Ethical issues in computational pathology

Authors: Tom Sorell, Nasir Rajpoot, Clare Verrill
Published in: J Med Ethics, 2022
Study Focus: Ethical issues surrounding whole slide image-based computational pathology.

Recent advancements in computational pathology highlight ethical concerns regarding:
- Data collection and processing conflicts with ethical norms.
- Data fusion practices against data ethics and biobanking standards.
- AI's 'black box' problem — obscured interpretability and accountability of AI algorithms.
- Dependency on profit-driven scanning technology manufacturers.

Digital Pathology Defined: Utilizing digital whole slide images (WSIs) instead of traditional physical slides.
Benefits:
- Cost-effectiveness, improved clinical workflow, and enhanced collaboration among experts.
- Facilitates teaching and quality assurance processes without physical slide transportation.
Collaborative Nature: Sharing sensitive data raises new ethical concerns within legal frameworks.

AI's Data Requirements: Significant data collection needs clash with data minimization principles in ethics and law.
Data Fusion: Challenges arise when combining varied data types which may violate ethical norms.
AI Opacity: 'Black box' nature of AI methods complicates accountability and understanding results even for experts.
Market Incentives: Scanning technology manufacturers may prioritize profit, influencing ethical integrity in data usage and consent procedures.

Classification Abilities: Quickly discerning malignancy and predicting patient outcomes through data insights.
Data Integration: Analyzing heterogeneous data, such as biobanked samples and clinical notes, broadens diagnostic capabilities.
Quantitative Histo-morphometry (QH): Advanced analysis techniques for spatial inquiry of tumors addressing features like nuclear orientation and texture.
Distinctions between domain-inspired and domain-agnostic approaches:
- Domain Inspired: Specific to certain diseases, e.g., cancer types.
- Domain Agnostic: Useful across various disease types, assessing general characteristics like tissue architecture.

Moral Benefits: Potential for quicker diagnoses linked to timely treatments, improving patient survival rates.
Data Protection Concerns: WSIs, while not always directly identifying, can still provide inferential links to patient identities, invoking GDPR and ethical scrutiny.
Hunger for Data: Challenges in meeting AI's data needs while respecting laws that advocate personal data minimization.
Deidentification Issues: Practical anonymity is often not absolute; inferential data remains a concern in large datasets.

Repurposing Samples: Ethical implications arise from using existing pathology samples for new research avenues, especially without clear consent regarding secondary data uses.
Consent Complexity: Public consultations reveal support for broad consent approaches but raises concerns over understanding the implications of such consents for future use.
Tissue Retention Laws: The Human Tissue Act highlights the distinction between ethical uses of living versus deceased patients' samples, complicating secondary uses for research.

AI Ethics vs. Data Processing Law: The importance of transparency is emphasized, particularly around algorithms making significant health decisions.
Reliability of AI Diagnoses: Although AI can automate diagnostics, the issue of explainability remains a significant hurdle in the medical context to ensure informed consent.
Perception of AI in Healthcare: Despite complexities, public sentiment reflects support for effective algorithm-based healthcare solutions.

Commercial Involvement: Scanner manufacturers have vested interests in the proliferation of computational pathology, leading to conflicts between profit motives and ethical standards.
Transparency in Data Usage: NHS guidelines delineate acceptable commercial activities using patient data only for health benefits, promoting ethical conduct.
Public Perception: Broad support for advancements in diagnostic technology coexists with skepticism towards non-health-related data uses.

Ethical issues in computational pathology are multifaceted, balancing data ethics and commercial interests with patient welfare.
Potential benefits in diagnostics and treatment must be weighed against the ethical dilemmas posed by automated systems and data usage practices.