The future of artificial intelligence in digital pathology - results of a survey across stakeholder groups

Céline N. Heinz, Amelie Echle, Sebastian Foersch, Andrey Bychkov, Jakob Nikolas KatherbioRxiv 2021 · DOI

TL;DR

A survey of 75 computational pathology experts reveals that predicting treatment response from routine pathology slides is seen as the most promising future application of AI. This, along with predicting genetic alterations and survival, represents the "advanced" AI tasks that stakeholders believe will have the biggest impact.

What Problem Does This Paper Address?

Digital pathology, the digitization of glass slides, is rapidly transforming how pathology is done. This shift opens the door to powerful AI tools that can extract more information from these images than ever before. However, with so many potential applications – from basic tasks like cell detection to complex ones like predicting treatment response – it's unclear which areas are most ripe for clinical integration and where research efforts should be focused. This paper aims to clarify this by surveying experts to gauge their opinions on the most promising AI applications in digital pathology, particularly for solid tumors.

Methods

The researchers conducted an anonymous online survey of 75 computational pathology experts from academia and industry. The survey had two main parts:

  1. Demographics and Background: Participants provided information on their age, professional background, and experience with AI and computer programming.
  2. Application Prioritization: Participants rated various AI applications in digital pathology on a scale of 1 to 10 (1 being least interesting/promising, 10 being most). These applications were categorized based on previous work into "enabling technologies," "basic tasks," and "advanced tasks," with some new categories added.

The data was analyzed using descriptive statistics (median, IQR, mean, SD) and Kruskal-Wallis tests to compare group responses. Spearman correlation was used to assess relationships between category ratings.

Datasets

Not discussed in the paper. The study relies on survey responses, not on training or evaluating AI models on specific datasets.

Key Insights and Ideas

  • "Advanced" AI is King: The survey strongly suggests that applications moving beyond current pathologist capabilities – specifically predicting treatment response, survival, and genetic alterations directly from images – are considered the most valuable and promising.
  • Clinical Translation is Key: While basic tasks like cell detection are foundational, the real excitement lies in AI that can directly inform clinical decisions and patient outcomes.
  • Expert Consensus Matters: The study highlights the value of gathering expert opinions to guide research and development in a rapidly evolving field.
  • Demographics Don't Drastically Shift Priorities: While some minor differences were observed (e.g., medical professionals ranking survival prediction slightly higher), the overall consensus on the most promising applications remained remarkably consistent across different age groups and professional backgrounds.

Key Results

  • Top Application: Predicting treatment response directly from H&E images of solid tumors was ranked highest by all participants, with a mean score of 9.09 out of 10.
  • High-Ranking Advanced Tasks: Predicting genetic mutations (mean 8.64) and gene expression (mean 8.63) directly from H&E images also scored very high.
  • Least Promising: Detecting large structures in H&E images was ranked as the least promising application (mean 6.83).
  • Free-Text Insights: Participants emphasized the need for AI to automate time-consuming tasks, improve quality control, and integrate seamlessly into clinical workflows. They also stressed the importance of data availability, sharing, and interpretability of AI models.

Limitations and Future Work

The authors acknowledge several limitations:

  • Non-Response Bias: Participants were recruited through professional networks and conferences, which might have excluded individuals with different views or less engagement with digital pathology.
  • English-Speaking Focus: The survey was limited to English speakers, potentially missing perspectives from other regions.
  • Snapshot in Time: The field of AI in pathology is evolving rapidly, so these findings represent a snapshot of opinions in 2021 and may change over time.

The authors suggest that their survey provides a valuable starting point for future, larger-scale studies to validate these findings and guide the development of AI in digital pathology. They anticipate a continued focus on developing new computational pathology solutions that can be integrated into existing digital pathology environments, with a particular emphasis on advanced tasks that extend pathologist capabilities.

Summary updated

Abstract

Abstract Artificial intelligence (AI) provides a powerful tool to extract information from digitized histopathology whole slide images. In the last five years, academic and commercial actors have developed new technical solutions for a diverse set of tasks, including tissue segmentation, cell detection, mutation prediction, prognostication and prediction of treatment response. In the light of limited overall resources, it is presently unclear for researchers, practitioners and policymakers which of these topics are stable enough for clinical use in the near future and which topics are still experimental, but worth investing time and effort into. To identify potentially promising applications of AI in pathology, we performed an anonymous online survey of 75 computational pathology domain experts from academia and industry. Participants enrolled in 2021 were queried about their subjective opinion on promising and appealing sub-fields of computational pathology with a focus on solid tumors. The results of this survey indicate that the prediction of treatment response directly from routine pathology slides is regarded as the most promising future application. This item was ranked highest in the overall analysis and in sub-groups by age and professional background. Furthermore, prediction of genetic alterations, gene expression and survival directly from routine pathology images scored consistently high across subgroups. Together, these data demonstrate a possible direction for the development of computational pathology systems in clinical, academic and industrial research in the near future.
BibTeX
@misc{heinz2021future,
  author = {Céline N. Heinz and Amelie Echle and Sebastian Foersch and Andrey Bychkov and Jakob Nikolas Kather},
  title = {{The future of artificial intelligence in digital pathology - results of a survey across stakeholder groups}},
  year = {2021},
  doi = {10.1101/2021.12.16.472990},
  howpublished = {bioRxiv},
}
Figure