RichardObi · RichardObi · Nov 29, 2023 · Nov 27, 2023 · Nov 27, 2023 · Nov 27, 2023
diff --git a/README.md b/README.md
diff --git a/config/global.json b/config/global.json
@@ -952,7 +952,7 @@
     "00007_INPAINT_BRAIN_MRI": {
         "execution": {
             "package_name": "00007_INPAINT_BRAIN_MRI",
-            "package_link": "https://zenodo.org/record/7042217/files/00007_INPAINT_BRAIN_MRI.zip?download=1",
+            "package_link": "https://zenodo.org/records/10214796/files/00007_INPAINT_BRAIN_MRI.zip?download=1",
             "model_name": "inp_gen",
             "extension": ".pth",
             "image_size": [
@@ -3251,7 +3251,7 @@
                 "pytorch-lightning==1.4.7",
                 "pandas",
                 "comet-ml",
-                "monai",
+                "monai<=1.0.1",
                 "grad-cam",
                 "matplotlib",
                 "monai[skimage]",
@@ -3354,7 +3354,145 @@
                 "view: default=\"la\" help=",
                 "subcat: default=\"2ch\", help="
             ],
-            "comment": ""
+            "comment": "Conditional WGAN-GP Model for Cardiac Image generation with age offset (Trained on UK Biobank). A conditional wasserstein generative adversarial network with gradient penalty (WGAN_GP) that generates MRI cardiac images. The pixel dimension of the generated images is 256x256. The uploaded ZIP file contains the files model.ckpt (model weights), __init__.py (image generation method and utils), a requirements.txt, and the used GAN training config file. A sample_image.png is provided for example generation."
         }
-    }
+    },
+    "00023_PIX2PIXHD_BREAST_DCEMRI": {
+         "execution": {
+            "package_name": "00023",
+            "package_link": "https://zenodo.org/records/10215478/files/00023.zip?download=1",
+            "model_name": "30_net_G",
+            "extension": ".pth",
+            "image_size": [
+               512, 512
+            ],
+            "dependencies": [
+               "numpy",
+               "torch",
+               "torchvision",
+               "pillow"
+            ],
+            "generate_method": {
+               "name": "generate",
+               "args": {
+                  "base": [
+                     "model_file",
+                     "num_samples",
+                     "output_path",
+                     "save_images"
+                  ],
+                  "custom": {
+                     "input_path": "input/",
+                     "image_size": "512",
+                     "gpu_id": "0"
+                  }
+               }
+            }
+         },
+         "selection": {
+            "performance": {
+               "SSIM": 0.726,
+               "MSE": 34.88,
+               "NSME": null,
+               "PSNR": 32.91,
+               "IS": null,
+               "FID": 28.71,
+               "turing_test": "",
+               "downstream_task": {
+                  "CLF": {
+                     "trained_on_fake": {
+                        "accuracy": null,
+                        "precision": null,
+                        "recall": null,
+                        "f1": null,
+                        "specificity": null,
+                        "AUROC": null,
+                        "AUPRC": null
+                     },
+                     "trained_on_real_and_fake": {},
+                     "trained_on_real": {}
+                  },
+                  "SEG": {
+                     "trained_on_fake": {
+                        "dice": 0.687,
+                        "jaccard": null,
+                        "accuracy": null,
+                        "precision": null,
+                        "recall": null,
+                        "f1": null
+                     },
+                     "trained_on_real_and_fake": {
+                        "dice": "0.797"
+                     },
+                     "trained_on_real": {
+                        "dice": "0.790"
+                     }
+                  }
+               }
+            },
+            "use_cases": [
+               "segmentation",
+               "tumour localization",
+               "classification",
+               "simulation"
+            ],
+            "organ": [
+               "breast"
+            ],
+            "modality": [
+               "dce-mri",
+               "mri",
+               "t1",
+               "t1-weighted",
+               "fat-saturated"
+            ],
+            "vendors": [],
+            "centres": [
+               "Duke Hospital"
+            ],
+            "function": [],
+            "condition": [],
+            "dataset": [
+               "DUKE"
+            ],
+            "augmentations": [],
+            "generates": [],
+            "height": 512,
+            "width": 512,
+            "depth": 1,
+            "type": "pix2pixHD",
+            "license": "BSD License",
+            "dataset_type": "DCE-MRI",
+            "privacy_preservation": "",
+            "tags": [
+               "dce-mri",
+               "postcontrast",
+               "synthesis",
+               "breast",
+               "mri",
+               "treatment",
+               "i2i",
+               "pix2pixHD",
+               "SPIE"
+            ],
+            "year": 2023
+         },
+         "description": {
+            "title": "Pre- to Post-Contrast Breast MRI Synthesis for Enhanced Tumour Segmentation",
+            "provided_date": "11.2023",
+            "trained_date": "2023",
+            "provided_after_epoch": 30,
+            "version": "1.0",
+            "publication": "https://doi.org/10.48550/arXiv.2311.10879",
+            "doi": [
+               "https://doi.org/10.48550/arXiv.2311.10879"
+            ],
+            "inputs": [
+                "input_path: default=input/, help=the path to .png breast DCE-MRI images that are translated from pre-contrast to the first DCE post-contrast sequence. ",
+                "image_size: default=[512, 512], help=list with image height and width. ",
+                "gpu_id: default=0, help=the gpu to run the model on."
+            ],
+            "comment": "Pix2Pix model for DCE-MRI slice generation from pre-contrast image input (Trained on Duke Breast MRI Dataset). A pix2pixHD mmodel that generates DCE-MRI axial slices based on checkpoint after 30 training epochs. \nThe pixel dimension of the generated images is 512x512. Several generated 2d slices can be merged together to create a 3D MRI volume with tumour tissue highlighted by synthetic contrast. \nThe uploaded ZIP file contains the files 30_net_G.pth (model weights), __init__.py (image generation method and utils), a requirements.txt, and further code below the /src folder for handling of model, data, and training utils. Sample input images are provided as an example for image generation."
+         }
+  }
 }
diff --git a/docs/source/_static/samples/00023.png b/docs/source/_static/samples/00023.png
diff --git a/docs/source/_static/samples/c-dcgan/model8_samples.png b/docs/source/_static/samples/c-dcgan/model8_samples.png
diff --git a/docs/source/_static/samples/c-dcgan/sample_benign_synthetic-1.png b/docs/source/_static/samples/c-dcgan/sample_benign_synthetic-1.png
diff --git a/docs/source/_static/samples/c-dcgan/sample_benign_synthetic-2.png b/docs/source/_static/samples/c-dcgan/sample_benign_synthetic-2.png
diff --git a/docs/source/_static/samples/c-dcgan/sample_benign_synthetic-3.png b/docs/source/_static/samples/c-dcgan/sample_benign_synthetic-3.png
diff --git a/docs/source/_static/samples/c-dcgan/sample_malignant_syn-1.png b/docs/source/_static/samples/c-dcgan/sample_malignant_syn-1.png
diff --git a/docs/source/_static/samples/c-dcgan/sample_malignant_syn-2.png b/docs/source/_static/samples/c-dcgan/sample_malignant_syn-2.png
diff --git a/docs/source/_static/samples/c-dcgan/sample_malignant_syn-3.png b/docs/source/_static/samples/c-dcgan/sample_malignant_syn-3.png
diff --git a/docs/source/model_documentation.md b/docs/source/model_documentation.md
@@ -235,6 +235,7 @@ PGGAN Model for Patch Generation of Polyps with Corresponding Segmentation Masks
 
 ```python
 # create samples with this model
+from medigan import Generators
 Generators().generate(
     model_id="00009_PGGAN_POLYP_PATCHES_W_MASKS",
     gpu_id=None, 
@@ -638,4 +639,35 @@ inputs= [
     "view: default=\"la\" help=",
     "subcat: default=\"2ch\", help="
 ]
+```
+
+# 00023_PIX2PIXHD_BREAST_DCEMRI
+
+Pix2Pix model for DCE-MRI slice generation from pre-contrast image input (Trained on Duke Breast MRI Dataset) \
+<sub> **Note:** A pix2pixHD mmodel that generates DCE-MRI axial slices based on checkpoint after 30 training epochs. 
+The pixel dimension of the generated images is 512x512. Several generated 2d slices can be merged together to create a 3D MRI volume with tumour tissue highlighted by synthetic contrast. 
+The uploaded ZIP file contains the files 30_net_G.pth (model weights), __init__.py (image generation method and utils), a requirements.txt, and further code below the /src folder for handling of model, data, and training utils. Sample input images are provided as an example for image generation. </sub>
+
+
+| Output type                    | Modality | Model type | Output size |  Base dataset   |           Output examples            |    `model_id`      |  Hosted on   |  Reference  |
+|--------------------------------|:--------:|:----------:|:-----------:|:------------:|:------------------------------------:|:------:|:------:|:------:|
+| DCE-MRI sequence 1 axial slice | DCE-MRI  | pix2pixHD  |   512x512   |     [Duke Dataset](https://sites.duke.edu/mazurowski/resources/breast-cancer-mri-dataset/)      | ![sample](_static/samples/00023.png) |  `00023_PIX2PIXHD_BREAST_DCEMRI`  | [Zenodo (10210944)](https://zenodo.org/doi/10.5281/zenodo.10210944)  | [Osuala et al (2023)](https://doi.org/10.48550/arXiv.2311.10879)  | 
+
+
+```python
+# create samples with this model
+from medigan import Generators
+Generators().generate(
+    model_id="00023_PIX2PIXHD_BREAST_DCEMRI",
+    input_path= "input/",
+    image_size=[512, 512],
+    gpu_id=0,
+)
+
+# model specific parameters
+inputs = [
+    "input_path: default=input/, help=the path to .png breast DCE-MRI images that are translated from pre-contrast to the first DCE post-contrast sequence. ",
+    "image_size: default=[512, 512], help=list with image height and width. ",
+    "gpu_id: default=0, help=the gpu to run the model on.",
+] 
 ```
diff --git a/src/medigan/contribute_model/model_contributor.py b/src/medigan/contribute_model/model_contributor.py
@@ -253,6 +253,9 @@ def push_to_zenodo(
             self.zenodo_model_uploader = ZenodoModelUploader(
                 model_id=self.model_id, access_token=access_token
             )
+        # Update in case previous access token gave an error
+        self.zenodo_model_uploader.access_token = access_token
+
         return self.zenodo_model_uploader.push(
             metadata=self.metadata,
             package_path=self.package_path,
@@ -301,6 +304,8 @@ def push_to_github(
             self.github_model_uploader = GithubModelUploader(
                 model_id=self.model_id, access_token=access_token
             )
+        # Update in case previous access token gave an error
+        self.github_model_uploader.access_token = access_token
 
         return self.github_model_uploader.push(
             metadata=self.metadata,

diff --git a/src/medigan/contribute_model/zenodo_model_uploader.py b/src/medigan/contribute_model/zenodo_model_uploader.py
@@ -203,9 +203,9 @@ def upload(self, file_path: str, filename: str, bucket_url: str) -> dict:
                 params=self.params,
             )
 
-        if not r.status_code == 200:
+        if not r.status_code == 200 and not r.status_code == 201:
             raise Exception(
-                f"{self.model_id}: Error ({r.status_code}!=200) during Zenodo ('{bucket_url}') upload (step 2: uploading model as zip file): {r.json()}"
+                f"{self.model_id}: Error ({r.status_code}!= any of (200, 201) ) during Zenodo ('{bucket_url}') upload (step 2: uploading model as zip file): {r.json()}"
             )
         return r
 
@@ -231,9 +231,9 @@ def upload_descriptive_data(self, deposition_id: str, data: dict) -> dict:
             data=json.dumps(data),
             headers=ZENODO_HEADERS,
         )
-        if not r.status_code == 200:
+        if not r.status_code == 200 and not r.status_code == 201:
             raise Exception(
-                f"{self.model_id}: Error ({r.status_code}!=200) during Zenodo ('{deposition_url}') upload (step 3: updating metadata): {r.json()}"
+                f"{self.model_id}: Error ({r.status_code}!= any of (200, 201) ) during Zenodo ('{deposition_url}') upload (step 3: updating metadata): {r.json()}"
             )
         return r
 

diff --git a/src/medigan/execute_model/model_executor.py b/src/medigan/execute_model/model_executor.py
@@ -276,7 +276,10 @@ def _import_package_as_lib(self):
                 self.serialised_model_file_path = f"{MODEL_FOLDER}/{self.model_id}/{self.model_name}{self.model_extension}"
             except Exception as e:
                 logging.error(
-                    f"{self.model_id}: Error while importing {self.package_name} from /{self.model_id}: {e}"
+                    f"{self.model_id}: Error occurred while trying to import "
+                    f"'{MODEL_FOLDER}.{self.model_id}.{self.package_name}'."
+                    f"Fallback import of '{MODEL_FOLDER}.{self.model_id}' also failed. "
+                    f"Please make sure the module '{MODEL_FOLDER}' is not imported from elsewhere in your syspath: {e}"
                 )
                 raise e
 

diff --git a/tests/model_integration_test_manual.py b/tests/model_integration_test_manual.py
@@ -6,8 +6,8 @@
 import logging
 
 MODEL_ID = "YOUR_MODEL_ID_HERE"
-MODEL_ID = "00002_DCGAN_MMG_MASS_ROI"  # "00007_BEZIERCURVE_TUMOUR_MASK"
-NUM_SAMPLES = 10
+MODEL_ID = 23  # "00023_PIX2PIXHD_BREAST_DCEMRI" #"00002_DCGAN_MMG_MASS_ROI"  # "00007_BEZIERCURVE_TUMOUR_MASK"
+NUM_SAMPLES = 2
 OUTPUT_PATH = f"output/{MODEL_ID}/"
 try:
     from src.medigan.generators import Generators
@@ -21,11 +21,30 @@
     model_id=MODEL_ID,
     num_samples=NUM_SAMPLES,
     output_path=OUTPUT_PATH,
+    input_path="input/",
+    gpu_id=0,
+    image_size=448,
+    install_dependencies=True,
 )
 
-generators.get_as_torch_dataloader(
+data_loader = generators.get_as_torch_dataloader(
     model_id=MODEL_ID,
     num_samples=NUM_SAMPLES,
     output_path=OUTPUT_PATH,
+    input_path="input/",
+    gpu_id=0,
+    image_size=448,
     # prefetch_factor=2, # debugging with torch v2.0.0: This will raise an error for torch DataLoader if num_workers == None at the same time.
 )
+
+print(f"len(data_loader): {len(data_loader)}")
+
+if len(data_loader) != NUM_SAMPLES:
+    logging.warning(
+        f"{MODEL_ID}: The number of samples in the dataloader (={len(data_loader)}) is not equal the number of samples requested (={NUM_SAMPLES})."
+    )
+
+#### Get the object at index 0 from the dataloader
+data_dict = next(iter(data_loader))
+
+print(f"data_dict: {data_dict}")
diff --git a/tests/test_model_executor.py b/tests/test_model_executor.py
@@ -88,6 +88,15 @@
         3,
     ),
     ("00022_WGAN_CARDIAC_AGING", {}, 3),
+    (
+        "00023_PIX2PIXHD_BREAST_DCEMRI",
+        {
+            "input_path": "input",
+            "gpu_id": 0,
+            "image_size": 448,
+        },
+        3,
+    ),
 ]
 
 
@@ -252,7 +261,14 @@ def test_get_dataloader_method(self, model_id):
         data_loader = self.generators.get_as_torch_dataloader(
             model_id=model_id, num_samples=self.num_samples
         )
-        self.logger.debug(f"len(data_loader): {len(data_loader)}")
+        self.logger.debug(f"{model_id}: len(data_loader): {len(data_loader)}")
+
+        if len(data_loader) != self.num_samples:
+            logging.warning(
+                f"{model_id}: The number of samples in the dataloader (={len(data_loader)}) is not equal the number of samples requested (={self.num_samples}). "
+                f"Hint: Revise if the model's internal generate() function returned tuples as required in get_as_torch_dataloader()."
+            )
+
         #### Get the object at index 0 from the dataloader
         data_dict = next(iter(data_loader))
 
@@ -296,7 +312,7 @@ def _check_if_samples_were_generated(
     ):
         # check if the number of generated samples of model_id_1 is as expected.
         file_list = glob.glob(self.test_output_path + "/*")
-        self.logger.debug(f"{len(file_list)} == {self.num_samples} ?")
+        self.logger.debug(f"{model_id}: {len(file_list)} == {self.num_samples} ?")
         if num_samples is None:
             num_samples = self.num_samples