Comparative Assessment of SAMM and MONAI Algorithms in Spleen Image Segmentation: Approaches and Metrics

Salvatierra-Guillermo D.P.; Zavaleta Jave V.O. · IEEE Andescon, ANDESCON 2024 - Proceedings (2024) · DOI: 10.1109/andescon61840.2024.10755771

Resumen

This study presents a comparative assessment of two prominent open-source segmentation algorithms, SAMM (Segment Any Medical Model) and MONAI (Medical Open Network for AI), in the context of spleen image segmentation. While both algorithms leverage artificial intelligence for medical image analysis, they differ significantly in their design and performance. Using a dataset of 10 thoracic CT scans from the 'Task09 _Spleen' collection, we evaluated these algorithms on a laptop with standardized hardware specifications. The segmentation quality was assessed using three key metrics: Dice Similarity Index (DSC), Average Hausdorff Distance (AHD), and Volumetric Similarity (VS), all computed using the MISeval library. Our results consistently demonstrate MONAI's superior performance across all metrics, with mean scores of 0.956 for DSC, 6.258 for AHD, and 0.984 for VS, compared to SAMM's 0.719, 29.474, and 0.858 respectively. These findings not only highlight MONAI's enhanced precision and efficacy in spleen segmentation but also underscore its potential as a more reliable tool for medical image analysis in similar contexts.

Instituciones: Pontificia Universidad Católica del Perú