edited shap functions and notebook

notebooks/custom_functions/shap_functions.py

 import numpy as np
 import cv2
+import matplotlib.pyplot as plt
+import shap
+
 
 
 __all__ = [
     'standardize',
     'get_background',
-    'get_organoidlike_background'
+    'get_organoidlike_background',
+    'print_shap_images'
 ]
 
 def standardize(input_arr):
@@ -14,8 +18,13 @@ def standardize(input_arr):
     x /= (np.std(x, keepdims=True) + 1e-6)
     return x
 
-def get_background(img):
-
+def compute_background(img):
+    if len(img.shape)==4:
+        img = img[0]
+    elif len(img.shape)==3:
+        pass
+    else:
+        raise Exception("Shape of the imput image must be equal to 3 or 4!")
     gray = cv2.cvtColor(img.astype(np.uint8), cv2.COLOR_RGB2GRAY)
     ret, thresh = cv2.threshold(gray, 100, 255,  cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
     ret, labels = cv2.connectedComponents(thresh)
@@ -24,6 +33,12 @@ def get_background(img):
 
     background_label = np.argsort(areas)[-2]
     organoid_label = np.argsort(areas)[-1]
+    
+    return img, labels, areas, background_label, organoid_label
+    
+def get_background(img):
+
+    img, labels, areas,_ , background_label = compute_background(img)
 
     background_area = areas[background_label]
     background_mask = (labels == background_label)
@@ -31,16 +46,39 @@ def get_background(img):
     return np.expand_dims((np.ones_like(img)* mean_background), axis = 0)
 
 def get_organoidlike_background(img):
-    gray = cv2.cvtColor(img.astype(np.uint8), cv2.COLOR_RGB2GRAY)
-    ret, thresh = cv2.threshold(gray, 100, 255,  cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
-    ret, labels = cv2.connectedComponents(thresh)
     
-    areas = np.sum((labels == np.arange(ret).reshape(-1,1,1)), axis = (1,2))
-
-    background_label = np.argsort(areas)[-2]
-    organoid_label = np.argsort(areas)[-1]
+    img, labels, areas, organoid_label,_ = compute_background(img)
 
     organoid_area = areas[organoid_label]
     organoid_mask = (labels == organoid_label)
     mean_background = (img[:,:,0] * organoid_mask).sum()/organoid_area
-    return np.expand_dims((np.ones_like(img)* mean_background), axis = 0)
\ No newline at end of file
+    return np.expand_dims((np.ones_like(img)* mean_background), axis = 0)
+
+def print_shap_images(model = None,
+                      image_to_process = None,
+                      reference_image = None,
+                      image_name = None,
+                      background = None,
+                      config_dict = None,
+                      scores_dict = None,
+                      output_file_path = None
+                     ):
+    
+    explainer = shap.DeepExplainer(model, background)
+    shap_values = explainer.shap_values(image_to_process)
+    shap.image_plot(shap_values,
+                reference_image.astype(np.float32),
+                show = False
+               )
+    title = image_name
+    title = " ".join([title, "   Network type:", config_dict['keras_model_name']])
+    title = "\n".join([title, "Network prediction:"])
+    title = " ".join(
+        [title, 
+         str(np.round(model.predict(image_to_process)[0,0], decimals=3)),
+         "    Threshold: ",
+         str(np.round(scores_dict['threshold'], decimals = 3))
+        ]
+    )
+    plt.suptitle(title, y=0.9)
+    plt.savefig(output_file_path, dpi = 600)
\ No newline at end of file