Upload folder using huggingface_hub

dicom_processor.py

@@ -64,7 +64,10 @@ def anonymize_dicom(ds: pydicom.dataset.FileDataset) -> pydicom.dataset.FileData
     # Wipe other fields
     for tag in PHI_TAGS:
         if tag in ds:
-            ds.data_element(tag).value = "ANONYMIZED"
+            if 'Date' in tag: # VR DA requires YYYYMMDD
+                ds.data_element(tag).value = "19010101"
+            else:
+                ds.data_element(tag).value = "ANONYMIZED"
             
     return ds
 

explainability.py

@@ -146,7 +146,12 @@ class ExplainabilityEngine:
         # 3. Constrain
         # Resize seg_mask to match gradcam_map (both should be HxW float 0..1)
         if gradcam_map is None:
-            return {"heatmap": None, "original": None, "confidence": "LOW"}
+            return {
+                "heatmap_array": None, 
+                "heatmap_raw": None, 
+                "reliability_score": 0.0, 
+                "confidence_label": "LOW"
+            }
 
         # Ensure shapes match
         if seg_mask.shape != gradcam_map.shape:
@@ -184,8 +189,16 @@ class ExplainabilityEngine:
             inputs = {k: v.to(self.model.device) for k, v in inputs.items()}
             
             # Wrapper
+            # Robust get for attention_mask (some processors might not return it for image-only flows, though text is here)
+            input_ids = inputs.get('input_ids')
+            attention_mask = inputs.get('attention_mask')
+
+            if input_ids is None:
+                logger.error("Explainability: Missing input_ids in processor output")
+                return None
+
             model_wrapper_cam = HuggingFaceWeirdCLIPWrapper(
-                self.model, inputs['input_ids'], inputs['attention_mask']
+                self.model, input_ids, attention_mask
             )
             
             target_layers = [self.model.vision_model.post_layernorm]
@@ -196,7 +209,13 @@ class ExplainabilityEngine:
                 reshape_transform=reshape_transform
             )
             
-            grayscale_cam = cam(input_tensor=inputs['pixel_values'], targets=None)
+            # GradCAM needs pixel_values
+            pixel_values = inputs.get('pixel_values')
+            if pixel_values is None:
+                 logger.error("Explainability: Missing pixel_values")
+                 return None
+
+            grayscale_cam = cam(input_tensor=pixel_values, targets=None)
             grayscale_cam = grayscale_cam[0, :]
             
             # Smoothing