added a function to save heatmaps

nnunetv2/inference/export_prediction.py

@@ -4,6 +4,7 @@
 
 import numpy as np
 import torch
+import matplotlib.pyplot as plt
 from acvl_utils.cropping_and_padding.bounding_boxes import bounding_box_to_slice
 from batchgenerators.utilities.file_and_folder_operations import load_json, isfile, save_pickle
 
@@ -12,6 +13,47 @@
 from nnunetv2.utilities.plans_handling.plans_handler import PlansManager, ConfigurationManager
 
 
+
+def save_heatmap(probabilities, output_path):
+    # Check if the probabilities array is 4D and handle each channel and slice separately
+    if probabilities.ndim == 4:  # Shape: (num_channels, num_slices, width, height)
+        num_channels = probabilities.shape[0]
+        num_slices = probabilities.shape[1]
+        output_dir = os.path.dirname(output_path)
+
+        # Create a directory for the channel and slice images if it doesn't exist
+        os.makedirs(output_dir, exist_ok=True)
+
+        for channel in range(num_channels):
+            for slice_idx in range(num_slices):
+                slice_output_path = os.path.join(output_path, f"channel_{channel}_slice_{slice_idx}.png")
+                plt.figure(figsize=(10, 8))
+                plt.imshow(probabilities[channel, slice_idx], cmap='hot', interpolation='nearest', origin='lower')
+                plt.colorbar()
+                plt.savefig(slice_output_path)
+                plt.close()
+    elif probabilities.ndim == 3:  # Handle the 3D case separately (num_slices, width, height)
+        num_slices = probabilities.shape[0]
+        output_dir = os.path.dirname(output_path)
+
+        os.makedirs(output_dir, exist_ok=True)
+
+        for i in range(num_slices):
+            slice_output_path = os.path.join(output_path, f"slice_{i}.png")
+            plt.figure(figsize=(10, 8))
+            plt.imshow(probabilities[i], cmap='hot', interpolation='nearest', origin='lower')
+            plt.colorbar()
+            plt.savefig(slice_output_path)
+            plt.close()
+    else:
+        # If the array is already 2D, save it directly
+        plt.figure(figsize=(10, 8))
+        plt.imshow(probabilities, cmap='hot', interpolation='nearest', origin='lower')
+        plt.colorbar()
+        plt.savefig(output_path)
+        plt.close()
+
+
 def convert_predicted_logits_to_segmentation_with_correct_shape(predicted_logits: Union[torch.Tensor, np.ndarray],
                                                                 plans_manager: PlansManager,
                                                                 configuration_manager: ConfigurationManager,
@@ -97,6 +139,12 @@ def export_prediction_from_logits(predicted_array_or_file: Union[np.ndarray, tor
         segmentation_final, probabilities_final = ret
         np.savez_compressed(output_file_truncated + '.npz', probabilities=probabilities_final)
         save_pickle(properties_dict, output_file_truncated + '.pkl')
+
+        # Save heatmaps of the probabilities
+        heatmap_output_path = f"{output_file_truncated}_heatmap"
+        os.makedirs(heatmap_output_path, exist_ok=True)
+        save_heatmap(probabilities_final, heatmap_output_path)
+
         del probabilities_final, ret
     else:
         segmentation_final = ret