Upload tumor_features.py

utils/tumor_features.py

@@ -2,13 +2,11 @@ from scipy.ndimage import label, find_objects
 import numpy as np
 import cv2
 
-
 IMAGE_SPACING_X = 0.7031
 IMAGE_SPACING_Y = 0.7031
 IMAGE_SPACING_Z = 2.5 
 
 
-
 def compute_largest_diameter(binary_mask):
     
     # Label connected components in the binary mask
@@ -37,19 +35,44 @@ def compute_largest_diameter(binary_mask):
     return largest_diameter / 10 # IN CM
 
 
+def compute_shape(binary_mask):
 
-
-def generate_features(img, liver, tumor):
+    contours, _ = cv2.findContours(binary_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
     
-    contours, _ = cv2.findContours(mask_image, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+    shape, margin = None, None
+
+    if contours:
+        tumor_contour_area = cv2.contourArea(contours[0])
+
+        tumor_contour_perimeter = cv2.arcLength(contours[0], True)
+        epsilon = 0.02 * tumor_contour_perimeter
+        approx = cv2.approxPolyDP(contours[0], epsilon, True)
+        num_sides = len(approx)
 
+        # determine the shape based on the number of sides
+        if num_sides < 5: shape = "Round"
+        else: shape = "Irregular"
+
+        # determine the margin characteristics based on solidity
+        hull = cv2.convexHull(contours[0])
+        hull_area = cv2.contourArea(hull)
+        tumor_solidity = tumor_contour_area / hull_area
+        if tumor_solidity > 0.95: margin = "Smooth"
+        elif tumor_solidity > 0.80: margin = "Irregular"
+        else: margin = "Spiculated"
+
+    return shape, margin
+
+def generate_features(img, liver, tumor):
     
+    # shape, margin = compute_shape(tumor)
+    shape, margin = "irregular", "irregular"
     features = {
         "lesion size (cm)": compute_largest_diameter(tumor),
-        "lesion shape": "irregular",
+        "lesion boundary": margin,
+        "lesion shape": shape, 
         "lesion density (HU)": np.mean(img[tumor==1]),
         "involvement of adjacent organs:": "Yes" if np.sum(np.multiply(liver==0, tumor)) > 0 else "No"
     }
-
     
     return features