Clinical Scorecard: AI Won’t Replace the Pathologist – But…

At a Glance

Key Highlights

• Routine AI implementation remains limited in many pathology departments, despite substantial published advances.
• The article identifies pre-analytic and post-analytic workflow support as current areas of greatest value.
• AI-based case prioritization should reflect clinical urgency; lack of an alert should not justify leaving a case at the bottom of the queue.
• A prostate cancer prognostic model combining H&E images with clinical information was validated using randomized-trial samples and outcomes and received FDA authorization for prognostic use in a defined population.
• The interview describes laboratory-specific validation and continued post-deployment monitoring as necessary because local conditions and software can change.

Guideline-Based Recommendations

Diagnosis

• Use AI as support for pathologist interpretation, not as a replacement for pathologist judgment.
• Prefer diagnostic support that presents inspectable evidence, such as relevant references or prior cases, over unsupported AI-generated answers.

Management

• Validate AI tools on local specimens and workflows, including section thicknesses, scanners, staining protocols, patient populations, and difficult cases.
• Assess clinically significant false negatives and evaluate the pathologist and tool together, including whether use improves performance or encourages overreliance.
• Measure effects across the complete workflow, from case receipt to delivery of an accurate, finalized report.

Monitoring & Follow-up

• After deployment, monitor discrepancies and failed analyses and review samples of both flagged and unflagged cases.
• Retain a fixed reference set and rerun it after software updates; reassess when scanners, stains, or patient populations change.

Risks

• Performance on one test dataset does not establish universal reliability; models may fail under local pre-analytic variation or reach correct results for the wrong reasons.
• Prioritization systems must not allow cases without alerts to be neglected.
• Treatment-benefit findings described for AI in prostate cancer came from retrospective analyses of completed trials, not prospective trials in which AI guided treatment.

Patient & Prescribing Data

The article describes selected applications, including a prostate cancer prognostic model authorized for use in a defined patient population and investigational TROP2 testing in lung cancer; it does not provide general prescribing guidance.

Retrospective analyses of completed trials identified patients who may benefit from hormonal therapy, but the article notes that prospective trials in which AI guided treatment have not been conducted for these findings.

Clinical Best Practices

• Keep the pathologist in control and ensure that evidence and audit trails remain accessible.
• Use local validation that includes difficult cases and close review of clinically significant false negatives.
• Continue evaluation after implementation, including after relevant hardware, staining, population, or software changes.
• Demonstrate local operational benefits and establish a viable reimbursement or funding model for analytic tools.

Related Resources & Content