Exploring foundation models for multi-class muscle segmentation in MR images of neuromuscular disorders: A comparative analysis of accuracy and uncertainty
SAM's fine-tuned encoder/decoder outperformed nnU-Net 3D in Moderate and Severe cases (DSC: 0.886 vs 0.883 and 0.857 vs 0.850) and was comparable in Early (DSC: 0.925). MedSAM did not show an advantage over SAM. Regarding UQ, SAM exhibited superior calibration in Moderate and Severe groups (ECE: 3.6% vs. 5.1% and 3.3% vs. 7.1%), CONCLUSIONS:: In conclusion, our findings demonstrate that fine-tuning SAM yields superior performance, considering both accuracy and UQ metrics, highlighting its enhanced reliability in challenging NMD imaging scenarios.
Authors
- Giuseppe Baselli (ORCID: https://orcid.org/0000-0003-2978-1704)
- Marco Moscatelli (ORCID: https://orcid.org/0000-0002-6745-474X)
- Marina Grisoli (ORCID: https://orcid.org/0000-0002-7663-446X)
- Domenico Aquino (ORCID: https://orcid.org/0000-0002-1289-9573)
- Alessandro Brusaferri (ORCID: https://orcid.org/0000-0003-0838-9826)
- Giovanna Rizzo (ORCID: https://orcid.org/0000-0002-6341-1304)
- Alfonso Mastropietro (ORCID: https://orcid.org/0000-0003-3358-7111)
- Nicola Casali
Institutions
- Tecnologie Avanzate (Italy) (IT)
- Fondazione IRCCS Istituto Neurologico Carlo Besta (IT)
- Politecnico di Milano (IT)
Publication Details
- Journal
- Computer Methods and Programs in Biomedicine
- Published
- 2025-08-28
- DOI
- https://doi.org/10.1016/j.cmpb.2025.109035
- Citations
- 1
- Primary Topic
- Extracellular vesicles in disease
- Type
- article
- Field-Weighted Citation Impact
- 0.81
Funders
- Fondazione Cariplo
- Regione Lombardia
- Ministero dell'Università e della Ricerca