01Statement
Path-iQ includes AI-powered modules that assist pathologists in reviewing digital pathology cases. These modules are deliberately designed as assistive tools. They do not produce diagnostic conclusions, treatment recommendations, or autonomous clinical decisions.
02Assistive role
Our AI surfaces regions of interest, segments structures, quantifies findings (e.g. cell counts, tumour percentages), and prepares draft outputs for pathologist review. Path-iQ AI:
- Does not produce a final diagnosis.
- Does not order treatment, recommend therapy, or contact patients.
- Does not bypass the pathologist sign-out step.
- Is one input among many that the reviewing pathologist considers.
03Indications & intended scope
Each AI model is evaluated for clinical validity within stated indications — typically defined by organ system, stain (e.g. H&E, IHC), and modality. Outside those indications, the AI is suppressed, flagged, or both. The user interface displays the active model's indications.
Examples of typical indications include specific organ-stain combinations for tumour identification, mitosis counting, and IHC scoring. The active set varies by customer configuration.
04Validation & performance
Models are evaluated on representative datasets before release. Performance is documented (sensitivity, specificity, agreement with reference pathologists) and reviewed by clinical and engineering teams. Customers receive a performance summary for each enabled model during onboarding, including:
- Indications and out-of-scope flags.
- Performance metrics with confidence intervals on the validation set.
- Known failure modes and edge cases.
- Validation methodology and reference standard.
05Versioning & auditability
Every AI output is logged with the model version, parameters, and inputs used to generate it. Model updates are versioned, reviewed, and communicated to customers. Updates that change clinical performance are accompanied by a release note and, where applicable, a re-validation summary.
06Override
Pathologists can accept, modify, or reject any AI output. The pathologist's signed-out conclusion — not the AI output — is the diagnostic record. The audit trail captures the AI output, the pathologist's modifications, and the final signed conclusion.
07Bias & fairness
We evaluate models across diverse patient populations and tissue types where data permits. Where data is sparse for a population or scenario, we document the gap and flag the limitation in onboarding materials. We collect feedback from pathologists and labs to identify failure modes and retrain or revalidate models with appropriate clinical and engineering review.
08Training data
Models are trained on licensed datasets and, where authorised, on customer-derived data under explicit written authorisation set out in a Data Processing Agreement. Path-iQ does not use customer case data, patient data, or pathologist-authored content to train general-purpose AI models without that explicit authorisation. Models can be trained or fine-tuned in customer-specific tenants for high-data-residency engagements.
09Contact
For questions about AI models, performance summaries, or to report a suspected AI failure, contact rahul.wasnik@pathiq.cloud.