysis tools, as well as developing computational technologies support the development of companion diagnostics in immuno-oncology, with an emphasis on the tumor microenvironment. These new technologies are enhancing the detection, segmentation, feature extraction, and tissue classification within immune therapy.

A DP system, which will integrate morphology-based feature extraction and molecular technology, will serve physicians as they attempt to understand the tumor microenvironment activity to allow them to predict response, or resistance to, immune check-point inhibitors. A plethora of unique molecular tests are available today for whole genome and whole transcriptome sequencing of both normal and cancer tissues. The detection and presence of protein biomarkers in patient tumor cells will enable personalized cancer care for patients with actionable insights for potential treatment options. A tumor sample, i.e., a biopsy sample, is required for testing and the accurate identification of cellular alterations, which requires digitization from whole slide scanners to isolate the individual tumor cells from the biopsy specimen.

In conclusion, a systematic application of a methodology to harness histopathology data of the tumor, and its corresponding molecular signatures, could be made possible with the use of DP to address immunotherapy complexities.

— Anagha Jadhav, MD
Director of Digital Pathology

Pathology goes mobile

The medical community has seen a surge in the development of advanced mobile technologies enabled through smart phones and other mobile devices. Medical applications on mobile devices are changing the face of medicine as we know it—and digital pathology (DP) is the perfect vehicle to enable these radical changes for cancer detection, interpretation, and treatment. In DP, mobile devices are being used to acquire and transmit digital images to bypass traditional forms of telepathology. The smart phone utility has presented a unique opportunity for DP consultations in rural and developing regions, and can act as a mobile management system for digital images and metadata.

New mobile applications are designed to target the clinical workflow for all users, including histotechnicians, pathologists, and telepathologists. These applications will provide critical time-savings and value, specifically for time-sensitive consultations. Mobile applications will be aimed at hospitals and hospital networks, clinical labs, reference labs, and consulting pathologists. Mobile applications revolving around DP will be seamlessly integrated into standard laboratory workflows for uploading digital pathology images, creating case hierarchies with multiple case slides and metadata, updating case histories, and annotating specimen descriptions. These applications would provide users with “anytime, anywhere access to the lab,” 24/7, and empower the pathologist to collaborate and consult with other experts globally, in real-time.

— Isha Doshi
Systems Analyst

The outlook for digital pathology using deep learning

With the advancements in digital pathology (DP), high volumes of quality digitized data are available for algorithm developers, scientists, and computational pathologists around the world, and the integration of telepathology has allowed for real-time collaboration to continue the development of these algorithms. With the advent of cloud computing and high-end processor resources, the environment is conducive for novel approaches to image analysis challenges.

We can acknowledge the limitations of handcrafted features using conventional machine learning approaches—deep learning (DL) takes feature engineering to the next level by automating the process. There are DL methodologies to directly learn from the raw data and map to the intended goals. The combination of handcrafted features and DL-discovered features can result in reproducible and accurate outcomes for prediction or classification objectives. In theory, the system could be partly supervised for initial training done with the labelled ground truth, followed by the system using this training for unsupervised learning.

The development of convoluted neural networks (CNNs, modelled by the human brain), as a deep network for analyzing and classifying image patterns, has revolutionized medical imaging through DL. According to recent studies, CNNs are the basis of some outstanding breakthroughs in the analysis of DP images with applications, for example, for mitosis detection, nucleus segmentation, gland segmentation, and metastasis detection.

— Guru Kamble
Head of Imaging