Related Experiment Video
Updated: May 19, 2026
10:28
Gene Regulation and Targeted Therapy in Gastric Cancer Peritoneal Metastasis: Radiological Findings from Dual Energy CT and PET/CT
Published on: January 22, 2018
Multi-modal segment anything model (mmSAM) for tumor segmentation in multi-tracer oncologic PET/CT
Zhijie Fang1, Yibin Liu1, Tao Tan2
1Biomedical Imaging Laboratory (BIG), Department of Electrical and Computer Engineering, Faculty of Science and Technology, University of Macau, Taipa, Macau SAR, China.
EJNMMI Physics
|May 17, 2026
Summary
A novel multi-modal Segment Anything Model (mmSAM) shows promise for lesion segmentation in PET/CT scans. Fine-tuning mmSAM significantly improves accuracy for cross-tracer studies in oncologic imaging.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Oncologic Imaging
Background:
- Accurate lesion segmentation is crucial for dosimetry in targeted radionuclide therapy.
- Whole-body multi-tracer PET/CT imaging provides valuable diagnostic information.
- Existing segmentation methods may have limitations in multi-modal and cross-tracer scenarios.
Purpose of the Study:
- To propose and evaluate a novel 2D multi-modal Segment Anything Model (mmSAM) for lesion segmentation.
- To assess the performance of mmSAM in whole-body multi-tracer PET/CT images.
- To investigate the transferability and effectiveness of mmSAM with and without fine-tuning on different tracers.
Main Methods:
- Utilized the AutoPET 2024 dataset with 18F-PSMA, 18F-FDG, and 68Ga-PSMA PET/CT images.
- Trained and validated the mmSAM on the 18F-PSMA dataset, incorporating both PET and CT inputs.
- Evaluated mmSAM's transferability on 18F-FDG and 68Ga-PSMA datasets, with and without fine-tuning.
- Compared mmSAM against standard 2D SAM, 3D nnUNet, and thresholding methods.
- Quantified performance using metrics like Dice, HD95, SUVmean error, MTV error, TPR, PPV, and FDR.
Main Results:
- mmSAM outperformed other methods on the primary 18F-PSMA dataset (Dice: 0.76, HD95: 1.92 mm).
- Without fine-tuning, mmSAM showed varying performance on cross-tracer datasets (18F-FDG Dice: 0.61, 68Ga-PSMA Dice: 0.77).
- Fine-tuning significantly improved mmSAM's performance on cross-tracer datasets (18F-FDG Dice: 0.65, 68Ga-PSMA Dice: 0.81).
- mmSAM consistently achieved high TPR (100%) across all evaluations.
Conclusions:
- The proposed mmSAM demonstrates significant potential for lesion segmentation in multi-tracer oncologic PET/CT imaging.
- Fine-tuning is a critical step to enhance segmentation accuracy when applying mmSAM across different radiotracers.
- mmSAM offers a promising tool for improving dosimetry and treatment planning in targeted radionuclide therapy.
