.png){kind=icon}
Return
Report Supervision
.png){kind=icon}
PRE
AI
DOI：10.1016/j.media.2026.104257.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
•
Radiology reports directly supervise tumor segmentation training.
•
Volume Loss and Ball Loss teach segmentation models to segment tumors matching reports in tumor count, sizes, and locations.
•
Reports scale segmentation datasets when masks are few or many.
•
R-Super gains up to +15% F1-Score and DSC over mask-only training.
•
R-Super improves tumor detection at unseen external hospitals and for small tumors.
Abstract
Segmentation models can surpass radiologists, classification models, and vision–language models in tumor detection. Furthermore, segmentation models outline tumors, allowing radiologists to better verify and trust the AI output. However, their main limitation is the scarcity of tumor masks: creating one 3D tumor mask takes up to 30 min, so most public CT datasets contain only a few hundred masks, and even the largest private datasets contain only a couple of thousand. Tumor masks are not produced in clinical routine, but radiology reports are. Public datasets contain tens of thousands of CT-Report pairs, and hospitals contain hundreds of thousands. These reports describe tumors in detail, providing large-scale, informative training data. Here, we introduce Report Supervision (R-Super), a training framework that uses reports to directly supervise and improve tumor segmentation. R-Super introduces new loss functions that teach segmentation models to segment tumors that match report descriptions of tumor count, sizes, and locations. Reports are only used for training. We evaluated R-Super on kidney and pancreatic tumor segmentation, exploring diverse training data sizes, up to 41,418 CT-Report pairs plus 3488 CT-Mask pairs. On external validation, R-Super increased tumor detection F1-Score and segmentation DSC by up to +15% with respect to mask-only training. It also surpassed alternative methods such as CLIP and multi-task learning. It enabled pancreatic tumor detection on non-contrast CT without requiring tumor masks from non-contrast CT. In summary, by leveraging numerous readily available reports to supplement scarce masks, R-Super strongly improves AI performance when very few training masks are available (e.g., 50), and when many masks are available (e.g., 3488), unlocking scale in tumor segmentation. Project: https://github.com/MrGiovanni/R-Super.
Keywords:
Tumor detection
Tumor segmentation
Radiology reports
Weak supervision
Computed tomography
Journal
IF:
11.8
Papers:
3.9K
Citations:
2.4W
Organization
View more organizations
Cited Papers
Rubik's Cube plus : A self-supervised feature learning framework for 3D medical image analysis.png)
MEDICAL IMAGE ANALYSIS
MEDICAL IMAGE ANALYSISIF11.8
PREAI
ShareSaveShareSaveShareSaveLarge-scale pancreatic cancer detection via non-contrast CT and deep learningNATURE MEDICINE
OAAI
NATURE MEDICINEIF50
OAAIShareSaveAbdomenAtlas: A large-scale, detailed-annotated, & multi-center dataset for efficient transfer learning and open algorithmic benchmarkingMEDICAL IMAGE ANALYSIS
MEDICAL IMAGE ANALYSISIF11.8
PREAI
ShareSaveShareSaveShareSave
ShareSave View more