cs.CVSep 29, 2026

Merlin Plus: A Large-Scale, Multi-Cancer, Image-Mask-Report Dataset

Authors: Pedro R. A. S. Bassi, Wenxuan Li, Szymon Plotka, Ruby Honjol, Jakub Przado, Xinze Zhou, Kang Wang, Yang Yang, +5 more

Organizations: Johns Hopkins University · Harvard Medical School · Massachusetts General Hospital · Jagiellonian University · Stanford University · Warmian-Masurian Cancer Center · University of California, San Francisco · Johns Hopkins Medicine

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Multi-cancer segmentation in computed tomography (CT) is fundamentally limited by the scarcity of tumor masks across different organs. We present Merlin Plus, the first large-scale CT dataset with radiologist-created tumor masks across 9 organs. Merlin Plus extends the Merlin dataset by adding 1,153 per-voxel tumor masks and longitudinal metadata. To create these tumor masks, we developed a report-based active-learning framework in which radiology reports identify tumor cases for annotation and support training of a tumor segmentation model. The model generates initial masks, which radiologists review and correct to produce the final masks, reducing annotation burden while maintaining high-quality annotations. Besides tumor masks, the longitudinal metadata in Merlin Plus enables temporal modeling of cancer progression. By directly addressing the major bottleneck of limited multi-cancer segmentation masks, Merlin Plus supports scalable multi-organ cancer detection, segmentation, and longitudinal analysis in CT. Dataset is available at: https://github.com/MrGiovanni/MerlinPlus

Figures & tables

Figure 1: Examples of different tumor masks provided in Merlin Plus . Masks are 3D, and were created by 4 radiologists, assisted by report-based active learning.
Figure 2: Report-based Active Learning. Our active learning strategy has 6 steps. Steps 2-6 are repeated cyclically. (1) we train a tumor segmentation model from radiology reports (no mask); (2) the segmentation model generates tumor segmentation masks; (3) we automatically refine these masks to be more consistent with the tumor descriptions in radiology reports; (4) and we calculate a report-confidence score, which flags mask errors by comparing masks and reports; (5) masks with the lowest scores are revised by radiologists; and (6) we fine-tune the segmentation model using all available CT-Report pairs and CT-Mask pairs.

Explore similar work

CardsList

  1. RT-Super: Learning Tumor Segmentation from Longitudinal Images and Reports

    Sep 28, 2026Pedro R. A. S. Bassi, Wenxuan Li, Hanxue Gu +11Tumor SegmentationLongitudinal Imaging

    Johns Hopkins University · Harvard Medical School · Massachusetts General Hospital +4

  2. Weakly-supervised Kidney Tumor Classification from CT Scans with Multi-Instance Learning and Anatomical Filtering

    Sep 7, 2026Joonas Ariva, Dmytro FishmanMultiple Instance LearningCancer Detection

    University of Tartu, Tartu, Estonia · STACC, Tartu, Estonia · Better Medicine, Tartu, Estonia

  3. ORACLE-CT: Anatomy-Aware Support Pooling for CT Classification

    Jun 3, 2026Lavsen Dahal, Yubraj Bhandari, Geoffrey Rubin +1Semi-Supervised Medical Image SegmentationGlobal Average Pooling

    Electrical and Computer Engineering, Pratt School of Engineering, Duke University, Durham, 27708, North Carolina, USA · Department of Mathematics, Trinity College of Arts & Sciences, Duke University, Durham, 27708, North Carolina, USA · Department of Radiology and Imaging Sciences, University of Arizona College of Medicine, Tucson, 85004 Arizona, USA +1