MAPSeg: self-supervised colorectal polyp segmentation via memory-augmented framework and synthetic polyp simulation
-
By
-
Claudia Delprete
-
Domenico Buongiorno
-
Roberto Maria Scardigno
-
Elena Sibilano
-
Antonio Brunetti
-
Vitoantonio Bevilacqua
-
May 14, 2026
-
Clinical Scorecard: MAPSeg: A Memory-Augmented Framework for Self-Supervised Segmentation of Colorectal Polyps Using Synthetic Simulation
At a Glance
| Category | Detail |
|---|
| Condition | Colorectal Cancer |
| Key Mechanisms | Self-supervised learning and anomaly detection using synthetic simulation of polyps (SIMPO). |
| Target Population | Patients undergoing colonoscopy for colorectal cancer screening. |
| Care Setting | Clinical settings utilizing colonoscopy for polyp detection. |
Key Highlights
- MAPSeg outperforms existing unsupervised methods by 23% in Intersection over Union and 12% in DICE score.
- Utilizes SIMPO to generate realistic polyp shapes from normal mucosa images.
- Reduces dependency on manual annotations for polyp segmentation.
- Demonstrates strong generalization across multiple out-of-distribution benchmarks.
- Addresses challenges in polyp detection due to variability in colorectal mucosa.
Guideline-Based Recommendations
Diagnosis
- Colonoscopy remains the gold standard for colorectal cancer screening.
Management
- Implement MAPSeg for enhanced polyp segmentation during colonoscopy.
Monitoring & Follow-up
- Regular assessment of segmentation accuracy and performance metrics.
Risks
- Potential for missed polyps due to variability in appearance and detection challenges.
Patient & Prescribing Data
Individuals at risk for colorectal cancer, particularly those undergoing routine screening.
Early detection through effective polyp segmentation can significantly improve treatment outcomes.
Clinical Best Practices
- Incorporate advanced AI techniques like MAPSeg to improve polyp detection rates.
- Utilize self-supervised learning frameworks to minimize reliance on annotated datasets.
- Regularly update training protocols to include diverse datasets for robust model performance.
Related Resources & Content
