Guest Post - Wanted: Your Opinion on Artificial Intelligence in Pathology

Today features a guest post from Dr. Phedias Diamandis of the University of Toronto:

Phedias Diamandis, MD, PhD

There is a growing body of evidence highlighting the utility of Artificial Intelligence (AI) in pathology. AI is a type of mathematical algorithm that allows computers to carry out human-like tasks considered “intelligent”, such as recognizing diagnostic histologic patterns or counting mitotic figures on digital H&E slides.  This has the potential to radically transform the clinical practice of pathologists. This anonymous survey was developed to understand the familiarity, enthusiasm, and concerns pathologists may have regarding this technology in their practice. Some anonymous demographic information is also requested to understand the relationship of these variables (e.g. specific age groups, practice types and speciality) to the responses of the variables. It is expected that findings of this study will help guide researchers to design AI workflows that meet the specific needs of the study participants.

This survey will take roughly 10-15 minutes to complete. It is anonymous and the investigators of the study have no conflicts of interest.

The survey can be accessed at the following link: https://www.surveymonkey.com/r/AIinpathology

Thank you for your participation! Please circulate to your colleagues. Everyone's opinion counts!

Automated lesion detection and classification in pathology: Upper Panels: A form of artificial intelligence known as deep convolutional neural networks (CNNs) is currently showing impressive results at pattern recognition tasks traditionally carried out by highly skilled humans. In this example, a full digital slide image is analyzed to highlight brain regions infiltrated by this oligodendroglioma. The IDH1-R132H immunostains (right upper panel) highlights where the tumor actually is for comparison. Lower panels: High power view of the same tumor showing it's infiltrative nature. The CNN clearly detects the infiltrative nature of this lesion. The right most panel shows the computer output of another metastatic lesion with a much more circumscribed border. In addition to identifying the lesions, CNNs are beginning to show promise at classifying tumors with different clinical outcomes.