Speaking the Same Language

Negative stereotypes about pathologists have long plagued the field. The average person may imagine a pathologist as someone who spends most of their time practicing medicine, alone, in front of a microscope – someone who only communicates through report writing, who chose this specialty because they don't want the challenges that come with treating patients. Now, imagine another stereotypical professional – an introvert sat at their computer, writing code to analyze images. The professional is very good at what they do, but chooses not to be bothered by other people. A computer scientist. 

Computational pathology is a new medical discipline that combines pathology with computer science, and more precisely, computer vision. In the classical workflow, pathology images come from glass slides that are viewed through the eyepiece of a microscope. Now, in the form of whole slide images, they can be viewed on the computer screen – but more importantly, they can be analyzed with computer algorithms!

There is an opportunity to automate repetitive tasks such as cell counting, and even unlock molecular properties of analyzed tissue without the need to perform expensive tests. To maximize this potential, we need one thing: a close collaboration between pathologists and computer scientists. The pathologist needs to learn about image analysis tools and the scoring systems. The computer scientist needs to understand the pathology workflow, terminology, and have a basic understanding of different tissue components.

When collaborating on a computational pathology project, both the pathologist and the computer scientist take out their respective notes on computer vision and pathology and hope to work together smoothly… However, they soon realize there is a rather big expertise gap – and therefore a communication gap – between the two professions. 

I experienced this divide when I started working at a digital pathology company. Brilliant programmers and bioimage analysis scientists were doing their best to develop tissue image analysis algorithms without actually understanding the tissue. When I joined I had to learn a lot about image analysis, the tools, and the strengths and limitations of different methods – including classical and deep learning image analysis approaches and the post-processing of images.

I also realized that to optimize the analysis for a particular output, I had to live with the fact that the algorithm was not always perfect in the irrelevant parts of tissue. (For a pathologist, it is a big visual discordance when tissue image analysis markups do not match the actual tissue – we are trained for years to spot things that don’t fit).

I realized that this knowledge gap was not unique to my workplace. My computer vision colleagues also needed to acquire the relevant pathology knowledge to work independently on their projects. We just needed to know enough about each other’s fields to figure out the best solution for our tissue image analysis problems. We needed a bridge. This missing bridge was a result of a siloed approach to science, combined with the highly specialized and individual nature of the professions involved in the discipline of computational pathology. So, I created a pathology training program for non-pathologists involved in tissue image analysis.

I founded my blog – The Digital Pathology Place – in 2019. Fast-forward four years, the platform has expanded to a podcast, a YouTube channel, and a series of courses. My mission is to bridge the gap between pathology and computer science, advance digital pathology, and ultimately improve patient care.

When I started my digital pathology journey, it was a bit of a lonely place, but now, the landscape looks a lot better. The recent Pathology Vision Conference organized by the Digital Pathology Association (DPA) sold out to over 800 people! The vendors are engaging in conversations with users, and the users are initiating new developments in this space. Pathologists are starting their own digital pathology companies and are an integral part of digital pathology startups. The DPA is constantly expanding and there are now five digital pathology podcasts (there were none when I started podcasting in 2019)! The digital pathology vendors are providing excellent webinars, training, and courses.

Even though collective expertise is growing, there are always more beginners. And as a community, it is our responsibility to spread knowledge and bring everyone up to speed as fast as possible. Constant communication between computer scientists and pathologists is key to maintaining a seamless digital pathology workflow.