Skip to contents

Visualize ROI Overlap and Dice Coefficient

Source: R/visualization.r
plot_roi_overlap.Rd

Visually compares two segmentation masks (ROIs) by overlaying them on the original thermal image. This function is primarily used for validation purposes: to compare an automated segmentation (filled layer) against a manual ground truth (contour line). It automatically calculates and displays the Dice Similarity Coefficient (DSC) in the title.

Usage

plot_roi_overlap(
  img_obj1,
  img_obj2,
  title = NULL,
  color = "green",
  alpha = 0.5,
  line_color = "white",
  palette = "inferno"
)

Arguments

img_obj1

A 'BioThermR' object. Typically the Automated/Predicted segmentation. This object must contain the raw thermal matrix. Its mask will be plotted as a filled area.

img_obj2

A 'BioThermR' object. Typically the Manual/Ground Truth segmentation. Its mask will be plotted as a contour outline. Dimensions must match img_obj1.

title

String. Custom title for the plot. If NULL (default), the title shows "ROI overlap (DICE: X.XXX)".

color

String. Fill color for the img_obj1 mask. Default is "green".

alpha

Numeric. Transparency level for the img_obj1 mask (0 to 1). Default is 0.5.

line_color

String. Line color for the img_obj2 contour. Default is "white".

palette

String. Color palette for the background thermal image (passed to scale_fill_viridis_c). Default is "inferno".

Value

A ggplot object showing the overlay.

Details

The visualization consists of three layers:

  1. Background: The raw thermal image from img_obj1.

  2. Prediction (img_obj1): The processed mask from the first object, rendered as a semi-transparent filled raster (default green).

  3. Ground Truth (img_obj2): The processed mask from the second object, rendered as a contour outline (default white).

The function calculates the Dice Similarity Coefficient (DSC) using the formula: $$DSC = \frac{2 \times |X \cap Y|}{|X| + |Y|}$$ where X and Y are the set of pixels in the two masks. A DSC of 1 indicates perfect overlap.

Examples

# \donttest{
#' # Load raw data
img_obj <- system.file("extdata", "C05.raw", package = "BioThermR")
img <- read_thermal_raw(img_obj)
# Apply automated segmentation
img1 <- roi_segment_ebimage(img, keep_largest = TRUE)
#> Auto-Segmentation: Kept largest object ( 401 pixels )

# Simple background removal: Keep everything above 24 degrees
img2 <- roi_filter_threshold(img, threshold = c(33, Inf))
#> ROI Filter applied: Keeping range [33, Inf]

# Compare them
plot_roi_overlap(img_obj1 = img1,
                 img_obj2 = img2)

# }