LOGISMOS segmentation framework (Layered Optimal Graph Image Segmentation for Multiple Objects and Surfaces) facilitates highly efficient multi-dimensional, multi-layered, and multi-object optimum graph-based segmentation and surface editing on image data from various modalities (CT, MR, Ultrasound, OCT, etc.).

LOGISMOS - General approach to 3D, 4D, n-D image segmentation

Sonka team

1. Pre-segmentation that yields an approximate segmentation of the objects of interest and gives information about the topological structures of the target objects;
2. Mesh generation used to specify the structure of a graph G_B, called base graph that defines the neighboring relations among voxels on the sought (optimal) surfaces;
3. Graph construction yielding a weighted directed graph G in which each graph column corresponds to a list of vertices in G. The vertex costs of G can encode edge-based, region-based, and shape-based cost functions. Graph topology allows encoding of multi-surface/multi-object interactions;
4. Graph search yields optimal surfaces corresponding to structures of interest. The sought optimal surfaces are obtained by searching for an optimal closed set in G using efficient maximum flow graph-theory algorithms;
5. Highly efficient “Just-Enough Interaction” strategy for correction of local/regional mis-segmentations as/if needed.

• Application areas:
  • Cardiology
  • Ophthalmology
  • Orthopaedics
  • Neuroscience
  • Pulmonology
  • Radiation oncology
  • Surgery

• Publications
  • Seminal publication - Optimal Surface Segmentation in Volumetric Images—A Graph-Theoretic Approach (2006)
  • PubMed publications of the team - LOGISMOS
  • PubMed publications of the team - Graph based

• Code, library, application sharing
  • Multi-layer 3D human macular and ONH OCT LOGISMOS+JEI segmentation entitled Iowa Reference Algorithms and murine macular segmentation programs are available as downloads for research use at www.iibi.uiowa.edu/downloads, the name Iowa Reference Algorithm is used since it allows analysis of OCT images from scanners of five tested manufacturers.
  • Source code of Alpha Path-Moves is shared at github.com/sydbarrett/AlphaPathMoves
  • Developed with partial support from U01 CA140206 by R. Beichel's group, cancer-related PET image analysis modules (code, libraries, apps) for the popular 3D Slicer environment are shared at github.com/QIICR/PETTumorSegmentation. Full source code is publicly available, is free for use, and shows how to implement a graph-based segmentation method with JEI functionality.

Image Gallery

Left-to-right: Aortic CT, MR, CTA, MRA image analysis in 3D, 4D; Quantitative results of aortic analysis; 4D cardiac ultrasound; 4D cardiac ultrasound analysis.

Broad applicability of LOGISMOS n-D segmentation methods. Left-to-right: 3D liver tumor segmentation; 3D lung tumor segmentation; Prostate-bladder -- simultaneous segmentation of 2 objects in 3D CT; 3D cartilage MR segmentation -- normal and osteoarthritic joints, note cartilage thinning and ``holes'' in the right-most image.

Left-to-right: Motion artifact reduction with structure awareness in 4D CT (original images with artifacts on the left; Human macular 3D OCT - multi-layer 3D segmentation; Human peripapillary 3D OCT -- layer segmentation and 3D visualization of retinal layers around the optic nerve head; Multi-layer 3D segmentation of murine retinal OCT; First-ever 3D segmentation of human choroidal vasculature from retinal OCT using pilot implementation of the surface-and-region graph search.

Advancement of LOGISMOS flexibility. Left-to-right: 3D segmentation of human choroidal vasculature from retinal OCT using surface-and-region graph search; Simultaneous segmentation of hierarchically-structured interacting objects based on Alpha Path-Moves (ribcage+spine, liver, heart, left kidney); Knee MR segmentation (for longitudinal 4D analyses, central subplate shown); Caudate and putamen segmentation using automatically designed cost functions learned from examples – ground truth = yellow, LOGISMOS = green; White matter/gray matter cost function probability maps learned from cortical segmentation examples in 3D MRI, used for LOGISMOS-B cortical segmentation.

Left-to-right: LOGISMOS programming & development environment, showing MR/CT aortic analysis application; Simultaneous 4D knee MR bone+cartilage segmentation - 8 time-point 3D datasets registered and simultaneously segmented in 4D; Illustration of JEI on knee segmentation; IVUS 3D segmentation facilitated prediction of major adverse cardiac events in coronaries; Segmentation of coronary wall layers in 3D OCT images is a pre-requisite to prediction of cardiac allograft vasculopathy onset in heart-transplant patient; Retinal OCT image analysis with 3D JEI capability in an age-related macular degeneration patient; Brain tumor segmentation.

Videos

LOGISMOS+JEI in action: segment a brain tumor with simple shape, no editing required:

LOGISMOS+JEI in action: segment a brain tumor with complex shape using JEI and merging of objects:

The purpose of these videos is to demonstrate the capability of the algorithm, not to show the clinically correct segmentation.