Pivotal Reader Study for Liver Lesion CAD

Challenge

A developer of a computer-aided detection algorithm for liver lesions required a pivotal reader study capable of demonstrating a statistically meaningful improvement in detection sensitivity with algorithmic assistance. The case set had to span the full spectrum of lesion size, conspicuity, and underlying hepatic disease, and every case required a reference standard established independently of the algorithm's output. The sponsor also needed the reading environment, randomization scheme, and statistical analysis plan to satisfy FDA expectations for a De Novo classification request.

Our Solution

Intrinsic Imaging assembled a pathology- and follow-up-confirmed case collection enriched for small and subtle lesions, with truth established by an expert panel blinded to algorithm output. Board-certified abdominal radiologists read each case in randomized, counterbalanced sessions with and without CAD assistance, separated by washout periods sufficient to eliminate recall effects. All interpretations were captured in our Part 11 compliant reading environment, and the statistical analysis plan—covering area under the ROC curve, sensitivity, and specificity endpoints—was finalized before the first reading session began.

Results

• Aided sensitivity improvement demonstrated at p < 0.01
• De Novo classification granted
• Complete audit trail accepted without FDA information requests