Carcinoid tumors in the small bowel are rare, making it challenging to collect a sufficiently large dataset of lesions with diverse sizes for training detection models using computed tomography (CT) scans. This scarcity particularly affects the detection performance on relatively small or large tumors. To address this limitation, we propose a novel image synthesis method that can selectively augment underrepresented tumor sizes to enhance detection performance. Our method enables size-controllable tumor generation by integrating a tumor segmentation model and a size-aware loss into the training process. Specifically, one dimension of the input noise vector is designated to control the size of the synthesized tumors. These tumors, generated at desired sizes, are implanted into CT scans to enrich the training data for a tumor detection network. Our method produces tumors with clearer size distinctions, while maintaining comparable visual realism compared to a baseline synthesis method. In a visual Turing test, human observers could not reliably distinguish synthetic tumors from real ones. Lesion-level evaluation using free-response receiver operating characteristic (FROC) curves demonstrated that detection performance improved when synthetic tumors were included during training (P=0.04). This method offers a promising direction for improving the detection of rare tumors such as small bowel carcinoid tumors.

The carcinoid tumors in this patient population were very small and potentially difficult to diagnose. The presented method showed reasonable sensitivity at small numbers of FPs for lesion-level detection. It also achieved a promising AUC for patient-level detection. The method may have clinical application in patients with this rare and difficult to detect disease.