Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SAMM (Segment Any Medical Model): A 3D Slicer Integration to SAM

Yihao Liu, Jiaming Zhang, Zhangcong She, Amir Kheradmand, Mehran Armand

TL;DR

This work introduces SAMM, an open-source integration that brings the Segment Anything Model into medical image analysis via 3D Slicer. It presents a dual-component architecture with a SAM Server and Slicer-IPP that precompute per-volume embeddings and enable real-time, prompt-based segmentation across CT, MRI, and US data. The authors demonstrate end-to-end segmentation latency around 0.612 seconds per cycle and embedding times of approximately 162.9 seconds for a typical MRI volume, indicating practical performance for semi-automatic annotation. They discuss initialization stability and outline future directions including domain-specific fine-tuning, text prompts, and MONAI-based enhancements, positioning SAMM as a flexible platform for medical foundation model research.

Abstract

The Segment Anything Model (SAM) is a new image segmentation tool trained with the largest available segmentation dataset. The model has demonstrated that, with prompts, it can create high-quality masks for general images. However, the performance of the model on medical images requires further validation. To assist with the development, assessment, and application of SAM on medical images, we introduce Segment Any Medical Model (SAMM), an extension of SAM on 3D Slicer - an image processing and visualization software extensively used by the medical imaging community. This open-source extension to 3D Slicer and its demonstrations are posted on GitHub (https://github.com/bingogome/samm). SAMM achieves 0.6-second latency of a complete cycle and can infer image masks in nearly real-time.

SAMM (Segment Any Medical Model): A 3D Slicer Integration to SAM

TL;DR

This work introduces SAMM, an open-source integration that brings the Segment Anything Model into medical image analysis via 3D Slicer. It presents a dual-component architecture with a SAM Server and Slicer-IPP that precompute per-volume embeddings and enable real-time, prompt-based segmentation across CT, MRI, and US data. The authors demonstrate end-to-end segmentation latency around 0.612 seconds per cycle and embedding times of approximately 162.9 seconds for a typical MRI volume, indicating practical performance for semi-automatic annotation. They discuss initialization stability and outline future directions including domain-specific fine-tuning, text prompts, and MONAI-based enhancements, positioning SAMM as a flexible platform for medical foundation model research.

Abstract

The Segment Anything Model (SAM) is a new image segmentation tool trained with the largest available segmentation dataset. The model has demonstrated that, with prompts, it can create high-quality masks for general images. However, the performance of the model on medical images requires further validation. To assist with the development, assessment, and application of SAM on medical images, we introduce Segment Any Medical Model (SAMM), an extension of SAM on 3D Slicer - an image processing and visualization software extensively used by the medical imaging community. This open-source extension to 3D Slicer and its demonstrations are posted on GitHub (https://github.com/bingogome/samm). SAMM achieves 0.6-second latency of a complete cycle and can infer image masks in nearly real-time.
Paper Structure (6 sections, 4 figures)

This paper contains 6 sections, 4 figures.

Table of Contents

  1. Introduction
  2. Methodology
  3. Overall Architecture
  4. Slicer-IPP
  5. Experiments and Results
  6. Discussion and Conclusion

Figures (4)

  • Figure 1: Overall architecture of the integration of 3D Slicer and SAM. Images from different modalities, such as computed tomography (CT), magnetic resonance imaging (MRI), or ultrasound (US), are contained in a scalar volume node. The node (vtkMRMLScalarVolumeNode) is a class in the Visualization Toolkit (VTK) that represents a volume of the scalar data within the Medical Reality Markup Language (MRML) framework.
  • Figure 2: The affiliation of five tasks, namely are "send inference request" (SND_ INF), "receive inference request" (RCV_ INF), "complete SAM inference" (CPL_ INF), "receive mask transmission" (RCV_ MSK), and "apply mask" (APL_ MSK). Slicer-IPP and SAM Server use ZMQ/Numpy memory mapping package to enable real-time communication.
  • Figure 3: Example results for different image formats (CT, MRI, and US). The prompts with green points are for the regions to be selected, whereas the red points are for the regions to be removed.
  • Figure 4: (a) is the event plot of the 5 tasks (SND_ INF, RCV_ INF, CPL_ INF, RCV_ MSK, and APL_ MSK). Each row represents one task and each bar represents one event. A 16-second period (covers 60 cycles) from the initialization phase of SAMM (when all tasks are initially launched) to the stable phase (when all tasks run at a steady frequency) is shown. Events executed within the same segmentation cycle are marked with the same number on the bottom two panels. The time interval of five tasks between event 1 and 2 is marked on (b) while the time interval of five tasks between event 59 and 60 is marked on (c). The time intervals between two cycles are shown in the green boxes whereas the time latency of one complete cycle is highlighted in the purple box.