Pathology AI Standard of Care: Digital Pathology, Cancer Detection, and Liability

AI Transforms the Pathology Laboratory

#

Pathology, the cornerstone of cancer diagnosis, is undergoing a digital revolution. Whole slide imaging has transformed glass slides into gigapixel digital files, and AI algorithms now assist pathologists in detecting cancers, grading tumors, and identifying features invisible to the human eye. Paige AI’s 2021 FDA authorization marked the first-ever approval for AI in pathology, and the field has expanded rapidly since.

But when AI misses a cancer or misclassifies a tumor grade, who bears responsibility? The pathologist who trusted the algorithm? The laboratory that deployed it? The software developer who trained it on limited data?

This guide examines the standard of care for AI use in pathology, the regulatory framework governing digital pathology, and the emerging liability landscape for AI-assisted diagnosis.

FDA-Cleared Pathology AI Devices

#

The Digital Pathology Landscape

#

Unlike radiology (with 800+ FDA-cleared AI algorithms), pathology AI is earlier in its regulatory journey:

Major FDA-Cleared/Authorized Pathology AI:

Breakthrough Device Designations

#

Several pathology AI tools have received FDA Breakthrough Device designation:

Paige Breast Lymph Node:

• Breast cancer metastasis detection in lymph nodes
• Breakthrough designation granted

Paige PanCancer Detect:

• Multi-cancer detection from various anatomic sites
• First AI tool designated for identifying both common cancers and rare variants

What FDA Authorization Means (and Doesn’t Mean)

#

De Novo Authorization: Paige Prostate received De Novo authorization, a pathway for novel devices that don’t have a predicate but are low-to-moderate risk.

What Authorization Does NOT Guarantee:

• Performance in all patient populations
• Generalizability across all specimen types
• Equivalent performance to validation studies in real-world settings
• Detection of all cancer types or grades

Clinical Applications and Risk Areas

#

Prostate Cancer Detection and Grading

#

The Challenge: Prostate biopsy interpretation is among the most difficult tasks in surgical pathology:

• Subtle morphologic features distinguish benign from malignant
• Gleason grading determines treatment course
• Inter-observer variability is significant even among experts

AI Performance (Paige Prostate):

• 7.3% improvement in pathologist cancer detection (89.5% → 96.8%)
• 70% reduction in false-negative diagnoses
• 24% reduction in false-positive diagnoses
• 97.7% sensitivity, 99.3% specificity in validation

AI Performance (Ibex Prostate Detect):

• 99.6% positive predictive value for heatmap accuracy
• Pathologists without AI missed 13% of cancers that AI identified
• Detects perineural invasion (critical prognostic factor)
• Distinguishes low-grade from high-grade tumors

Liability Concerns:

• Missed cancers leading to delayed treatment
• Gleason grade errors affecting treatment decisions
• Over-reliance on AI in atypical cases
• Performance variations across tissue preparation methods

Breast Cancer Detection

#

Applications:

• Primary tumor detection in biopsies
• Lymph node metastasis identification
• Hormone receptor status assessment
• Grade determination

Paige Breast Suite:

• H&E-stained whole slide image analysis
• Biopsy and excision specimen support
• Macrometastases, micrometastases, and isolated tumor cell detection

Liability Concerns:

• Missed axillary lymph node metastases affecting staging
• False positives leading to unnecessary treatment
• Performance across breast cancer subtypes

Multi-Cancer Detection

#

Paige PanCancer (Breakthrough Designation):

• First AI capable of identifying cancers from multiple anatomic sites
• Detects both common cancers and rare variants
• Potential to catch unexpected findings

Risk Considerations:

• Broader scope means more potential for error
• Performance validation across all cancer types challenging
• “Safety net” function may create over-reliance

The Regulatory Framework

#

FDA Classification

#

Pathology AI devices are typically classified as:

Class II (De Novo or 510(k)):

• Most current pathology AI
• Moderate risk designation
• Post-market surveillance requirements vary

Class III (PMA Required):

• High-risk devices
• Would require prospective clinical trials
• Rare in pathology AI currently

CLIA Requirements

#

The Clinical Laboratory Improvement Amendments govern all clinical laboratory testing:

Quality Control:

• Monitor testing personnel, test system, and laboratory environment
• Applies to analytic phase of digital pathology

Validation Requirements:

• Calibrations and performance specification verification
• Equipment maintenance protocols
• Test result comparisons
• Corrective action procedures
• Backup plan for instrument failure
• Procedure manual documentation

Location Requirements:

• Primary diagnostic interpretations must be made in CLIA-certified locations
• Remote signout permitted only during declared emergencies (e.g., COVID-19)

CAP Validation Guidelines

#

The College of American Pathologists issued comprehensive WSI validation guidelines:

Minimum Requirements:

• At least 60 routine cases per application
• Intraobserver diagnostic concordance comparison
• Digitized vs. glass slides viewed at least 2 weeks apart
• Validation must emulate actual clinical environment
• Pathologists must be trained on the system

Laboratory-Specific Validation:

• Each institution must validate their own WSI system
• Cannot rely solely on manufacturer validation data
• Must account for local workflow and patient population

Laboratory Developed Test (LDT) Considerations

#

If Not FDA-Cleared as End-to-End System: Digital pathology tools may be considered LDTs if:

• Scanning instrument, viewing software, display, and analysis application not cleared together
• Subject to FDA’s Final Rule on LDTs
• May require compliance after phaseout period

The Liability Framework

#

The Pathologist’s Responsibility

#

Primary Interpretation: The pathologist remains responsible for the final diagnosis regardless of AI input.

Standard of Care Elements:

• Use AI as assistance, not replacement for judgment
• Understand AI limitations for specific case types
• Document AI use and clinical reasoning
• Recognize when AI recommendations may be unreliable

The Double Bind: Like radiologists, pathologists face competing pressures:

• If AI is followed and wrong → liability for failing to apply independent judgment
• If AI is overridden and diagnosis missed → AI output becomes evidence of what should have been seen

Laboratory Responsibility

#

Pre-Implementation:

• Validate AI per CAP guidelines before deployment
• Ensure CLIA compliance for digital pathology
• Train all pathologists on AI capabilities and limitations
• Establish quality monitoring protocols

Ongoing Obligations:

• Monitor concordance between AI and pathologist diagnoses
• Track performance across case types
• Report adverse events
• Update for software changes and known issues

Failure Points:

• Deploying AI without local validation
• Inadequate pathologist training
• No quality assurance program
• Ignoring performance degradation signals

Manufacturer Liability

#

Product Liability Theories:

• Design defect (AI trained on biased/limited data)
• Manufacturing defect (software bugs, version issues)
• Failure to warn (inadequate disclosure of limitations)

Challenges for Plaintiffs:

• “Black box” algorithms difficult to analyze
• Training data and methodology often proprietary
• FDA clearance cited as evidence of reasonable care

The “Black Box” Problem

#

Neural network-based pathology AI creates unique challenges:

Explainability Gap:

• Algorithms cannot be fully understood by manufacturers or pathologists
• Cannot explain why specific pixels triggered cancer detection
• Difficult to identify systematic biases

Liability Implications:

• Hard to prove specific algorithm error caused harm
• Difficult to apportion fault between AI and pathologist
• Expert testimony on AI function may be limited

Emerging Malpractice Patterns

#

Current State of Litigation

#

Direct pathology AI malpractice litigation remains limited but growing:

Observed Trends:

• Missed cancer diagnoses by machine-learning software increasingly cited in claims
• Product liability claims against AI developers growing
• Multiple defendants common (pathologist, laboratory, software company)

Analogous Cases from Diagnostic AI

#

While pathology-specific AI cases are emerging, patterns from related diagnostic AI provide guidance:

Common Allegations:

• AI failed to detect visible pathology
• Pathologist over-relied on AI “all clear”
• Laboratory deployed AI without adequate validation
• Software performed below claimed specifications

Potential Case Patterns:

Defense Strategies

#

For Pathologists:

• Documentation of independent review
• Appropriate use per indications
• Recognition of AI limitations in specific case
• Compliance with professional standards

For Laboratories:

• CAP validation documentation
• Training records
• Quality monitoring data
• Adverse event reporting compliance

For Manufacturers:

• FDA authorization as evidence of safety
• Proper labeling and limitations disclosure
• Training program adequacy
• Post-market surveillance compliance

Standard of Care for Pathology AI

#

What Reasonable Use Looks Like

#

Pre-Implementation:

• Conduct CAP-compliant validation (minimum 60 cases per application)
• Verify AI performance in your laboratory’s specimen types
• Understand training data demographics and limitations
• Train all pathologists on capabilities and limitations
• Establish clear use case boundaries

Clinical Use:

• AI recommendations are advisory, not determinative
• Pathologist applies independent clinical judgment to every case
• Document AI use and reasoning for concordance/discordance
• Consider AI limitations for atypical specimens or edge cases

Quality Assurance:

• Track concordance rates between AI and pathologist diagnoses
• Monitor for performance variations across case types
• Report adverse events to FDA MAUDE
• Regularly reassess AI performance
• Update for software changes

What Falls Below Standard

#

Implementation Failures:

• Deploying AI without laboratory-specific validation
• Using AI outside approved indications
• No training for pathology staff
• Absence of quality monitoring
• Operating outside CLIA requirements

Clinical Failures:

• Treating AI output as definitive diagnosis
• Ignoring AI findings without documented reasoning
• Over-relying on AI in atypical or complex cases
• Failing to recognize AI limitations for specific specimens

Systemic Failures:

• No AI oversight committee or governance
• Ignoring FDA safety communications
• Suppressing concerns about AI performance
• Failing to validate after software updates

CAP and Professional Society Guidance

#

College of American Pathologists

#

WSI Validation Guidelines (2013, Updated): CAP’s 12 guideline statements for whole slide imaging validation establish the foundation for digital pathology quality.

Key Principles:

• Validation must emulate actual clinical environment
• Minimum 60 cases per application
• Intraobserver concordance study required
• Pathologists must be trained on system
• Each laboratory must conduct own validation

Accreditation Standards:

• CAP inspects and accredits laboratories under CMS authority
• 21 discipline-specific checklists
• Digital pathology requirements increasingly specific
• AI use must meet quality control standards

Association for Pathology Informatics

#

Developing guidance on:

• AI algorithm validation
• Quality assurance for computational pathology
• Integration of AI into laboratory workflows

Digital Pathology Association

#

Provides:

• Regulatory information resources
• Implementation guidance
• Best practices for WSI deployment

Frequently Asked Questions

#

Related Resources

#

AI Liability Framework

#

• AI Misdiagnosis Case Tracker, Diagnostic failure documentation
• AI Product Liability, Strict liability for AI systems
• Radiology AI Standard of Care, Diagnostic imaging AI

Healthcare AI

#

• Healthcare AI Standard of Care, Overview of medical AI standards
• AI Medical Device Adverse Events, FDA MAUDE analysis
• Cardiology AI Standard of Care, Cardiovascular AI liability

Emerging Litigation

#

• AI Litigation Landscape 2025, Overview of AI lawsuits