AI in pathology: Enhancing clinical decision making

• Pathologists are a cornerstone of diagnostics, but the profession is facing mounting pressure with declining numbers of pathologists and an increasing workload
• Advances in AI-enabled tools could help alleviate pressure on pathologists and healthcare systems
• Widespread adoption of AI-enabled technologies requires change management, and can benefit from commercial partnerships

In recent years, pathologists have faced mounting pressure as the profession grapples with declining numbers in the workforce, an aging physician population, and an increasingly complex workload. This challenging environment takes a detrimental toll on pathologists, which can threaten the quality of patient care.¹ To address these strains, the field is increasingly turning toward AI in pathology as a means to enhance efficiency and support diagnostic decision-making.

In our latest interview, Healthcare Transformers spoke to Bruno Occhipiniti, CEO of Qritive, about his vision for empowering pathologists with AI-enabled tools designed to improve efficiency, accuracy, and confidence in decision making.

HT: What is the biggest problem you see in the field of pathology?

Bruno Occhipinti: There is an imbalance in care access in the world of diagnostics. If you look at the statistics, we’ve got roughly 100,000 pathologists globally, in 130 countries.² Meanwhile, we’ve got 20 million new cancer cases per year, and this is increasing year on year.³

This situation leads to a number of issues. For example, a patient may have to wait for a long time before getting their results. At times, they may have to pay more to get access to those results. Even more problematic is that, in some instances, the results they get are inaccurate. This is because pathologists are under immense pressure to deliver on the volume that they have to handle, and mistakes can be made. Then, if you look more broadly, you find that 47% of the population of the world doesn’t have access to diagnostics.⁴ So fundamentally speaking, you’ve got a care access issue.

HT: How does AI in pathology, in other words, solutions like yours address the access issues that you mention?

Bruno Occhipinti: The whole idea of these solutions is that they will be a clinical decision support system for pathologists. AI solutions can really accelerate the process of diagnosis and make pathologists capable of handling the large volumes that they are faced with. They gain in speed, and they gain in accuracy as well. It is really quite difficult to do both at the same time, but technology can enable that.

AI-powered solutions can minimize the time required to detect disease presence and grade it in whole slide images, simply because of the way data and analysis are managed. You can fast-track the whole process and get results to patients quicker. We use a model that can reduce the time taken to detect a metastatic deposit by 90%, and improve sensitivity to 100% from 95.9%.⁵

With AI, you can also improve on accuracy. We have done a number of studies with the Singapore General Hospital where we proved that we could gain up to 82% in accuracy.⁶ AI can analyze tissues and spot a very small presence of disease that may have been missed, then flag it for doctors, allowing them to make a more accurate decision.

You could say, “But I can just train more pathologists.” The problem is that it’s not a very easy task to become a pathologist. It’s a long period of study, then 5 to 10 years of practice to build experience before being comfortable enough to operate at speed. That’s where AI can facilitate the process and unlock access to diagnostics. AI will dramatically augment the capabilities of the pathologists so that they can do their job with a higher level of accuracy and at a higher speed.

HT: Have you faced any challenges to adoption in the market?  

Bruno Occhipinti: It’s a new approach to embrace tech and AI, and particularly in the field of diagnostics, so that comes with a number of challenges. Right now, you have a glass slide on which you have a sample of the body tissues, and it’s being analyzed on a microscope, which has been the reference tool for the past few decades. The conversion into a digital asset requires a slide scanner, and then once it’s a digital asset, AI can analyze it. But not every lab or hospital has this, so it requires investment.

The second challenge is the fact that for many of the hospitals and labs, when it comes to buying AI, it’s still very, very new to them, and they don’t really have an understanding of how to qualify, how to compare, and how to choose the solution. It seems very trivial, but it extends the sales cycle dramatically.

Another consideration is the regulatory aspect. There are certain countries that are extremely open and very advanced when it comes to defining the regulations around those products, and other locations where that’s far from the case. The categorization is either not there, or if it’s there, it can change overnight, so your regulatory approval can become invalid. This dramatically delays access to the market.

The most important aspect challenging adoption, I believe, is that these solutions are still new to the population of clinicians. There are clinicians who very quickly understood the benefits of having this kind of companion process, this kind of guidance and decision support. However, some doctors are struggling with the idea of having AI in pathology because there is no proper training for most doctors about how it works and what its capabilities are.

Some of them may have a fear of replacement, asking, “Am I promoting something that could affect my own role, and my own importance in the healthcare ecosystem?” So there’s a lot of work that needs to be done to educate clinicians and highlight what AI does well and what AI is not designed to do and will not do.

However, as many parts of the world are now facing shortages in pathologists, making it even harder to deal with the surge in cancer cases, the willingness to adopt AI-based solutions for diagnostics is evolving. The Asia-Pacific region, in particular, is struggling with only 6.8 pathologists for 1 million patients versus 48.8 in North America.⁷ Consequently, it makes it a favorable location for the adoption of digital pathology and AI in pathology.  Also, considering the different perceptions among healthcare institutions about architecture and cybersecurity risk, it is important to provide the flexibility to operate seamlessly on the cloud or on-prem. By the same token, the focus in terms of disease areas reflects the reality of cancer progression in Asia.  South East Asia in particular is expected to suffer a significant increase in cancer occurrences, with a growth of 89.2% among men, and 65.6% among women by 2050.⁸ In this context, the use of AI to enable affordable and accurate diagnostics at scale is even more pressing in this part of the world.

HT: Do you think change management has a role to play in driving adoption for AI in pathology?

Bruno Occhipinti: I think it’s vital. Obviously, technology is key, and you need to get a solution that makes sense, that has a performance level you expect, and the speed you expect, and so on. But the biggest challenge for me at the end of the day is adoption.

You want to make sure that whoever takes the decision to adopt the AI will fully understand how AI is going to operate in the system. It’s not just that you can plug and play and then things will magically become better. You need to make sure that the whole usage of AI has been clearly defined because you can use AI in different ways. You could say, “I just want AI to confirm my assessment, nothing else,” or, “I want to go full on digital,” and not use a microscope at all because you choose to analyze the digital whole slide image and not the physical glass slides, and you trust the AI’s level of sensitivity as a guidance in your own assessment. Those are very different perspectives. You have to really think through how you want the use of AI in pathology to fit into your environment, what makes sense, and be really comfortable that this is going to address all the needs that you typically have.

You are inviting people to use a completely new technology and solution. This requires change management to be taken very, very seriously, and actually even more seriously than the pure technical considerations. You have to put in place mechanics to ensure training is done properly and a support system that allows doctors to ask questions and get quick feedback. You have to ensure that there is transparency on how AI is being used within the system, and how it brings clarity to the data that you generate. These are complex issues that arise and have to be considered against your current mechanics.

HT: What do you see as the biggest benefits of the adoption of AI in pathology?

Bruno Occhipinti: With this type of technology, you can expect the time to treatment to be much quicker, and the risk of mistakes will be minimized quite dramatically. We’ve also done a pretty extensive study that showed that when you have AI in the loop, the discordance between specialists can be resolved much faster because AI plays this game of tiebreaker and aligns the different perspectives much quicker.⁶ So there are a number of obvious gains in terms of the patient journey itself. 

But there are other levels of change that we are only seeing the very beginning of. For example, we are developing a solution for prostate prognosis. So it’s not just about the detection of disease, but how AI is going to interpret the future evolution of the patient. With this kind of AI technology in place, you will be able to know which patients require very regular biopsies and which ones should be very, very carefully tracked, and in which ones we can delay biopsies. That will have a massive impact on the patient journey as well because for some of those patients, it means far fewer biopsies, which makes their lives better. It also benefits the healthcare ecosystem because it means less costs. 

Additionally, there are benefits for clinicians. It’s a very lonely task to be a pathologist, and make very pressurized decisions. When you look at lymph node metastasis and you try to spot metastatic deposits, you know that if you make a mistake, you’re going to compromise quite substantially the life of a patient. That’s a huge burden on any person. The fact that you have an additional pair of eyes, which in this case is AI, on top of the other specialists that you might be working with, brings comfort and support to a very lonely and extremely demanding effort. Whatever can help pathologists with that and make their task much more manageable is going to bring immense value to the clinicians and the patients they are serving across the world.

In our partnership with Roche, an open ecosystem approach makes these processes even easier for clinicians since they end up using one interface with algorithms from different AI providers, depending on their needs. Streamlining the interface also dramatically facilitates the pace of adoption.

HT: Is partnering with larger companies important for smaller companies like Qritive?

Bruno Occhipinti: Large multi-national companies that have an understanding of the customer are helpful for us because they can provide feedback from clinicians on the challenges that they are facing. This helps guide the type of technologies we will prioritize and the way we deliver those solutions to clinicians, so that there is easier adoption.

Global businesses have immense financial power, and they also have the global clout that we don’t have, so they are helping us to tap into some of that. They may present opportunities that we have not seen, and provide a chance to meet the decision makers at the right time. Also, because large companies are so strong in terms of branding, the fact that we are working alongside them has a halo effect on what we bring to the market because people will associate us with that established brand and feel more comfortable. If those established brands have chosen us, they know that they can trust us as well, and that is fundamental. At the end of the day, it’s healthcare, it’s about trust. So, endorsement by a large, trusted multinational can have a massive, massive impact on small AI providers.