RISE-MICCAI Journal Session – Dec 6, 2025

Abstract:

Early tumor detection saves lives. Each year, more than 300 million computed tomography (CT) scans are performed worldwide, offering a vast opportunity for cancer screening. However, detecting small or early-stage tumors on these CT scans is challenging, even for experts. Artificial intelligence (AI) segmentation models can assist by highlighting suspicious regions, but training such models typically requires a large number of tumor masks (detailed, voxel-wise outlines of tumors drawn by radiologists). Drawing these masks is costly, and public datasets contain few masks, covering tumors in few organs. In contrast, nearly every CT scan in clinical practice is already accompanied by radiology reports describing the tumor’s size, number, location, and appearance. We introduce R-Super, which trains AI to segment tumors that match tumor descriptions in radiology reports. R-Super uses reports to substitute or supplement masks in segmentation training.

Speaker Bio:

Pedro Bassi is a postdoctoral researcher at Johns Hopkins University in the CCVL Lab (Department of Computer Science), working with Prof. Alan Yuille and Prof. Zongwei Zhou. His research focuses on AI for medical computer vision and cancer detection, including AI training techniques, out-of-distribution generalization, scaling, and trustworthiness. Dr. Bassi received the 2025 MICCAI Best Paper Award (runner-up) and has first-author publications in major venues such as Nature Communications, NeurIPS, ICCV, ISBI, and MICCAI. He completed his PhD in Data Science and Computation at the University of Bologna (Italy), which included an internship at Johns Hopkins University (USA). He previously earned his master’s degree in Computer Engineering from UNICAMP (Brazil), where he also received his bachelor’s degree in Electrical Engineering.