Phase II Evaluation of a Novel Tumor Imaging Platform

Challenge

A medical device developer sought Phase II evidence that its novel imaging platform could help physicians distinguish malignant from non-malignant tissue across several tumor types. Unlike a therapeutic trial, the endpoint was diagnostic performance, which meant every case needed a reliable reference standard, and the novel images had to be evaluated by readers under conditions that neither overstated nor understated the device's contribution.

The sponsor also needed a protocol that captured the platform's images alongside conventional imaging and pathology without introducing bias through case selection or reading order, and a core lab experienced enough with diagnostic device development to advise on endpoint definitions that regulators would accept.

Intrinsic Imaging’s Solution

Intrinsic Imaging worked with the sponsor on protocol and charter development, defining the reference standard hierarchy, the case-mix requirements for malignant and non-malignant tissue types, and the blinded reading design. Site qualification and training covered acquisition of both the novel platform data and the conventional comparator imaging, and image transfer was configured to handle the platform's proprietary output format.

Readers evaluated cases in randomized order, blinded to the reference standard, using a structured case report form that captured a lesion-level malignancy assessment and confidence score. The charter specified how multiple lesions per patient would be handled and how the reference standard would be adjudicated when pathology and clinical follow-up disagreed.

Quality control and processing were centralized so that each case's platform data, comparator imaging and reference-standard documentation were linked and complete before reading, and the dataset was delivered in a form suitable for ROC-based analysis.

Intrinsic Imaging’s Results

• Approximately 420 lesions from 310 patients at 12 sites evaluated against a documented reference standard
• Blinded, randomized reading design with lesion-level confidence scoring delivered a dataset ready for sensitivity, specificity and ROC analysis
• Proprietary platform output integrated with conventional imaging and pathology records at case level with 100% linkage before reading
• Imaging charter and endpoint definitions structured to support the sponsor's subsequent pivotal study design