gaussian mixture model (gmm)  (MathWorks Inc)

Journal: Nmr in Biomedicine

Article Title: Cluster Analysis of VERDICT MRI for Cancer Tissue Characterization in Neuroendocrine Tumors

doi: 10.1002/nbm.70050

Figure Lengend Snippet: Scatterplot of R against f IC for all tumor voxels ( n = 11,519) and subjects (A; left side). The black contours show the 2D Gaussian mixture model (GMM) fit with each voxel data point color‐coded based on the probability of belonging to each component (blue, green, and red). Contours of the three individual GMM components are shown as smaller plots (right side). R and f IC maps of tumor ROIs were used to generate color‐coded posterior probability maps of each GMM component (B; Subject 6 shown as example).

Article Snippet: A 2D Gaussian mixture model (GMM) was then fitted to the f IC and R values of all tumor voxels using an algorithm based on the MATLAB function fitgmdist with 20 random initializations to avoid local optima and a regularization value of 3.5 × 10 −3 to avoid overfitting ( mathworks.com/matlabcentral/fileexchange/71496‐identification‐of‐subregions‐in‐parameter‐maps‐by‐gmm ) [ ].

Techniques:

Journal: Nmr in Biomedicine

Article Title: Cluster Analysis of VERDICT MRI for Cancer Tissue Characterization in Neuroendocrine Tumors

doi: 10.1002/nbm.70050

Figure Lengend Snippet: Model fit output of the Gaussian mixture model (GMM) of three clusters fitted to R and f IC for all tumor voxels. The table shows cluster ID as defined by histological analysis (Figures 4 and 6 ), mean values of R and f IC ( μ ) for each cluster, and the cluster fraction indicating the percentage of tumor voxel data that is associated with each Gaussian component.

Article Snippet: A 2D Gaussian mixture model (GMM) was then fitted to the f IC and R values of all tumor voxels using an algorithm based on the MATLAB function fitgmdist with 20 random initializations to avoid local optima and a regularization value of 3.5 × 10 −3 to avoid overfitting ( mathworks.com/matlabcentral/fileexchange/71496‐identification‐of‐subregions‐in‐parameter‐maps‐by‐gmm ) [ ].

Techniques:

Journal: Nmr in Biomedicine

Article Title: Cluster Analysis of VERDICT MRI for Cancer Tissue Characterization in Neuroendocrine Tumors

doi: 10.1002/nbm.70050

Figure Lengend Snippet: Gaussian mixture model (GMM) probability maps from the VERDICT cluster analysis of R and f IC (left columns) and classification maps from the histology analysis (right columns). The colors in the histology classification maps represent different tissue types: necrotic (red), fibrotic (blue), and viable cancer cells (green). Black pixels indicate areas where no stain was present. The colors in the VERDICT cluster maps represent the probability of each voxel belonging to the GMM clusters, with colors chosen for each cluster to best match with the histology maps.

Article Snippet: A 2D Gaussian mixture model (GMM) was then fitted to the f IC and R values of all tumor voxels using an algorithm based on the MATLAB function fitgmdist with 20 random initializations to avoid local optima and a regularization value of 3.5 × 10 −3 to avoid overfitting ( mathworks.com/matlabcentral/fileexchange/71496‐identification‐of‐subregions‐in‐parameter‐maps‐by‐gmm ) [ ].

Techniques: Staining