Standalone Validation of Fracture AI

Challenge

Before its fracture detection algorithm could be marketed, the developer needed standalone performance evidence on a dataset entirely independent of the algorithm's training data. Regulatory reviewers required the validation cohort to reflect real-world acquisition variability—multiple scanner manufacturers, imaging protocols, and patient demographics—and demanded confidence intervals tight enough to support the intended-use claims. Assembling such a cohort, with rigorously adjudicated truth for every examination, exceeded the developer's internal capacity.

Our Solution

Intrinsic Imaging curated a multi-site validation cohort drawn from geographically distinct institutions, deliberately balancing scanner vendors, protocols, and patient populations to mirror deployment conditions. Fellowship-trained musculoskeletal radiologists established the reference standard for every examination through independent dual review with adjudication, documenting fracture presence, location, and subtype. The algorithm's outputs were then compared against this standard under a pre-specified statistical analysis plan, with subgroup analyses by anatomical region and acquisition parameters.

Results

• Validation cohort of 4,800 examinations delivered
• Sensitivity and specificity claims substantiated
• 510(k) clearance received on first review cycle