Oncology AI Standard of Care: Cancer Diagnosis, Imaging Analysis, and Liability

AI Transforms Cancer Care

#

Artificial intelligence is reshaping every phase of cancer care, from early detection through treatment planning and survivorship monitoring. AI tools now analyze mammograms for breast cancer, pathology slides for prostate cancer, and imaging studies across multiple cancer types. But as AI becomes embedded in oncology workflows, critical liability questions emerge: When AI-assisted diagnosis misses cancer or delays treatment, who bears responsibility? When AI recommends treatment and outcomes are poor, what standard of care applies?

This guide examines the evolving standard of care for AI in oncology, the expanding landscape of FDA-cleared devices, and the liability framework for AI-assisted cancer care.

FDA-Cleared Oncology AI Devices

#

Pathology AI: Prostate Cancer Detection

#

Paige Prostate Detect made history as the first FDA-authorized AI in pathology:

Paige Prostate Detect:

• First FDA de novo authorization for AI in digital pathology (September 2021)
• Assists pathologists in detecting prostate cancer in biopsy slides
• 97.7% sensitivity, 99.3% specificity in independent testing
• Pathologists improved from 89.5% to 96.8% sensitivity with AI
• 70% reduction in false negative diagnoses
• 24% reduction in false positive diagnoses
• Validated on slides from 200+ institutions

CONFIDENT P Trial Results (2025):

• AI-assisted pathologists reported higher diagnostic confidence (80% vs 56%)
• 30% reduction in atypical small acinar proliferation (ASAP) reports
• 20% reduction in immunohistochemistry requests
• 40% reduction in second opinion requests
• 20% faster median reading and reporting time

Mammography AI: Breast Cancer Screening

#

Multiple FDA-cleared AI tools assist radiologists in breast cancer detection:

FDA-Cleared Mammography AI:

Performance Considerations:

• AI detection complements but doesn’t replace radiologist interpretation
• Double reading with AI shows improved sensitivity
• Recall rates and specificity vary by implementation

Dermatology AI: Skin Cancer Detection

#

DermaSensor:

• FDA-approved handheld device
• AI-powered spectroscopy technology
• Detects melanoma, basal cell carcinoma, squamous cell carcinoma
• Point-and-click design for primary care settings
• Noninvasive lesion analysis

Multi-Cancer Detection Platforms

#

Emerging AI Tools:

The Oncology AI Liability Framework

#

Missed or Delayed Cancer Diagnosis

#

Cancer misdiagnosis claims are among the highest-value medical malpractice cases:

Missed Diagnosis Scenarios:

Multi-Party Liability Analysis

#

Potential Defendants in Oncology AI Cases:

The “Augmentation vs. Replacement” Question

#

Critical Legal Distinction:

AI in oncology is designed to augment, not replace, physician expertise. When AI fails, the key question is whether the physician exercised appropriate independent clinical judgment.

Factors Courts May Consider:

• Did the physician independently review the imaging/slides?
• Did the physician consider clinical context AI cannot access?
• Was the AI output a “recommendation” or “diagnosis”?
• Did institutional protocols require human oversight?

Standard of Care Considerations

#

Professional Society Guidance

#

Medical societies are developing AI guidance for oncology practice:

Key Principles (Emerging Consensus):

1. AI augments but doesn’t replace clinical judgment
2. Physicians remain responsible for final diagnostic decisions
3. AI limitations must be understood by users
4. Quality assurance programs should monitor AI performance
5. Patient consent/disclosure for AI use is advisable

American Society of Clinical Oncology (ASCO):

• Supports AI integration with appropriate oversight
• Emphasizes human expertise in contextualizing AI recommendations
• Calls for transparency about AI use in patient care

American College of Radiology (ACR):

• Provides guidance on mammography AI integration
• Emphasizes double reading protocols
• Addresses liability and quality assurance

When AI Gets It Wrong

#

Case Study Patterns from AI Diagnostic Failures:

JAMA Health Forum documented across 691 AI devices:

• 489 adverse events reported
• 113 recalls issued
• 1 death attributed to AI device failure
• 43% of recalls occurred within first year of clearance
• Diagnostic errors were the most common adverse event category

Oncology-Specific Risks:

• False negatives leading to delayed cancer diagnosis
• False positives causing unnecessary biopsies/treatment
• AI trained on one population performing poorly on another
• Edge cases and rare cancer presentations outside training data

Physician Responsibilities

#

Before Using AI in Cancer Care:

• Understand AI capabilities and limitations
• Verify FDA clearance matches intended use
• Establish protocols for discordant AI/clinical findings
• Train staff on proper AI integration

During AI-Assisted Diagnosis:

• Exercise independent clinical judgment
• Consider patient-specific factors AI cannot access
• Document AI use and your clinical reasoning
• Address discordant findings explicitly

After AI Results:

• Ensure appropriate follow-up regardless of AI output
• Monitor for symptoms that may contradict AI assessment
• Track AI performance in your practice
• Report AI failures to manufacturers and FDA (MAUDE)

Emerging Liability Concerns

#

Bias and Disparities in Cancer AI

#

AI oncology tools may perpetuate or worsen health disparities:

Documented Concerns:

• Training data underrepresents racial/ethnic minorities
• AI may perform differently across demographic groups
• Socioeconomic factors affect access to AI-enhanced care
• Skin cancer AI may miss lesions on darker skin tones

Liability Implications:

• Disparate impact claims if AI performs worse for certain groups
• Failure to validate across populations
• Duty to disclose known performance disparities

Inadequate Clinical Trial Data

#

FDA clearance data raises concerns about AI validation:

Per JAMA Health Forum Analysis:

• 46.7% of devices failed to report study designs
• 53.3% failed to report training sample size
• 95.5% failed to report demographic information
• Only 1.6% reported data from randomized clinical trials
• Only 7.7% from prospective studies

Litigation Implications:

• Failure-to-warn claims for inadequate performance disclosure
• Design defect claims for insufficiently validated AI
• Negligent implementation claims against healthcare systems

Treatment Planning AI

#

Beyond diagnosis, AI now informs treatment decisions:

Emerging AI Applications:

• Treatment response prediction
• Radiation therapy planning
• Chemotherapy regimen optimization
• Recurrence risk scoring
• Clinical trial matching

Liability Considerations:

• AI treatment recommendations are advisory only
• Physician must exercise independent judgment
• Poor outcomes may trigger malpractice claims
• Unclear liability when AI suggests unconventional approaches

Case Examples and Analogies

#

Radiology AI Precedents

#

Mammography AI cases inform oncology AI liability:

Common Claim Patterns:

• AI-flagged abnormality dismissed by radiologist → Cancer progresses
• AI missed cancer that experienced radiologist would have caught → Delayed diagnosis
• AI recommended immediate follow-up, system failed to track → Lost to follow-up

Applicable Principles:

• AI doesn’t eliminate radiologist duty to exercise skill
• AI recommendations require physician evaluation
• Systems must track AI-generated findings to completion

Pathology Malpractice Context

#

Traditional Pathology Malpractice:

• Prostate cancer verdicts range from $3M to $120M+
• Breast cancer missed diagnosis among highest-value claims
• Tumor grade errors affect treatment decisions

AI-Specific Questions:

• Does AI use meet or exceed current standard of care?
• Can pathologist liability be reduced by demonstrating AI use?
• Does AI failure create claims against developers?

Frequently Asked Questions

#

Practical Guidance

#

For Healthcare Systems

#

Implementation Considerations:

• Conduct thorough vendor due diligence
• Verify FDA clearance matches intended use
• Establish clear protocols for AI integration
• Create quality assurance monitoring programs
• Train all users on AI capabilities and limitations

Risk Management:

• Review malpractice coverage for AI-related claims
• Consider contractual indemnification from AI vendors
• Establish incident reporting procedures
• Document AI validation in your patient population

For Oncologists and Pathologists

#

Clinical Integration:

• AI results are starting points, not final diagnoses
• Apply clinical judgment to all AI outputs
• Document independent clinical reasoning
• Consider AI limitations for each patient
• Address discordant findings explicitly

Documentation Best Practices:

• Record that AI was used
• Document the specific AI output
• Note clinical factors affecting interpretation
• Explain reasoning if overriding AI recommendation

For Patients

#

Questions to Ask:

• Is AI being used in my cancer diagnosis or treatment planning?
• What are the AI’s known limitations?
• Who reviews AI results before my diagnosis is finalized?
• How accurate is this AI tool?
• What happens if AI results conflict with clinical findings?

Related Resources

#

• Pathology AI, Digital pathology, Paige Prostate, slide analysis
• Radiology AI, Mammography AI, imaging analysis, detection tools
• Medical AI Device Cases, FDA adverse events, recalls, malpractice trends
• AI Misdiagnosis Case Tracker, Diagnostic AI failures across specialties
• AI Product Liability, AI LEAD Act, strict liability framework