This is Part Two in a series. In Part One, learn how precision pathology is connected to clinical trials and the important factors in patient selection and enrollment.
A critical challenge in any clinical trial is patient stratification to ensure that the patient group being studied is diverse, yet accurately reflective of the patient population who will benefit from the treatment being investigated. Stratification entails dividing the potential patient group into subgroups, also referred to as ‘strata’ or ‘blocks’. Each strata represents a particular section of the patient population. Artificial intelligence can be employed to optimize patient stratification into subgroups that represent patients with diverse genetic backgrounds and particular pathogenesis. Reveal Biosciences provides advanced AI-powered image analysis pipelines that can accurately separate patients that are more likely to respond to a treatment, informed by specific biomarkers as well as from biomarker-agnostic tissue features.
Biomarkers are becoming an integral part of clinical trials as they are considered key tools in the identification of patient sub-populations most likely to benefit or conversely to incur adverse reactions from a given treatment. Reveal Biosciences offers more than 250 optimized immunohistochemistry assays that can detect and enumerate biomarkers with high precision and accuracy. Similarly, imageDx™ provides AI-powered digital assays that can analyze immunofluorescence of whole slide images prepared in-house or prepared by other laboratories.
Example of an 18-plex immunofluorescent image of a non-alcoholic fatty liver disease (NAFLD) sample analyzed by imageDx™ PRISM digital assay is shown in Fig. 3. imageDx provides a suite of digital assays that can provide cluster analysis of biomarkers and their network interactions as well as detailed cell neighborhood and tissue microenvironment analysis. Reveal’s suite of digital assays can assist with quantifying cell heterogeneity which is instrumental in analyzing immune cell infiltration.
Reveal Biosciences also provides precision pathology tools to predict responders and non-responders based on features that are hidden within the tissue. For example, imageDx™ offers a deep-learning based digital assay that can predict mismatch repair deficiency (dMMR) and microsatellite instability from hematoxylin and eosin (H&E) stained tissue images (Fig. 4). Recent studies have shown checkpoint blockade’s effectiveness in patients with mismatch repair–deficient tumors. In such trials, our AI-powered dMMR prediction model can add significant value to expedite patient selection and stratification.
Fig. 4 shows an example of imageDx™ digital assay applied to a colorectal cancer H&E image to identify MMR deficiency based on visually undetectable signals. The dMMR model uses a complex deep-learning algorithm, and it can generate attention heatmaps to provide insight into the inner workings of the algorithm.
Endpoint, Interpreting Results, and Measuring Outcomes
It is not uncommon in a clinical trial design to use pathological parameters as primary and secondary outcome measures, either as standalone classifiers or in combination with clinical data or other confirmatory laboratory procedures. Therefore, it is important to have standardized protocols with stringent quality assurance to deliver reproducible pathological data to assess the outcomes across groups. The inter- and intra- observer variability due to different pathologists’ education, training, and skills can complicate standardization of patient enrollment and interpreting the endpoint results.
In order to develop standardized protocols in clinical trials, precision pathology tools built in imageDx™ are trained on heterogeneous datasets that are screened by AI algorithms for quality control to ensure reproducible results. For each clinical study, these digital assays are built in collaboration with pathologists involved in the study for the specific outcome measures of that study. Therefore, pathologists can seamlessly integrate the final custom imageDx assay into their workflow over the course of the study. This collaborative model building approach provides the flexibility to adjust the model, in case there was a need to change an outcome measure or protocol in the clinical trial. AI strategies such as transfer learning techniques can also be used to repurpose these digital assays for other studies.
About the Author
Dr. Misagh Naderi is the manager of AI and digital assays at Reveal Biosciences, where he serves as a liaison between domain experts in pathology, artificial intelligence, and biopharma. Misagh’s scientific expertise spans across computational biology, biomedical sciences, and chemical engineering with publications in protein modeling, in-silico drug design, and oncolytic-virology.
Prior to joining Reveal Biosciences, Misagh served as scientific advisor at two prominent international law firms, focusing on intellectual property and innovations at the intersection of biology, engineering, and artificial intelligence. He has a passion for science communication and helping companies leverage AI in providing solutions that would impact human health and wellbeing.