Code Snippets

Factors in establishing trust in clinical AI

Consistency

Accuracy

Low validation burden

Human review consumes:

• Time
  Often several minutes per output

• Staffing
  Additional navigators, clinicians, or quality personnel

• Budget
  Salary and operational expenses that scale with volume

Human review consumes:

• ~93 million CT scans in 2023

• —> Up from 68.7 million

• 39.8% of all radiology exams contain an incidental finding

Human review consumes:

• 90
  Pancreatic cancers identified

• 37x
  Growth in new patients

• 4,622
  Downstream exams generated

• 55%
  Pancreatic cancers found in stages I, II, and III

• 77%
  Successful return rate for high-risk patients

Validation Burden decreases when:

• Extracted information is precise and complete
  Findings are identified with sufficient detail to determine follow-up without re-reading the report.

• Clinical relationships are captured accurately
  Measurements, locations, and descriptors are correctly linked to the same finding.

• Temporal context is interpreted correctly
  Current findings are distinguished from prior ones, and change over time is understood.

• Outputs are deterministic and reproducible
  The same input produces the same result, supporting auditability and trust.

• Every data element is traceable to source text
  Clinicians can see exactly where information came from without manual reconstruction.

In scenarios where AI outputs require staff intervention:

Depending on the condition, 30–40%
This is some text inside of a div block.