Contributed Commentary by Amanda Hemmerich, IQVIA Laboratories
October 4, 2024 | It is no surprise that clinical trial sponsors are slowly embracing new tools and technologies to achieve greater levels of precision and efficiency needed in an increasingly demanding and competitive drug discovery and development landscape. Where there was once hesitation due to the unknowns of straying from historical analytic processes, there has been exponential growth in the application of digital pathology within clinical research and development efforts in recent years.
The biopharma industry’s strong interest in digital pathology stems from its continued focus on cancer care and related precision medicine. Oncology remains the largest therapeutic focus of R&D activity, with 2,143 Phase I to III trials started or planned in 2023 and 25% of these trials focused on examining novel oncology mechanisms, such as cell and gene therapies and other immunotherapies. This and the growing interest in artificial intelligence/machine learning to improve efficiencies in clinical trials have helped pathology experts embrace new tools and methodologies as new partners that extend their capabilities to provide quality patient care.
Digitized pathology solutions, including scanners, high-resolution imaging tools and AI-driven data management strategies, are providing deeper insights from each tissue sample. Noted below are several ways digitized pathology is enhancing clinical trial programs, especially by helping sponsors meet the unique needs of this era of precision medicine.
Deeper Insights From A Single Sample
Traditional pathology has used glass slides under a microscope with a single hematoxylin and eosin stain to determine the root cause of a patient’s disease. As the field of pathology evolved, tissue was examined with other stains to categorize the various cells under the microscope. Though useful, this practice had the potential to use most of the patient’s sample, which could lead to needing additional procedures to collect more tissue to gather more details for disease classification.
Now, with the ability to digitize glass H&E slides, we are continuously learning how to use AI/ML to extract more granular information from one H&E slide and allow more to be done with a single sample.
Deeper Insights From Classifiers
Expert lab teams can now rely on machine learning algorithms referred to as “classifiers” to assign a class label (e.g., tumor cells, CD8 positive T-cells, etc.) to each data input. These classifier algorithms receive training data that labels images. After sufficient training, the classifier can receive unlabeled images as data inputs and output classification labels for each image. Employing sophisticated mathematical and statistical methods, a trained classifier can generate predictions about the likelihood of a data input being classified in a given way.
These algorithms are developed with either a single indication in mind or as a deep neural network to be applied to multiple indications. Both methods have been useful in multiple clinical instances. For example, a trained classifier can quantify a variety of details from samples, including:
All of these intricate aspects help the pathologist develop additional care approaches for the patient.
Deeper Insights For Case Prioritization
With the tissue classifier, the computer can rank cases assigned to a pathologist to identify which ones may be of greater urgency to prioritize earlier in their day. This allows the pathologist to deliver critical results to the care team that much quicker.
Deeper Insights For Molecular Testing
Cancer care has come a long way, with personalized medicine recognizing that each patient has a particular molecular signature to their cancer that can affect treatment outcomes. The use of AI/ML tools intends to help the cancer care team identify this information as quickly as possible. First, AI/ML can help the pathologist by selecting the most appropriate area for molecular testing. It can identify the area on the digital image and tell the microdissection laser where exactly to dissect the sample for testing. Secondly, AI/ML can provide predictive algorithms for what the molecular signature may be with the H&E image alone.
Deeper Insights About Microenvironments
Pathologists are able to describe the tissue environment using both the H&E stain and additional immunohistochemistry stains. Though this has led to valuable insights within the biomarker field, in particular with the checkpoint inhibitors PD-1/PDL-1-targeted monoclonal antibodies, deeper understanding of other aspects within the microenvironment can exhaust the majority of the tissue sample. As trial sponsors begin to identify new biomarkers, researchers have also explored how AI/ML capabilities can help to extract more data from a single H&E stain. Clinical researchers are starting to rely on these tech-enabled solutions to quantify exact numbers of tumor-infiltrating lymphocytes and classify the type of lymphocytes from a single slide. At the IASLC 2024 World Conference on Lung Cancer, a presentation by AstraZeneca, Daiichi Sankyo and Roche Diagnostics demonstrated the utility of computation pathology for biomarker scoring. It utilized an algorithm to quantify protein expression from an IHC stain within tumor cells to determine which patient would benefit from treatment.
Another strategy for learning more about the microenvironment focuses on multiplex testing with digital methods on a single slide. Traditional pathology evaluation has needed multiple slides (H&E and IHC stain slides) to classify each cell, which requires the bulk of the tissue sample and potentially the entire sample. Some enhanced AI/ML tools have worked around using a single slide for all traditional immunofluorescence stains (similar to IHC stains). In these cases, the pathology team can initially process the slide, then scan it, process the slide again and scan it a final time. This allows additional cell classification with traditional stains.
Upholding Quality Results Via Digital Pathology Validation
As with all lab systems, it is critical that pathology labs validate their digital pathology workflow, from scanners to image management software, image analysis platforms, monitors, etc., to ensure consistency in sample handling and processing.
Greater Than The Sum Of Its Parts: Increasing Value For Patient Care
Advanced digitized pathology solutions are continuously being fine-tuned to offer deeper insights, ensuring precision medicine further focuses on root causes and locations of debilitating diseases. Being in a world of narrow — not general — AI requires specialized experts in pathology, technology and clinical trials to bring nuanced knowledge and skills together to accelerate precision medicine’s impact for patients in need.
These advances in technology are designed to enhance the pathologist’s well-trained eyes because they can provide the relational context and make sense of deeper calculations and measurements. Looking forward, this is only the beginning of what may come with digitized pathology. R&D stakeholders can expect to have expanding opportunities to enhance and tailor-fit image analysis processes and secure subtle, yet deeper, insights to meet the unique needs of sub-patient populations.
Amanda Hemmerich, Pathologist, Genomics, IQVIA Laboratories, is board certified in Anatomic and Clinical Pathology and Cytopathology. Amanda brings her expertise in the development of innovative digital pathology techniques and managing large-scale pathology services to her current role. She has deep experience in molecular testing, immunohistochemistry and genomic test validation and operationalization with a focus on gastrointestinal and liver pathology. She can be reached at amanda.hemmerich@iqvia.com.