April 12, 2023

Examining the Impact of AI on Pathology: A Deep-Dive into the CONFIDENT-P Study

Following the recent launch of highly advanced artificial intelligence (AI) tool ChatGPT, global interest in better understanding the scope of AI’s power has swelled. Though AI for pathology is not a new phenomenon, that interest has extended into the medical space; pathologists are more curious than ever to learn how pathology AI works and how it can offer them value. To explore this, University Medical Center (UMC) Utrecht and Paige have joined forces to conduct research on the use of pathology AI in clinical practice. The first study, called the CONFIDENT-Prostate study, is set to begin this month. The study will employ AI to help pathologists evaluate real-world prostate biopsies, with the aim of assessing how it can reduce the cost of expensive routine immunohistochemistry (IHC) staining, as well as how AI might impact pathologist reading times, diagnostic accuracy, and diagnostic confidence.

Study Design

The study will take place over the course of 9 months, during which time 90 prostate cancer patients will be enrolled. Two genitourinary (GU) sub-specialist pathologists will follow two diagnostic workflows: (1) their standard of care workflow where AI is not utilized, and (2) a diagnostic workflow in which they are assisted by the AI applications in the Paige Prostate Suite. This suite of products includes Paige Prostate Detect, Paige Prostate Grade and Quantify, and Paige Prostate PNI, which work to help pathologists identify, grade and quantify prostate cancers, as well as identify the presence of perineurial invasion (PNI).

It is important to note that the standard of care workflow at UMC Utrecht includes the use of both standard hematoxylin and eosin (H&E) and IHC assessment for every patient. Paige’s AI will run only on the H&E-stained slides. After full pathologist review, each case in both study arms will be marked as either “malignant,” “doubt,” or “benign,” with IHC subsequently being evaluated for “doubt” and “benign” cases to confirm diagnosis. In addition to the final outcome, pathologists will be tracking metrics such as time to diagnosis and their confidence levels.

Study Aims

The study is guided by several key hypotheses designed to show how AI might address some of pathology’s most common pain points. The first pain point is that while cancer cases are rising around the globe, fewer new pathologists are entering the workspace, leaving many patients facing delayed diagnosis and thus a delay to their treatment, as pathologists are the first to influence the course of their care journey. If AI can enhance pathologist’s reading efficiency, speeding up case review time, more patients would reap the benefits of prompt turnaround times.

Of course, producing results faster is inconsequential without ensuring those diagnoses are as accurate as possible. Therefore, the study will examine how AI might impact pathologists’ accuracy and confidence, particularly in labeling cases as “malignant.” In the case of prostate cancer, there are several specific elements of diagnosis that can be particularly challenging, such as tumor grading and quantification. Previous studies, including a nationwide study conducted by the UMC Utrecht team, have shown that these processes can be highly subjective, which may have a direct impact on patient care. If AI could automate grading and quantification, and therefore reduce subjectivity, pathologists could feel more confident in delivering a diagnosis that would allow patients to receive the most effective treatment.

The primary hypothesis being evaluated is whether AI could reduce the need for IHC testing. This is a common testing method at many institutions and is used upfront by UMC Utrecht on every prostate cancer biopsy. Reducing the need for this additional test, as well as enhancing slide read efficiency and improving diagnostic confidence, could reduce the mean cost of case review. If AI could make diagnostic processes more affordable, patients at institutions of all shapes and sizes around the globe would be better able to access the best quality care.

Finally, a common pain point among pathologists is the fear that transitioning from traditional workflows to the use of new technology like AI would be challenging or even negatively impact the overall diagnosis. If adopting AI could instead prove to be a fairly fast and comfortable process, more hospital systems might be inclined to implement this technology, meaning that again, more patients would benefit.

The Paige Prostate Suite has been shown to offer benefits to pathologist’s efficiency and confidence in several previous clinical studies, and we are optimistic that it will prove an effective tool for supporting UMC Utrecht’s diagnostic outputs. Certainly, the study results will reveal important insights into the best path forward for labs looking to implement AI. In this way, we hope that the study will inspire more labs to onboard AI, and make pathologists feel more comfortable undergoing a transition to the use of AI.

“We already know that AI has the potential to aid pathologists in making better diagnoses,” said Rachel Flach MD, study co-Principal Investigator at UMC Utrecht. “The aim of the CONFIDENT-P study is to assess how pathologists, the diagnostic workflow, and the laboratory as a whole can benefit from AI-assisted pathology”

The Power of the UMC Utrecht Team

We are additionally excited about this study because of the wonderful team we’ve partnered with to drive it. UMC Utrecht was among some of the first adopters of digital pathology in the world, and our current collaboration builds on pioneering work from UMC Utrecht which demonstrated large systemic variability in Gleason scoring of prostate biopsies nationwide. Now, in evaluating efficiency, confidence, cost, and other important outcomes AI could offer, we hope to again transform how AI is viewed and used by pathologists.

Importantly, the teams at UMC Utrecht and Paige hope to demonstrate to pathologists that digital pathology and AI are NOT going to replace them, but instead support them through the challenging aspects of diagnosis to help them do their job with greater efficiency, reproducibility, and confidence. In doing so, we believe we can inspire more labs to go digital and implement AI. Achieving more confident diagnoses means better treatment for patients, and therefore better outcomes, which is a mission that all healthcare providers share. We are looking forward to continuing to work with UMC Utrecht to see how our study can support that common goal.

“As the UMC Utrecht aims to be not only a development site, but also an implementation site for AI, we are thrilled to start this study with Paige. Together, we hope to answer some of the questions that are needed to help move forward with implementation of AI-assisted pathology,” added Paul van Diest, Professor and Head, Department of Pathology, UMC Utrecht.