Weakly supervised detection of pheochromocytomas and paragangliomas in CT using noisy data

David Oluigbo, Tejas Sudharshan Mathai, Bikash Santra, Pritam Mukherjee, Jianfei Liu, Abhishek Jha, Mayank Patel, Karel Pacak, Ronald M. Summers

Abstract: Highlights•We present the first approach to detect Pheochromocytomas and Paragangliomas (PPGLs), which are rare adrenal and extra-adrenal tumors.•Data from 255 patients who underwent CT imaging of the chest, abdomen, and pelvis was collected at the NIH.•2D measurements of tumors in the volumes were converted into weak 3D bounding box annotations.•Two different nnUNet models were trained to detect PPGLs via a proxy segmentation task.•Our 3D model achieved a detection precision of 70% and sensitivity of 64.1%, and outperformed a 2D model (p < 0.05).