update prostate config

Author: taigw

examples/prostate/config/data/image_test_gt_seg.csv

@@ -0,0 +1,11 @@
+ground_truth,segmentation
+label/promise_40.nii.gz,promise_40.nii.gz
+label/promise_20.nii.gz,promise_20.nii.gz
+label/promise_44.nii.gz,promise_44.nii.gz
+label/promise_45.nii.gz,promise_45.nii.gz
+label/promise_41.nii.gz,promise_41.nii.gz
+label/promise_48.nii.gz,promise_48.nii.gz
+label/promise_10.nii.gz,promise_10.nii.gz
+label/promise_31.nii.gz,promise_31.nii.gz
+label/promise_15.nii.gz,promise_15.nii.gz
+label/promise_09.nii.gz,promise_09.nii.gz

examples/prostate/config/data/image_test_names.txt

@@ -3,13 +3,11 @@ metric = dice
 label_list = [1]
 organ_name = prostate
 
-segmentation_folder_list  = [result]
-segmentation_postfix = 
-segmentation_format  = nii.gz
-
-ground_truth_folder_list  = [/home/disk2t/data/prostate/promise12/preprocess/train/label]
-ground_truth_postfix = 
-ground_truth_format  = nii.gz
-
-patient_file_names  = config/data/image_test_names.txt
-
+ground_truth_folder_root  = /home/disk2t/data/prostate/promise12/preprocess/train
+segmentation_folder_root  = result
+evaluation_image_pair     = ./config/data/image_test_gt_seg.csv
+
+ground_truth_label_convert_source = None
+ground_truth_label_convert_target = None
+segmentation_label_convert_source = None
+segmentation_label_convert_target = None

examples/prostate/config/data/image_train_names.txt

@@ -7,6 +7,7 @@ train_csv = config/data/image_train.csv
 valid_csv = config/data/image_valid.csv
 test_csv  = config/data/image_test.csv
 
+load_pixelwise_weight = False
 # modality number
 modal_num = 1
 
@@ -59,6 +60,7 @@ device_name = cuda:0
 
 batch_size  = 4
 loss_function = dice_loss
+class_weight  = [0.5, 1.0]
 
 # for optimizers
 optimizer     = Adam

examples/prostate/config/data/image_valid_names.txt

@@ -59,38 +59,35 @@ def random_split_dataset():
     with open(test_names_file, 'w') as f:
         f.writelines(lines[:1] + test_lines)
 
-def obtain_patient_names():
-    """
-    extract the patient names from csv files
-    """
-    split_names = ['train', 'valid', 'test']
-    for split_name in split_names:
-        csv_file = 'config/data/image_{0:}.csv'.format(split_name)
-        if(not os.path.isfile(csv_file)):
-            continue
-        with open(csv_file, 'r') as f:
-            lines = f.readlines()
-        data_lines = lines[1:]
-        patient_names = []
-        for data_line in data_lines:
-            patient_name = data_line.split('/')[-1]
-            patient_name = patient_name.split('.')[0]
-            print(patient_name)
-            patient_names.append(patient_name)
-        output_filename = 'config/data/image_{0:}_names.txt'.format(split_name)
-        with open(output_filename, 'w') as f:
-            for patient_name in patient_names:
-                f.write('{0:}\n'.format(patient_name))
-        
+def get_evaluation_image_pairs(test_csv, gt_seg_csv):
+    with open(test_csv, 'r') as f:
+        input_lines = f.readlines()[1:]
+        output_lines = []
+        for item in input_lines:
+            gt_name = item.split(',')[1]
+            gt_name = gt_name.rstrip()
+            seg_name = gt_name.split('/')[-1]
+            output_lines.append([gt_name, seg_name])
+    with open(gt_seg_csv, mode='w') as csv_file:
+        csv_writer = csv.writer(csv_file, delimiter=',', 
+                            quotechar='"',quoting=csv.QUOTE_MINIMAL)
+        csv_writer.writerow(["ground_truth", "segmentation"])
+        for item in output_lines:
+            csv_writer.writerow(item)
+
+
 if __name__ == "__main__":
     # create cvs file for promise 2012
     fields      = ['image', 'label']
-    data_dir    = 'data/promise12/preprocess'
+    data_dir    = '/home/disk2t/data/prostate/promise12/preprocess/train'
     output_file = 'config/data/image_all.csv'
     create_csv_file(data_dir, output_file, fields)
 
     # split the data into training, validation and testing
     random_split_dataset()
 
-    #obtain image names the splitted dataset
-    obtain_patient_names()
+    # obtain ground truth and segmentation pairs for evaluation
+    test_csv    = "./config/data/image_test.csv"
+    gt_seg_csv  = "./config/data/image_test_gt_seg.csv"
+    get_evaluation_image_pairs(test_csv, gt_seg_csv)
+