Casey Laris, CTO of Reveal Biosciences, lead a webinar entitled “Revolutionize Pathology using Artificial Intelligence” as part of the Science Exchange Webinar Series.
Highlights from this webinar include:
- Reveal Biosciences specializes in data-driven pathology. The company offers a range of histopathology, immunohistochemistry (IHC), in situ hybridization (ISH), whole slide imaging and advanced image analysis services.
- ImageDx is a cloud-based pathology platform combining image management, training, inference and visualization for whole slide images and scalable image analysis.
- Reveal’s computational team develops data-driven machine learning technology for the analysis of whole slide pathology images. This technology is used to generate quantitative data from pathology samples for preclinical research, clinical trials, and as decision support tools to assist pathologists diagnosing patients.
Quality data requires quality images. Reveal has artificial intelligence that automatically performs image QC to detect and exclude tissue artifacts and segment cell nuclei. These pre-processing steps significantly improve the statistical performance of downstream disease- or assay-specific analysis applications.
- Artifact detection – image QC involves automatically detecting and excluding any regions classified as out of focus, tissue artifacts, folds, tears, etc. to improve the quality of data.
- Nuclear segmentation – a pixel level deep learning nuclear segmentation approach leveraging Reveal’s large repository of whole slide images, results in higher accuracy than the more traditional machine learning-based approach that is common within the industry. This is particularly true in tissue with dense nuclei or regions containing nuclei of different shapes and sizes.
A broad range of disease-specific or assay-specific models are then deployed as required including models focused on immuno-oncology, NASH, neuroscience, IHC, ISH, and more. Multi-modal models are also available that combine pathology data with data from other sources (clinical outcome, RNA-Seq, biomarker etc.).
At the end of the presentation, participants asked a range of questions including:
Q. Can ImageDx be used to study biomarker localization?
A. Yes, ImageDx can provide data from a range of biomarker assays including IHC and ISH. Additional data such as co-localization, proximity etc. can also be provided.
Q. Do your models take into account lab-to-lab variation?
A. Variation in staining intensity between different labs is common, and is something we have extensive experience with and data to address. We have specifically developed modern AI training workflows that integrate variation in staining intensity to prepare for and minimize reported quantification variance.
Q. How long does data analysis take?
A. ImageDx can process images in parallel, therefore reducing the time required to process large amounts of data. This approach allows multiple images to be analyzed in the same time as it takes to analyze a single image. This massively parallel approach offers a great deal of scale and flexibility for incorporating the highest performing models into the analysis of large datasets.
Q. Does ImageDx work for different image formats?
A. ImageDx works well with all the common image formats including images from Aperio, 3DHistech, Phillips, Hamamatsu, and others. Images can be uploaded to the site for analysis or storage.
To find out more, contact: [email protected]
Casey Laris, CTO leads the machine learning development of AI-based diagnostics at Reveal. He is a world leader in high-throughput biological computer vision and has been at the forefront of analyzing big data sets from automated microscopy in pathology. Casey developed patented high-throughput microscopy technology at UCSD and productized it in a venture-backed start-up that was later acquired by Beckman-Coulter. There he was Global Product Manager for high content microscopy, reagents and software, leading both the world-wide roll-out and follow-on clinical ready instrumentation development. Casey helped further develop these computer vision tools at Sanford-Burnham Medical Research Institute with commercialization by Biovia. Casey lead the applications and engineering team at Vala Sciences to create several novel automated high throughput microscope based systems. Casey holds a BS from UC San Diego.