/// <summary>
/// Calculates OA grade from LBP features using pretrained PCA an regression with given model.
/// </summary>
/// <param name="features">MRELBP features.</param>
/// <param name="mod">Model that includes PCA eigenvectors, mean feature, and regression weights.</param>
/// <param name="path">Path to save results.</param>
/// <param name="samplename">Name of analysed sample. This is saved with estimated grade to results.csv</param>
/// <param name="grade"></param>
public static bool FeaturesToGrade(double[,] features, Model mod, string path, string samplename, out string grade)
{
// Centering
double[,] dataAdjust = Processing.SubtractMean(features, mod.mean);
// Whitening and PCA matrix
int w = mod.eigenVectors.GetLength(0);
double[,] transform = new double[w, mod.nComp];
for (int i = 0; i < w; i++)
{
for (int j = 0; j < mod.nComp; j++)
{
transform[i, j] = mod.eigenVectors[i, j] / mod.singularValues[j];
}
}
double[,] PCA = dataAdjust.Dot(transform);
// Regression
double[] grades = PCA.Dot(mod.weights).Add(mod.intercept);
double[] logistic = PCA.Dot(mod.weightsLog).Add(mod.interceptLog);
bool log;
if (logistic[0] > 0.5)
{
Console.WriteLine("Logistic regression estimated sample as degenerated.");
log = true;
}
else
{
Console.WriteLine("Logistic regression estimated sample as healthy / mildly degenerated.");
log = false;
}
// Convert estimated grade to string
if (grades[0] < 0)
{
grade = "0.00";
}
else if (grades[0] > 3)
{
grade = "3.00";
}
else
{
grade = grades[0].ToString("###0.##", CultureInfo.InvariantCulture);
}
return(log);
}
/// <summary>
/// Calculates OA grade prediction.
/// </summary>
/// <param name="mod">Loaded model.</param>
/// <param name="features">LBP features.</param>
/// <param name="volume">Data array.</param>
/// <returns>Returns string containing the OA grade</returns>
public static string Predict(Model mod, ref int[,] features, ref Rendering.renderPipeLine volume)
{
// Default variables
int threshold = 50;
int[] size = { 400, 30 };
//int threshold = 5;
//int[] size = { 10, 3 };
// Load default model
string state = LoadModel(ref mod);
// Surface extraction
Processing.SurfaceExtraction(ref volume, threshold, size, out int[,] surfacecoordinates, out byte[,,] surface);
// Mean and std images
Processing.MeanAndStd(surface, out double[,] meanImage, out double[,] stdImage);
//
// LBP features
//
LBPLibrary.Functions.Save(@"C:\Users\sarytky\Desktop\trials\mean.png", meanImage, true);
LBPLibrary.Functions.Save(@"C:\Users\sarytky\Desktop\trials\std.png", stdImage, true);
features = LBP(meanImage.Add(stdImage));
// PCA
double[,] dataAdjust = Processing.SubtractMean(features.ToDouble());
double[,] PCA = dataAdjust.Dot(mod.eigenVectors.ToDouble());
// Regression
double[] grade = PCA.Dot(mod.weights).Add(1.5);
//double sum = CompareGrades(grade);
return("OA grade: " + grade[0].ToString("####.##", CultureInfo.InvariantCulture));
//return "Sum of differences between pretrained model and actual grade: " + sum.ToString("###.###", CultureInfo.InvariantCulture);
}