Spatial Atlas-Informed Loss Ensembles for Whole-Body PSMA PET/CT Metastasis Segmentation

Not provided

segmentationatlas-integrationdeep-learningmedical-imaging

2610.40834

Builder Relevance

70%

2h ago

Abstract

This study investigates the integration of a population metastasis atlas into model training to improve lesion detection in prostate metastasis segmentation.

Reality Card

Core Claim

The SAILS method improved whole-body Dice and lesion sensitivity in prostate metastasis segmentation without requiring atlas registration at inference time.

Method / Result

SAILS improved whole-body Dice to 0.531 from 0.520 (p=0.006) and increased pooled lesion sensitivity to 0.857 from 0.843 (p=0.004).

Limitations

The increase in false positives raises concerns about the balance between sensitivity and specificity.

Paper to code

Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.

No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.

← Back to all papers