Introduction

This vignette illustrates a minimal example of a single patient vs. group (1 vs Group) SCC analysis with the neuroSCC package.

Ensure you have executed the steps from the “Getting Started” vignette, or follow the conditional steps below to create required data.

Data Preparation (Conditional)

If you haven’t executed the previous vignette, let’s conditionally recreate necessary matrices.

cat("Creating required matrices from sample data...\n")
#> Creating required matrices from sample data...

databaseControls <- databaseCreator(pattern = "^syntheticControl.*\\.nii\\.gz$", control = TRUE, quiet = TRUE)
databasePathological <- databaseCreator(pattern = "^syntheticPathological1\\.nii\\.gz$", control = FALSE, quiet = TRUE)

matrixControls <- matrixCreator(databaseControls, paramZ = 35, quiet = TRUE)
matrixPathological <- matrixCreator(databasePathological, paramZ = 35, quiet = TRUE)

normalizedMatrix <- meanNormalization(matrixControls)
#> 
#>  Mean before normalization: 4.345851
#> 
#>  Normalization completed.
normalizedPathological <- meanNormalization(matrixPathological)
#> 
#>  Mean before normalization: 4.243371
#> 
#>  Normalization completed.

niftiPath <- system.file("extdata", "syntheticControl1.nii.gz", package = "neuroSCC")
contours <- neuroContour(niftiPath, paramZ = 35, levels = c(0), plotResult = FALSE)

cat("Data prepared successfully.\n")
#> Data prepared successfully.

Single Patient Data and Cloning

Prepare data for a single patient and generate synthetic Poisson clones to enable SCC computation:

# Select the single pathological patient
SCC_AD <- normalizedPathological[1, , drop = FALSE]
SCC_CN <- normalizedMatrix

# Toy parameters for Poisson clone generation (modifiable by user)
numClones <- 2
factorLambda <- 0.1

# Generate synthetic clones
SCC_AD_clones <- generatePoissonClones(SCC_AD, numClones, factorLambda)

# Combine the patient with generated clones
SCC_AD_expanded <- rbind(SCC_AD, SCC_AD_clones)
SCC_AD_expanded <- meanNormalization(SCC_AD_expanded)
#> 
#>  Mean before normalization: 1.065060
#> 
#>  Normalization completed.

SCC Estimation (Conditional Execution)

This step requires the external ImageSCC package, currently not on CRAN. Ensure you have this package installed. If not, install it using:

if (requireNamespace("ImageSCC", quietly = TRUE)) {
  message("'ImageSCC' package is available.")
} else {
  message("This vignette requires the 'ImageSCC' package.")
  message("You can install it from GitHub with:")
  message("  remotes::install_github('FIRST-Data-Lab/ImageSCC')")
}
#> 'ImageSCC' package is available.

The SCC computation is typically computationally intensive. Here, we skip actual computation and load a precomputed result from the package (SCCcomp.RData):


# Check for Triangulation package
triangulation_available <- requireNamespace("Triangulation", quietly = TRUE)

if (triangulation_available) {
  message("'Triangulation' package is available.")
} else {
  message("'Triangulation' package not available.")
  message("Install with: remotes::install_github('FIRST-Data-Lab/Triangulation')")
}
#> 'Triangulation' package is available.

# Proceed only if both packages are available
if (triangulation_available) {

  # Try loading precomputed SCC object from data/
  if (!exists("SCCcomp", inherits = FALSE) &&
      "SCCcomp" %in% data(package = "neuroSCC")$results[, "Item"]) {
    
    message("Loading precomputed SCC object from package data...")
    suppressMessages(data("SCCcomp", package = "neuroSCC"))

  } else if (!exists("SCCcomp", inherits = FALSE)) {
    
    message("Precomputed object not found. Running SCC estimation...")

    # 1. Prepare contour and triangulation
    Z <- as.matrix(contours[[1]][, c("x", "y")])
    VT <- Triangulation::TriMesh(contours[[1]], n = 15)
    V <- as.matrix(VT[[1]])
    Tr <- as.matrix(VT[[2]])

    # 2. Run SCC estimation
    SCCcomp <- ImageSCC::scc.image(
      Ya = SCC_AD_expanded,
      Yb = SCC_CN,
      Z = Z,
      d.est = 5,
      d.band = 2,
      r = 1,
      V.est.a = V,
      Tr.est.a = Tr,
      V.band.a = V,
      Tr.band.a = Tr,
      penalty = TRUE,
      lambda = 10^{seq(-6, 3, 0.5)},
      alpha.grid = c(0.10, 0.05, 0.01),
      adjust.sigma = TRUE
    )
  }
}
#> Loading precomputed SCC object from package data...

Extracting Significant Points and Evaluating Metrics

Once the SCC computation results (SCCcomp) are available, we can extract significant points and compute relevant performance metrics:

# Extract significant points
significantPoints <- getPoints(SCCcomp)

# Load example true ROI data from the package
roi_path <- system.file("extdata", "ROIsample_Region2_18.nii.gz", package = "neuroSCC")
trueROI <- processROIs(roi_path, region = "Region2", number = "18", save = FALSE)
#> Loading NIfTI file...

# Get dimensions for total coordinates
dimensions <- getDimensions(roi_path)
totalCoords <- expand.grid(x = 1:dimensions$xDim, y = 1:dimensions$yDim)

# Calculate performance metrics
metrics <- calculateMetrics(
  detectedPoints = significantPoints$positivePoints,
  truePoints = trueROI,
  totalCoords = dimensions,
  regionName = "SinglePatient_vs_Group"
)

# Display metrics
print(metrics)
#>                   region sensitivity specificity      PPV      NPV
#> 1 SinglePatient_vs_Group    85.12872    95.62596 97.22839 78.10603

Conclusion