Fig 1.
Schematic of the proposed GD-GLCM image training method.
SLP denotes Single-Layer Perceptron, MLP denotes Multi-Layers Perceptron, and HL denotes Hidden Layers.
Fig 2.
Cortical brain region image of a fetal term sheep exposed to global hypoxia ischemia (a) shows sampling region in the parasagittal cortex of the near-term fetal sheep brain, (b) shows a NeuN- positive stained image of the cortical brain region of a term sheep exposed to global hypoxia consisting both healthy and dying cells, (c) images of healthy cells and, (d) images of dying cells.
Fig 3.
Colour coded gradient-magnitude and gradient-direction maps of healthy and dying cells (a) Gradient-magnitude maps of 5 healthy cells, (b) gradient-magnitude maps of 5 dying cells, (c) gradient-direction maps of 5 healthy cells and (d) gradient-direction maps of 5 dying cells.
Fig 4.
Example of the GLCM heatmap derived from GM values.
GM values of a healthy cell (a, b, c, d) and a dying cell (e, f, g, h) when generating 16 GLCMs/cell.
Fig 5.
Example of the GLCM heatmap derived from GD values.
GD values of a healthy cell (a, b, c, d) and a dying cell (e, f, g, h) when generating 16 GLCMs/cell.
Fig 6.
Healthy and dying cells with their GM and GD colourmaps in pre and post NLM filtered states.
(a) shows two healthy and two dying cells before application of NLM filter, (b) shows those cells after application of NLM filter, (c) shows GM colourmaps of those cells before NLM filter application, (d) shows GM colourmaps of those cells after NLM filter application. (e) shows GD colourmaps of those cells before NLM filter application, and (f) shows GD colourmaps of those cells after NLM filter application.
Fig 7.
Typical error curve for training GD input datasets where GLCM size is 64×64 for MLP classifier.
Fig 8.
Sensitivity, selectivity, and accuracy bar plots of an SLP classifier.
Sensitivity (a, b, c), selectivity (d, e, f) and accuracy (g, h, i) bar plot of an SLP classifier. (Magenta dashed bar, standard method; Red slashed bar, GM-GLCM image training; Blue crossed bar, GD-GLCM image training method. The x-axis defines the size of GLCM arrays passed to the SLP network (namely, 8×8, 16×16, 32×32 and 64×64). The caption label defines how many GLCMs arrays were passed to the SLP network for training/cell (namely 16,32 or 60).
Fig 9.
ROC curves for healthy and dying cells of an SLP classifier.
Average ROC curves for healthy and dying cells using the optimised 32 GLCMs/cell category and GLCM size of 64×64 for the SLP classifier (Red dashed line–healthy cells; Blue solid line–dying cells). (a) using standard method with 64×64 GLCM size, (b) using GM-GLCM with 64×64 GLCM size, (c) using GD-GLCM with 64×64 GLCM size.
Fig 10.
AUC for healthy and dying cells bar plot of an SLP classifier.
(Magenta dashed bar, standard method; Red dashed bar, GM-GLCM method; Blue crossed bar, GD-GLCM method).
Fig 11.
Sensitivity, selectivity, and accuracy bar plots of one hidden layer MLP classifier.
Sensitivity (a, b, c), selectivity (d, e, f) and accuracy (g, h, i) bar plot of one hidden layer MLP classifier. (Magenta dashed bar, standard method; Red slashed bar, GM-GLC method; Blue crossed bar, GD-GLCM method). The x-axis defines the size of GLCM arrays passed to the SLP network (namely, 8×8, 16×16, 32×32 and 64×64). The caption label defines how many GLCMs arrays were passed to the SLP network for training/cell (namely 16,32 or 60).
Fig 12.
ROC curves for healthy and dying cells of one hidden layer MLP classifier.
Average ROC curves for healthy and dying cells using the optimised 32 GLCMs/cell category and GLCM size of 64×64 for the one hidden layer MLP classifier (Red dashed line–healthy cells; Blue solid line–dying cells). (a) using standard method with 64×64 GLCM size, (b) using GM-GLCM with 64×64 GLCM size, (c) using GD-GLCM with 64×64 GLCM size.
Fig 13.
AUC for healthy and dying cells bar plot of one hidden layer MLP classifier.
(Magenta dashed bar, standard method; Red dashed bar, GM-GLCM method; Blue crossed bar, GD-GLCM method).
Fig 14.
Sensitivity, selectivity, and accuracy bar plots of two hidden layer MLP classifier.
Sensitivity (a, b, c), selectivity (d, e, f) and accuracy (g, h, i) bar plot of two hidden layer MLP classifier. (Magenta dashed bar, standard method; Red slashed bar, GM-GLCM method; Blue crossed bar, GD-GLCM method). The x-axis defines the size of GLCM arrays passed to the SLP network (namely, 8×8, 16×16, 32×32 and 64×64). The caption label defines how many GLCMs arrays were passed to the SLP network for training/cell (namely 16, 32 or 60).
Fig 15.
ROC curves for healthy and dying cells of two hidden layer MLP classifier.
Average ROC curves for healthy and dying cells using the optimised 32 GLCMs/cell category and GLCM size of 64×64 for the two hidden layer MLP classifier (Red dashed line–healthy cells; Blue solid line–dying cells). (a) using standard method with 64×64 GLCM size, (b) using GM-GLCM with 64×64 GLCM size, (c) using GD-GLCM with 64×64 GLCM size.
Fig 16.
AUC for healthy and dying cells bar plot of two hidden layer MLP classifier.
(Magenta dashed bar, standard method; Red dashed bar, GM-GLCM method; Blue crossed bar, GD-GLCM method).