Contributed Commentary by David West, Jr.

July 5, 2017 | Pathology departments and translational research centers have amassed invaluable, untapped information sitting on glass. A wealth of potential medical discoveries are hidden in dusty old cabinets in laboratory storerooms. Digital pathology offers an opportunity to scan this tissue into whole slide images (WSI), so the slides can now be viewed by pathologists anywhere in the world, enabling collaboration between pathologists in a way never before possible.

But the availability of a WSI via Internet-connection is only the first stage of what is likely to become a much bigger shift. Once digitized, it becomes possible to apply a variety of computational tools to assist pathologists in analyzing the tissue.

The advent of slide digitization has shifted pathology into the era of computational medicine by creating the opportunity to quantify and integrate tissue data to supplement the existing cancer model centered around human expertise, corresponding patient history, and “-omic” data.

There are two sparks making this transformation possible. The first is processing power, democratized by cloud computing, which means efficient and plentiful computing resources can be applied to problems in a way never before possible.

The second is the introduction of Convolutional Neural Networks (CNNs). In 2011, three researchers used a new kind of machine learning to win an international image classification competition. Alex Krizhevsky and Ilya Sutskever (under the supervision of Geoffrey Hinton) didn’t just win the competition, they beat five other prestigious teams in dramatic fashion. Fueled by the availability of efficient computing power, this new approach has opened up a new field of deep learning, especially when applied to images and pattern recognition.

Because of these advances, it is possible to train intelligent algorithms to recognize broad or specific patterns on a whole slide image, and translate features evident in the tissue into prediction (e.g. metastasis, recurrence, etc.) and classification (i.e. staging, grading, differential diagnosis) data. The technology is still in its early days, but we’ll likely see a broad range of applications of deep learning for pathology.

While there is much to be discovered through image analysis, integrating what can be seen with findings from other fields of analysis and patient history could be equally powerful. The era of precision medicine was ushered in via high-throughput molecular assays, identifying patients with certain phenotypes and genotypes that correspond with significant improvement in overall and disease-free survival after certain neoadjuvant treatments. Such molecular biomarkers effectively stratify patient populations for treatments such as PD-L1 checkpoint inhibition, which has shown to significantly improve overall and disease-free survival for advanced melanoma patients with CD8+ T-cells located in the invasive tumor margin. A larger vision of computational medicine would marry the image analysis capabilities of digital pathology to the capabilities of molecular assays – leveraging data encoded in images to drive precise surgical procedures and therapeutic plans.

With the collaboration of partners in the clinical realm and in informatics, deep learning powered digital pathology tools will soon be able to activate digital slides to address problems in the clinic, create opportunities for translational research and data licensing, and inform disease prognosis and therapeutic plans.

David West, Jr. is the CEO of Proscia, a digital pathology software company. Mr. West founded the company in 2014, and has played a key role in business development, sales, and marketing, as well as the early design and development of Proscia’s software. David graduated Johns Hopkins University where he studied Biomedical Engineering with a focus on Computational Biology. He can be reached at info@proscia.com