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

Advances in scanning and storage hardware have enabled widespread use of whole slide images (digital pathology), creating opportunities for machine learning in oncological diagnostics. Convolutional neural networks can detect, segment, or classify cancer in WSI, sometimes matching or exceeding pathologist accuracy, potentially improving efficiency, objectivity, and diagnostic consistency for tasks like metastasis screening, Gleason grading, and tumor budding detection. Despite promise, key challenges remain: building trust in perceived “black-box” models; ensuring robustness and generalizability via high-quality diverse training data; conducting large-scale (preferably prospective) peer-reviewed validation demonstrating impact on patient care; defining integration into routine workflows and responsibilities; addressing ethical concerns; and achieving regulatory certification. Early involvement of end-users (pathologists) is critical to tailor implementation strategies, yet literature predominantly reflects developer perspectives. This study explores international end-user opinions and first experiences regarding barriers and facilitators to inform validation, implementation, and communication for broader acceptance.

A narrative review identified 14 reviews on barriers and facilitators for CPath clinical use. Most frequently cited were strengths of CPath (efficiency gains, improved accuracy/standardization). Commonly reported barriers included insufficient quality of evidence supporting CPath outcomes, limited generalizability, and potential lack of trust or acceptance among pathologists. Facilitators included clarifying AI training processes, transparency, and the need for fully digitized pathology workflows to enable integration.

Study design: Three-part exploratory study comprising (1) a narrative literature study to identify potential barriers and facilitators; (2) an international eSurvey to gauge initial opinions; and (3) online semi-structured interviews for in-depth exploration, structured by the implementation framework of Flottorp et al. Study population and recruitment: Targeted end-users (pathologists and pathology residents) in the Netherlands and internationally. The eSurvey was disseminated via the Dutch Pathology Association eNewsletter and to members of the International Tumor Budding Consensus Conference (ITBCC). Additional interview candidates were recruited from eSurvey respondents (who provided contact details) and via LinkedIn to ensure variation in attitudes. Data collection: The eSurvey included five statements regarding validation requirements, perceived promise of AI, need to understand algorithm functioning, willingness to use currently, and expected use within 5 years, plus demographics and professional characteristics. Semi-structured interviews (Zoom/MS Teams) began with a short demo video of two CPath algorithms (mitosis detection; Gleason grading). Questions probed six domains of the Flottorp framework: innovation factors; individual professional factors; professional interactions; incentives and resources; capacity for organizational change; and social, political and legal factors (patient factors excluded). Written informed consent was obtained; interviews lasted 32–44 minutes. Data analysis: eSurvey analyzed with descriptive statistics. Interviews were audio/video recorded, transcribed verbatim, and returned to interviewees for member checking. Qualitative coding was performed in ATLAS.ti by two researchers independently, mapped to the Flottorp framework domains; discrepancies were resolved through discussion and a third researcher’s review. Codes were refined via axial coding to produce a concise overview of barriers and facilitators.

eSurvey (n=70; 38 Netherlands, 32 international):

• 87% (61/70) would currently use CPath algorithms as a support tool in oncology diagnostics if available.
• 83% (58/70) expected to use CPath in daily practice in 5 years.
• 67% (47/70) perceived CPath as the future promise in clinical pathology, with higher agreement among Dutch respondents (82%, 31/38) vs international (50%, 16/32).
• Prospective validation required before clinical use: international 94% (30/32) vs Dutch 53% (20/38).
• Need to fully understand algorithm functioning before use: Dutch 32% (12/38) vs international 78% (25/32). Interviews (15 pathologists + 1 resident; diverse subspecialties, average experience 14 years): Identified 65 barriers and 130 facilitators; 29 barriers and 72 facilitators mentioned in ≥2 interviews. Key barriers:
• Quality of evidence: doubts about reliability due to inter-observer variability in reference standards; uncertainty about clinical impact; difficulty attributing outcomes to CPath; time investment for prospective trials; uncertainty about local validation requirements.
• Feasibility/compatibility: development challenges for rare cancers; added effort to guide algorithms (pre/post tissue analysis); dependence on pre-digitization workflow quality; unclear leading role within workflows.
• Source/conflicts: commercial supplier conflicts of interest; limited medical context understanding by vendors.
• Individual factors: perception of black-box; limited CPath knowledge/experience; small tolerated error margins; concerns about loss of diagnostic skills or overreliance; fear of job loss.
• Resources/finances: high investment needs; budget constraints; insufficient staining quality; lack of full digital workflows.
• Legal/policy: unclear liability for pathologists when using CPath; lack of clarity on applicable regulations. Key facilitators:
• Evidence/validation: internal and external validation; inclusion of clinical outcomes; non-inferiority or retrospective vs prospective studies (no consensus on methodology); laboratory-level validation; studies on time savings; reliable suppliers and ongoing development.
• Workflow integration: easy IT integration; background analysis; triaging cases by urgency; flexibility for leading or supportive roles depending on task; speed and user-friendliness; auto-populating structured reports; connections to LMS/PACS; ability to assign CPath to cases.
• Clinical value: decreased repetitive tasks; improved efficiency and diagnostic accuracy; better standardization; refined prognostic factors; discovery of new prognostic markers; improved treatment decisions; potential workload reduction.
• User adoption: generally positive attitudes; stepwise trust-building through exposure; basic understanding of CPath; outcome control by pathologists; intention to use for quantification tasks (e.g., mitosis/Ki67 counting, Gleason grading, lymph node screening).
• Professional interactions: clinician encouragement; clinician trust in pathologists’ discretion; reporting CPath use in pathology reports.
• Organizational capacity: central guidance by national pathology associations (guidelines); centralized updates aligned with clinical guidelines; monitoring and feedback to suppliers; education in residency and continuing education (timing debated).
• Regulation: FDA approvals as enabler; calls for global regulation; pathologist autonomy aligned with accountability. Areas of disagreement: sufficient validation methodology (retrospective/non-inferiority vs prospective trials), preferred role of CPath in workflow (supportive vs leading), and timing of CPath introduction in training.

The study addressed the research question by revealing a wide spectrum of end-user opinions on barriers and facilitators for adopting CPath in oncological pathology. Most concerns focus on the innovation itself—particularly the quality and type of evidence needed and compatibility with existing digital workflows. Differences between Dutch and international respondents highlight how local context (e.g., maturity of digital workflows) influences attitudes toward validation requirements and explainability. The findings underscore that end-users desire both rigorous validation tied to intended clinical use and practical integration strategies, including performance monitoring and clear liability frameworks. Despite knowledge gaps and black-box concerns, pathologists see potential efficiency and quality gains and support a stepwise exposure and education approach. The results align with broader literature calling for appropriate evidence standards, human-AI interaction research, and post-market monitoring, and they emphasize the need for professional guidance and regulatory clarity to achieve safe, effective clinical implementation.

Clinical adoption of CPath is shaped by diverse end-user views and multifaceted challenges in evidence generation, workflow integration, education, and regulation. To progress, the field should define well-specified use cases and audiences, conduct appropriate validation studies (potentially using effectiveness-implementation hybrid designs), and undertake quantitative studies to prioritize key determinants of adoption. Combining validation and implementation efforts and engaging stakeholders will be essential to achieve broad acceptance amid rapid technological advances.

• Limited direct experience: Most interviewees had little or no hands-on experience with CPath; many views reflect expectations rather than practice.
• Early digitalization stage: Pathology is still transitioning to digital workflows; full digitization is often a prerequisite for CPath implementation.
• Selection bias and representativeness: Recruitment likely favored digitally advanced contexts (e.g., the Netherlands), with underrepresentation of international non-academic settings and variable digital adoption levels.
• Lack of existing clinical guideline recommendations on CPath limited the basis for some judgements.
• Small sample relative to the global pathology community; qualitative design limits generalizability.