3D oncology lesion segmentation across follow-up studies
Tracking and 3D segmenting the same tumor lesions across multiple time points using a prompt-based AI-assisted annotation tool.
An oncology AI project required identifying the same lesions across successive follow-up studies for each patient, starting from a pre-segmented reference exam. The dataset combined several hundred studies with multiple series each and required strict lesion identity preservation over time.
Segment every referenced lesion in 3D across all patient studies, preserve labels between time points and record uncertainty or missing lesions in a structured way suitable for downstream longitudinal analysis.
A team of radiologists was trained on the client's AI-assisted annotation platform, using bounding-box and point prompts followed by interactive refinement. A per-series status workflow (annotated / doubt / rejected) was applied consistently, and lesion identifiers from the reference exam were carried forward across studies.
The client received longitudinal, patient-level 3D segmentations aligned with their reference exams, with transparent tracking of uncertainty. Time per case dropped as the team gained fluency with the prompt-based tools.