The Devil’s Advocate

“Regular” pathologists have a somewhat nebulous view of digital pathology – it’s an emerging technology that may be able to help them with some aspects of their workflow. Many of them have not completely embraced digital pathology yet, though. They are waiting to see more widespread adoption and ease of regulatory burden – perhaps quite wisely. Computational pathology is an even more uncertain prospect; most people don’t yet understand the term very well. Many pathologists do understand digital image analysis and artificial intelligence, but that doesn’t necessarily mean that they are believers in the technology – at least, not yet.

Why would anyone feel hesitant about, or even threatened by, digital and computational pathology? In my opinion, the main reason is that many pathologists feel that computers are not ready to make complex diagnoses yet – which is certainly true; they’re only equipped to provide assistance to the pathologist whose job it is (and whose job will remain) to make such diagnoses. Others feel that by supporting such technology, they put their own jobs at stake; why should they “train” the computers to take away their work? Still others worry that digitization may result in the outsourcing of their jobs to other countries, or to commercial labs with established computational pathology pipelines. Viewed from the perspective of the individual pathologist, especially one without the resources to remain fully up to date with new technologies, it’s no surprise that computational pathology is intimidating!

I think, though, that the true advantages and disadvantages of these technologies are slightly different. Computational pathology’s key benefit is that it offers an integrated view of the patient. The software can produce high-quality images that include annotations and clinical metadata, so that the pathologist can see the complete picture. Along with computer-based quantitative and qualitative image analysis, the overall practice of pathology becomes more objective – and thereby also more reproducible. And, of course, every pathologist is familiar with the tedium of counting and segmenting; why not assign these tasks to computers, so that we can do the more challenging, high-level diagnostic tasks?

I’m not worried that a computer will take my job. Why? Well, the kinds of computers that can provide practical diagnostic assistance are not cheap. They also require large numbers of annotated images to learn, so training them is no simple task. And many of the diagnoses we make in the clinical setting are too complex – or too dependent on morphology and molecular data – to entrust to even the most advanced computer.

My only real concern, in fact, is that humans may begin to rely on computers too much. We must take care not to become complacent and stop checking the computer’s results properly. At the end of the day, the responsibility for the patient’s diagnosis still falls at our feet, not the virtual feet of our software.

Often, those who develop or market new technologies feel compelled to tell us how good those technologies are, but that isn’t always the right approach; hearing how computers can do everything a pathologist can do – or even better – is threatening for some. It’s far more useful to learn how technology can help us in our daily workflow, how it can assist us in sharing and analyzing images, and ultimately improve the speed and quality of our diagnoses.

Those of us who regularly use computational pathology tools can help, too. We can talk about the practical benefits we’ve encountered. I, for example, particularly benefit from diagnostic algorithms, telepathology, and help with quantitating biomarkers and structures. Such tools have helped me to reduce error and facilitate a digital workflow. They also enable researchers and clinical teams – whether down the hall, in a different building, or on opposite sides of the globe – to share and review images instantly, then use algorithms to obtain valuable insights that ultimately lead to a more informed, more detailed diagnosis and personalized care plan for each patient. In my experience, combining pathologists and computers only results in better pathology with powerful tools for diagnosis and decision support.