Pathologists’ first opinions on barriers and facilitators of computational pathology adoption in oncological pathology: an international study

The past decade has witnessed significant advancements in digital pathology, driven by improved whole slide image (WSI) scanning and storage technologies. This progress has facilitated the application of machine learning, particularly convolutional neural networks (CNNs), to extract diagnostic information from WSIs, a field known as computational pathology (CPath). Successfully developed CNN algorithms can automatically detect, segment, and classify cancer in WSIs, often achieving accuracy comparable to or exceeding that of human pathologists, especially in oncological pathology. This increased efficiency could potentially reduce pathologists' workload by automating repetitive tasks like screening for metastases. CPath also offers the potential to enhance the accuracy, speed, and objectivity of diagnoses for biomarkers with significant inter-observer variability, such as Gleason grading of prostate cancer and the detection of tumor buds in colorectal cancer. Furthermore, CPath might uncover new diagnostic clues previously unrecognized by pathologists. Despite the promising potential of CPath, several significant challenges hinder its widespread adoption in clinical practice. These challenges include building trust in CPath (often viewed as "black boxes"), developing robust and generalizable algorithms trained on high-quality data, conducting large-scale validation studies demonstrating improved patient care, integrating CPath into daily workflows, addressing ethical concerns, and establishing legal certification. While many studies address these challenges from a developer's perspective, there's a scarcity of research exploring the adoption perspectives of pathologists, the end-users of these technologies. Given the global nature of these challenges, an international perspective is crucial for informing effective implementation strategies. This study aims to explore international perspectives on the future role of CPath in clinical practice by examining the opinions and first-hand experiences of pathologists regarding the barriers and facilitators to adoption. This understanding will be vital in guiding the development of validation studies, implementation strategies, and communication initiatives to achieve broader stakeholder acceptance.

A narrative literature review identified 14 review studies describing barriers and facilitators to CPath clinical use. These studies predominantly highlighted the strengths of CPath. Recurring themes included barriers related to the quality of evidence supporting CPath outcomes and potential lack of trust or acceptance among pathologists. Facilitators commonly mentioned included clarification on AI training and the need for fully digitized pathology workflows.

This study employed a mixed-methods approach, combining a literature review, an e-survey, and semi-structured interviews. The literature review focused on identifying barriers and facilitators to CPath adoption, informing the design of the subsequent e-survey and interview guide. The e-survey, disseminated through the Dutch Pathology Association eNewsletter and the International Tumor Budding Consortium (ITBCC) email list, included five statements assessing pathologists' attitudes towards CPath and their expectations for future use. The survey collected demographic data, including age, gender, professional role (pathologist or resident), type of laboratory (academic or non-academic), and involvement in AI development. Participants who provided their email addresses were invited to participate in semi-structured interviews. The interview guide, developed based on the literature review, used the implementation framework of Flottorp et al. to categorize influencing factors. Six of the seven domains of the framework were explored: Innovation factors, Individual professional factors, Professional interactions, Incentives and resources, Capacity for organizational change, and Social, political, and legal factors. The "Patient factors" domain was omitted as pathologists do not directly interact with patients. Interviews were conducted online via Zoom or Microsoft Teams and were audio- or video-recorded with informed consent. A total of 16 pathologists and pathology residents participated in the interviews, representing a range of specializations and experience levels. The interviews were transcribed verbatim, and qualitative data analysis was conducted using ATLAS.ti software. Thematic analysis was performed to identify recurring themes and patterns in the responses, mapping these to the Flottorp et al. framework domains. The e-survey data was analyzed using descriptive statistics.

The e-survey (70 respondents) showed a generally positive attitude towards CPath: 87% of respondents would use CPath if available, and 83% anticipated using it within five years. However, significant differences emerged between Dutch and international respondents, with Dutch pathologists showing a more positive attitude and less emphasis on the need for prospective validation studies. The interviews (16 participants) yielded 65 barriers and 130 facilitators to CPath implementation. Key findings include: **Barriers:** * **Quality of Evidence:** Significant concerns were raised about the reliability of CPath algorithms, particularly the use of pathologist expertise (subject to inter-observer variability) as a gold standard in algorithm development. Questions were raised about the real-world impact and the need for local and prospective validation. * **Feasibility:** Developing CPath algorithms for rare cancers was seen as challenging due to the large datasets required for training. Additional effort required for manual selection of regions and correction of CPath output were also noted. * **Compatibility:** The quality of CPath algorithms was seen as being dependent on the quality of prior steps in the workflow before slide digitization, and there were concerns about conflicts of interest with commercial CPath suppliers. * **Individual Professional Factors:** A critical attitude was observed, with pathologists emphasizing the need for sufficient evidence of clinical benefit before widespread adoption. A lack of knowledge and experience with CPath was widespread, and several expressed concern about the "black box" nature of the algorithms. There was also a fear of losing jobs or diagnostic skills due to reliance on CPath. * **Professional Interactions:** Clinicians' encouragement and trust in pathologists' judgement regarding CPath use were seen as facilitators. Clear communication between pathologists and clinicians was also crucial. * **Incentives and Resources:** Insufficient staining quality and limited resources were identified as potential barriers. Having a single supplier for the digital workflow, including CPath, was highlighted as a facilitator. * **Capacity for Organizational Change:** Centralized guidance and updates of CPath algorithms, along with prospective monitoring and feedback mechanisms, were viewed as important facilitators. * **Social, Political, and Legal Factors:** Uncertainty regarding liability in cases of CPath errors and a lack of awareness about applicable legislation were major concerns. Clear regulations and FDA/CE approvals were seen as facilitators. **Facilitators:** * The main facilitators centered around the potential strengths of CPath: workload reduction, improved treatment decisions, discovery of new prognostic factors, and the development of more comprehensive algorithms. Proper development and reliable validation were also crucial. * A step-by-step approach to building trust and familiarity was suggested, including education and real-world simulations.

This study offers a comprehensive overview of pathologists' perspectives on CPath adoption, highlighting diverse opinions and the complexity of implementation. The findings show that many factors, particularly related to evidence quality, compatibility with workflows, and individual professional factors, influence adoption. The disparity between Dutch and international respondents in the e-survey suggests that national contexts and levels of digitalization in pathology play a significant role. This mirrors findings from other studies indicating a lack of consensus on AI adoption even among experienced pathologists. The need for robust validation studies is emphasized, particularly those focused on demonstrating clinical benefits. Future research must address the need for better educational support and the integration of CPath into digital workflows, considering the various intended uses and the human-AI interaction dynamics. Performance monitoring and feedback mechanisms are crucial for ensuring the safe and reliable use of CPath, requiring improved data infrastructure and collaboration among pathology associations.

This study reveals the diverse opinions surrounding CPath adoption, highlighting the complexity of implementation. Further quantitative studies are needed to prioritize key factors for adoption, focusing on specific CPath algorithms and their target user groups. A combined approach of validation and implementation studies using a hybrid design is recommended to address the challenges identified and ensure widespread stakeholder acceptance within the rapidly evolving field of computational pathology.

This study has some limitations. The recruitment strategy may have introduced selection bias, as it only included a small percentage of the international pathology community, and there was a disproportionate representation of Dutch pathologists who had more experience with digital workflows. Many interviewees lacked direct experience with CPath, which led their perspectives to be mostly based on expectations. The study also lacks the perspectives of patients in assessing the clinical outcomes of CPath use. Finally, the lack of clear recommendations on the use of CPath algorithms in clinical guidelines is a limitation.