by Christos Evangelou, MSc, PhD – Medical Writer and Editor
Cancer diagnosis based on pathological analysis of tissue biopsies can be a time-consuming process and requires highly trained pathologists. However, the recent development of artificial intelligence (AI) tools that are trained on manually annotated tissue images has opened new avenues for faster diagnosis through automated classification of tissue sections.
In a recent study, researchers at the Institute of Molecular Pathology and Immunology of the University of Porto (Ipatimup) evaluated the diagnostic performance of pathologists with and without the assistance of an AI tool, PaigeProstate, in detecting and grading prostate cancer, with the aim of demonstrating the potential benefits of AI-assisted diagnosis.1
The study showed that the use of the AI tool significantly reduced the requirement for immunohistochemistry (IHC) studies and second opinions, as well as decreased the median time for reading and reporting each slide.
The findings of the study suggest that AI tools could help improve the efficiency of cancer diagnosis by assisting pathologists in detecting and grading prostate cancer. By reducing the workload on pathologists, AI tools like PaigeProstate could also help to address the shortage of trained pathologists in some areas.
“We found that the use of AI tools could result in a reduction in time for reporting and in the consumption of resources, including IHC and second opinion requests. AI tools could also help us preserve tissue for further analysis, like molecular studies,” said António Polónia, MD, PhD, surgical pathologist at Ipatimup Diagnostics and co-chair of the Digital and Computational Pathology Working Group of the European Society of Pathology.
The report was published in the journal Virchows Archiv.
Evaluating the Diagnostic Performance of Pathologists With and Without the Assistance of PaigeProstate, A New AI Tool for Prostate Cancer Diagnosis
In this study, the researchers evaluated the usefulness of PaigeProstate, a clinical-grade AI tool designed to assist pathologists in detecting, grading, and quantifying prostate cancer. PaigeProstate is a deep learning tool trained using a weakly supervised approach of convolutional neuronal networks (CNNs) that runs in the proprietary, Food and Drug Administration (FDA)-approved viewer called FullFocus. The tool is specifically trained to detect acinar adenocarcinoma in prostate biopsies.
Commenting on the novelty of their study, Dr. Polónia said: “Previous studies compared the evaluation of pathologists against AI algorithms alone. We are probably the first, or one of the firsts, that compared the performance of pathologists without AI against that of pathologists with AI, which is a more realistic given that AI should not be used alone.”
The Use of AI Can Save Time and Money by Reducing the Need for Immunohistochemistry Studies and Second Opinions
The study evaluated the diagnostic performance of four pathologists in two phases. In phase 1, the pathologists examined prostatic core needle biopsies unaided, and in phase 2, they were assisted by PaigeProstate. The diagnostic accuracy for prostate cancer was 95.00% in phase 1 and 93.81% in phase 2. The intraobserver concordance rate between the two phases was 98.81%.
The reported rate of atypical small acinar proliferation was approximately 30% lower in phase 2 than in phase 1. Furthermore, pathologists requested significantly fewer IHC studies (approximately 20% less) and second opinions (approximately 40% less). These findings suggest that the use of AI tools such as PaigeProstate can save time and money by reducing uncertainty and the requirement for follow-up analyses and second opinions.
The Use of AI Can Accelerate Prostate Cancer Diagnosis by Reducing the Time Required for Slide Reading and Reporting
The researchers also compared the time required for slide reading and reporting by pathologists with and without the help of the AI tool. They found that the use of PaigeProstate significantly reduced the time needed for pathologists to read and report on prostate biopsies. The median time required for reading and reporting each slide was 139.00 seconds in phase 1 (without AI assistance) and 108.50 seconds in phase 2 (with AI assistance). This 21.94% decrease in the time required per slide was statistically significant (P < 0.001).
For cancer cases, pathologists required 18.74% longer during phase 1 than phase 2 (P < 0.00), and for negative cases, the time required for reporting was 18.41% longer during phase 1 than phase 2 (P < 0.001). These findings suggest that AI-assisted diagnosis can significantly improve the efficiency of cancer diagnosis by reducing the time required for pathologists to read and report on biopsies.
“In this study, we show that the synergic use of PaigeProstate contributes to increased efficiency while maintaining highly accurate diagnostic standards,” said Dr. Polónia when asked about the importance of these findings. “In the setting of cancer diagnosis, time is an important element, and turnaround time has an important impact on health care costs and on-time treatments that can be accelerated with the use of AI,” he added.
Although the study showed that the use of AI in prostate cancer diagnosis could increase the efficiency of pathologists, no significant difference in diagnostic accuracy was observed with and without the use of AI.
Dr. Polónia explained that the non-significant effect of using AI on diagnostic accuracy is because the pathologists were highly experienced, and the accuracy was very high both with and without AI. “The use of AI tools will probably have different impact on diagnostic accuracy when used by different groups of pathologists, such as younger and less experienced pathologists,” he argued. However, this will have to be confirmed in subsequent studies.
“The system is not — and will probably never be — perfect and did not show perfect agreement preferentially on positive cases. This has to be improved in the future. But the system is already mature enough to be used in the clinical setting to help pathologists detect very small lesions,” he added.
Further research is needed to evaluate the generalizability of these findings and explore how AI-assisted diagnosis can improve other aspects of cancer diagnosis beyond detection and grading, such as treatment planning or patient outcomes.
- Eloy C, Marques A, Pinto J, et al. Artificial intelligence–assisted cancer diagnosis improves the efficiency of pathologists in prostatic biopsies. Virchows Arch. 2023;(0123456789). doi:10.1007/s00428-023-03518-5