Image Segmentation Evaluation

Evaluating segmentation algorithms using SimpleITK filters (Python) and results of statistical analysis (R) in order to demonstrate the statistical significance of comparisons between used models.

---

Table of contents

• Evaluation
  • Preparing images for evaluation
  • Get Measures
  • Show GT
• Statistical Analysis

---

Evaluation

The python library can be found here.

Preparing images for evaluation

In order to apply and evaluate segmentation models, images must be in a dictionary with the following structure:

dict[Rater][Image] = [list_of_segmentations]

Where

• Rater is rater's ground truth name
• Image is image's path
• list_of_segmentations is the list of GT segmentations.

For example:

data_dir = "/home/"
dataset_folders = sorted(glob.glob(os.path.join(data_dir,"subset*")))

images = []

segmentations = {"Rater_Name":{}}


for folder in dataset_folders:
    images = images + sorted(glob.glob(folder + "/*.mhd"))

for image in images:
        name = os.path.split(image)[1]
        segmentation_file_names = glob.glob(os.path.join(data_dir,'seg-lungs/',name))
        segmentations["Rater_Name"][image] = [segmentation_file_names]

GetMeasures

Get Measures of applying Neural Network to an image and its evaluation by SimpleITK filters. These measures will be saved in a csv file. Here is an example of how to apply this method.

rois_labels = {1:"Right_Lung", 2:"Left_Lung", 3:"Both_Lungs"}
GT_label_values = [2,3]
saving_path = "/path/results"
    
measures = GetMeasures(subset_name, segmentations, rois_labels, saving_path, GT_label_values = GT_label_values,
                       debug_mode = False, flip_image = True, flip_GT = True, image_extension = ".mhd", GT_extension = ".mhd")

Images and GTs must be in axial plane, as it is shown in the following pictures. If you need to rotate them, turn on flip_image and / or flip_GT.

ShowGT

This function shows all GTs that are in the dictionary segmentations and their labels.

ShowGT(segmentations,flip_GT)  

---

Statistical Analysis