Small fixes and more unit tests (#3)

fix small bugs, change urls to pretrained networks, add more unit tests and support newest pydicom
eeulig · Oct 10, 2024 · ddea399 · ddea399
1 parent 36821f6
commit ddea399
Show file tree

Hide file tree

Showing 16 changed files with 489 additions and 162 deletions.
diff --git a/.gitignore b/.gitignore
@@ -8,4 +8,5 @@ wandb/
 site/
 *.DS_Store
 __pycache__/
-*.psd
+*.psd
+.coverage
diff --git a/README.md b/README.md
@@ -9,6 +9,7 @@
 [![PyPI - Version](https://img.shields.io/pypi/v/ldct-benchmark?color=blue&cacheSeconds=!%5BPyPI%20-%20Version%5D(https%3A%2F%2Fimg.shields.io%2Fpypi%2Fv%2Fldct-benchmark))](https://pypi.org/project/ldct-benchmark/)
 ![License](https://img.shields.io/badge/MIT-blue?label=License)
 [![arXiv](https://img.shields.io/badge/2401.04661-red?label=arXiv)](https://arxiv.org/abs/2401.04661)
+[![DOI](https://img.shields.io/badge/10.1002%2Fmp.17379-red?label=MedPhys)](https://doi.org/10.1002/mp.17379)
 
 [GitHub](https://github.com/eeulig/ldct-benchmark) | [Documentation](https://eeulig.github.io/ldct-benchmark/)
 
@@ -49,15 +50,30 @@ We welcome contributions of novel denoising algorithms. For details on how to do
 
 ## Reference
 If you find this project useful for your work, please cite our [arXiv preprint](https://arxiv.org/abs/2401.04661):
-> Elias Eulig, Björn Ommer, & Marc Kachelrieß (2024). Benchmarking Deep Learning-Based Low Dose CT Image Denoising Algorithms. arXiv, 2401.04661.
+> Elias Eulig, Björn Ommer, & Marc Kachelrieß (2024). Benchmarking Deep Learning-Based Low-Dose CT Image Denoising Algorithms. arXiv, 2401.04661.
 
 ```bibtex
-@article{eulig2024ldctbench,
-    title={Benchmarking Deep Learning-Based Low Dose CT Image Denoising Algorithms}, 
-    author={Elias Eulig and Björn Ommer and Marc Kachelrieß},
-    year={2024},
-    eprint={2401.04661},
-    archivePrefix={arXiv},
-    primaryClass={physics.med-ph}
+@article{ldctbench-arxiv,
+  title = {Benchmarking Deep Learning-Based Low-Dose CT Image Denoising Algorithms}, 
+  author = {Elias Eulig and Björn Ommer and Marc Kachelrieß},
+  year = {2024},
+  eprint = {2401.04661},
+  archivePrefix = {arXiv},
+  primaryClass = {physics.med-ph}
+}
+```
+
+or [Medical Physics paper](https://doi.org/10.1002/mp.17379):
+> Elias Eulig, Björn Ommer, & Marc Kachelrieß (2024). Benchmarking Deep Learning-Based Low-Dose CT Image Denoising Algorithms. Medical Physics.
+
+```bibtex
+@article{ldctbench-medphys,
+  title = {Benchmarking deep learning-based low-dose CT image denoising algorithms},
+  author = {Eulig, Elias and Ommer, Björn and Kachelrieß, Marc},
+  journal = {Medical Physics},
+  year = {2024},
+  doi = {https://doi.org/10.1002/mp.17379},
+  url = {https://aapm.onlinelibrary.wiley.com/doi/abs/10.1002/mp.17379},
+  eprint = {https://aapm.onlinelibrary.wiley.com/doi/pdf/10.1002/mp.17379},
 }
 ```
diff --git a/README_PYPI.md b/README_PYPI.md
@@ -36,15 +36,30 @@ We welcome contributions of novel denoising algorithms. For details on how to do
 
 ## Reference
 If you find this project useful for your work, please cite our [arXiv preprint](https://arxiv.org/abs/2401.04661):
-> Elias Eulig, Björn Ommer, & Marc Kachelrieß (2024). Benchmarking Deep Learning-Based Low Dose CT Image Denoising Algorithms. arXiv, 2401.04661.
+> Elias Eulig, Björn Ommer, & Marc Kachelrieß (2024). Benchmarking Deep Learning-Based Low-Dose CT Image Denoising Algorithms. arXiv, 2401.04661.
 
 ```bibtex
-@article{eulig2024ldctbench,
-    title={Benchmarking Deep Learning-Based Low Dose CT Image Denoising Algorithms}, 
-    author={Elias Eulig and Björn Ommer and Marc Kachelrieß},
-    year={2024},
-    eprint={2401.04661},
-    archivePrefix={arXiv},
-    primaryClass={physics.med-ph}
+@article{ldctbench-arxiv,
+  title = {Benchmarking Deep Learning-Based Low-Dose CT Image Denoising Algorithms}, 
+  author = {Elias Eulig and Björn Ommer and Marc Kachelrieß},
+  year = {2024},
+  eprint = {2401.04661},
+  archivePrefix = {arXiv},
+  primaryClass = {physics.med-ph}
 }
-```
+```
+
+or [Medical Physics paper](https://doi.org/10.1002/mp.17379):
+> Elias Eulig, Björn Ommer, & Marc Kachelrieß (2024). Benchmarking Deep Learning-Based Low-Dose CT Image Denoising Algorithms. Medical Physics.
+
+```bibtex
+@article{ldctbench-medphys,
+  title = {Benchmarking deep learning-based low-dose CT image denoising algorithms},
+  author = {Eulig, Elias and Ommer, Björn and Kachelrieß, Marc},
+  journal = {Medical Physics},
+  year = {2024},
+  doi = {https://doi.org/10.1002/mp.17379},
+  url = {https://aapm.onlinelibrary.wiley.com/doi/abs/10.1002/mp.17379},
+  eprint = {https://aapm.onlinelibrary.wiley.com/doi/pdf/10.1002/mp.17379},
+}
+```
diff --git a/docs/index.md b/docs/index.md
@@ -38,16 +38,31 @@ Therefore, the **aim** of this project is to
 We welcome contributions of novel denoising algorithms. For details on how to do so, please check out our [contributing guide](https://github.com/eeulig/ldct-benchmark/blob/main/CONTRIBUTING.md){:target="_blank"} or reach out to [me](mailto:elias.eulig@dkfz.de).
 
 ## Reference
-If you find this project useful for your work, please cite our [arXiv preprint](https://arxiv.org/abs/2401.04661){:target="_blank"}:
-> Elias Eulig, Björn Ommer, & Marc Kachelrieß (2024). Benchmarking Deep Learning-Based Low Dose CT Image Denoising Algorithms. arXiv, 2401.04661.
+If you find this project useful for your work, please cite our [arXiv preprint](https://arxiv.org/abs/2401.04661):
+> Elias Eulig, Björn Ommer, & Marc Kachelrieß (2024). Benchmarking Deep Learning-Based Low-Dose CT Image Denoising Algorithms. arXiv, 2401.04661.
 
 ```bibtex
-@article{eulig2024ldctbench,
-    title={Benchmarking Deep Learning-Based Low Dose CT Image Denoising Algorithms}, 
-    author={Elias Eulig and Björn Ommer and Marc Kachelrieß},
-    year={2024},
-    eprint={2401.04661},
-    archivePrefix={arXiv},
-    primaryClass={physics.med-ph}
+@article{ldctbench-arxiv,
+  title = {Benchmarking Deep Learning-Based Low-Dose CT Image Denoising Algorithms}, 
+  author = {Elias Eulig and Björn Ommer and Marc Kachelrieß},
+  year = {2024},
+  eprint = {2401.04661},
+  archivePrefix = {arXiv},
+  primaryClass = {physics.med-ph}
+}
+```
+
+or [Medical Physics paper](https://doi.org/10.1002/mp.17379):
+> Elias Eulig, Björn Ommer, & Marc Kachelrieß (2024). Benchmarking Deep Learning-Based Low-Dose CT Image Denoising Algorithms. Medical Physics.
+
+```bibtex
+@article{ldctbench-medphys,
+  title = {Benchmarking deep learning-based low-dose CT image denoising algorithms},
+  author = {Eulig, Elias and Ommer, Björn and Kachelrieß, Marc},
+  journal = {Medical Physics},
+  year = {2024},
+  doi = {https://doi.org/10.1002/mp.17379},
+  url = {https://aapm.onlinelibrary.wiley.com/doi/abs/10.1002/mp.17379},
+  eprint = {https://aapm.onlinelibrary.wiley.com/doi/pdf/10.1002/mp.17379},
 }
 ```
diff --git a/ldctbench/data/LDCTMayo.py b/ldctbench/data/LDCTMayo.py
@@ -213,10 +213,10 @@ def __getitem__(self, idx: int) -> Dict[str, torch.Tensor]:
         # Load image
         sample = self.samples[idx]
         f_name = self._idx2filename(sample["slice"], sample["n_slices"])
-        x = pydicom.read_file(
+        x = pydicom.filereader.dcmread(
             os.path.join(self.path, sample["input"][2:], f_name)
         ).pixel_array.astype("float32")
-        y = pydicom.read_file(
+        y = pydicom.filereader.dcmread(
             os.path.join(self.path, sample["target"][2:], f_name)
         ).pixel_array.astype("float32")
 
@@ -242,10 +242,10 @@ def __init__(self, datafolder, data_norm):
             )
             for s in range(patient_dict["n_slices"]):
                 f_name = self._idx2filename(s + 1, patient_dict["n_slices"])
-                x = pydicom.read_file(
+                x = pydicom.filereader.dcmread(
                     os.path.join(datafolder, patient_dict["input"][2:], f_name)
                 ).pixel_array.astype("float32")
-                y = pydicom.read_file(
+                y = pydicom.filereader.dcmread(
                     os.path.join(datafolder, patient_dict["target"][2:], f_name)
                 ).pixel_array.astype("float32")
                 self.samples[-1]["x"].append(x)

diff --git a/ldctbench/data/prepare_dataset.py b/ldctbench/data/prepare_dataset.py
@@ -40,7 +40,9 @@
 
 
 def load_dicom(path):
-    slices = [pydicom.read_file(os.path.join(path, s)) for s in os.listdir(path)]
+    slices = [
+        pydicom.filereader.dcmread(os.path.join(path, s)) for s in os.listdir(path)
+    ]
     slices.sort(key=lambda x: float(x.ImagePositionPatient[2]))
     image = np.stack([s.pixel_array for s in slices])
     return image

diff --git a/ldctbench/hub/checkpoints.json b/ldctbench/hub/checkpoints.json
@@ -1,6 +1,6 @@
 {
     "resnet": {
-        "url": "https://my.hidrive.com/api/sharelink/download?id=p3CIkV4O",
+        "url": "https://downloads.eeulig.com/ldct-benchmark/resnet.pt",
         "checksum": "4fe7137fe5abe7c381d746914528eaf23f061204580d8f027faf4d5eaaca6714",
         "model_name": "Model",
         "args": {},
@@ -11,55 +11,55 @@
         "normalization": "meanstd"
     },
     "bilateral": {
-        "url": "https://my.hidrive.com/api/sharelink/download?id=YqiokoAv",
+        "url": "https://downloads.eeulig.com/ldct-benchmark/bilateral.pt",
         "checksum": "3e0722758e31c92fd11cc8bb9f5d2e93714a885bbe21183228fbe8946e79d5ff",
         "model_name": "Model",
         "args": {"cuda": true},
         "kwargs": {},
         "normalization": "meanstd"
     },
     "cnn10": {
-        "url": "https://my.hidrive.com/api/sharelink/download?id=LGCIEJ1u",
+        "url": "https://downloads.eeulig.com/ldct-benchmark/cnn10.pt",
         "checksum": "ee96e494c8eda02e946464df29cf43c14163e63bc4b1df6d8907ee33a1a18c3e",
         "model_name": "Model",
         "args": {},
         "kwargs": {},
         "normalization": "meanstd"
     },
     "dugan": {
-        "url": "https://my.hidrive.com/api/sharelink/download?id=ljCok94w",
+        "url": "https://downloads.eeulig.com/ldct-benchmark/dugan.pt",
         "checksum": "c6faad8376994fe22eadc0a723f81705fda6f435f40ef0f999666e871216ac5e",
         "model_name": "Model",
         "args": {},
         "kwargs": {"out_ch": 96},
         "normalization": "meanstd"
     },
     "qae": {
-        "url": "https://my.hidrive.com/api/sharelink/download?id=5LiIE1Em",
+        "url": "https://downloads.eeulig.com/ldct-benchmark/qae.pt",
         "checksum": "ca6ce543b476d4c3673acd2f3d1284d47d98483c41a61cc912f5214c80921363",
         "model_name": "Model",
         "args": {},
         "kwargs": {},
         "normalization": "meanstd"
     },
     "redcnn": {
-        "url": "https://my.hidrive.com/api/sharelink/download?id=2GCokzaQ",
+        "url": "https://downloads.eeulig.com/ldct-benchmark/redcnn.pt",
         "checksum": "058ae58264a00116e5620ed88047e687c7e93c8b732bfd6d7796e76bae0c74f3",
         "model_name": "Model",
         "args": {},
         "kwargs": {"out_ch": 96},
         "normalization": "meanstd"
     },
     "transct": {
-        "url": "https://my.hidrive.com/api/sharelink/download?id=6aiIEj6Q",
+        "url": "https://downloads.eeulig.com/ldct-benchmark/transct.pt",
         "checksum": "37b5af1cb4fdfaa0c1d67d5992a5e9c6da5b6afda59bf5274d1fe5db163f959e",
         "model_name": "Model",
         "args": {},
         "kwargs": {},
         "normalization": "meanstd"
     },
     "wganvgg": {
-        "url": "https://my.hidrive.com/api/sharelink/download?id=GHCoEb4H",
+        "url": "https://downloads.eeulig.com/ldct-benchmark/wganvgg.pt",
         "checksum": "1ad87d16aefea99a9eeb474c33127971131cc21a07e129bf464c764cf9ca6003",
         "model_name": "Model",
         "args": {},

diff --git a/ldctbench/hub/utils.py b/ldctbench/hub/utils.py
@@ -10,6 +10,11 @@
 from ldctbench.hub import Methods, load_model
 from ldctbench.utils.test_utils import denormalize, normalize
 
+SUPPORTED_TRANSFER_SYNTAX_UIDS = [
+    pydicom.uid.ExplicitVRLittleEndian,
+    pydicom.uid.RLELossless,
+]
+
 
 @torch.no_grad()
 def denoise_numpy(
@@ -162,9 +167,12 @@ def denoise_dicom(
     files = None
     if os.path.isdir(dicom_path):
         files = [
-            os.path.join(dicom_path, file)
-            for file in os.listdir(dicom_path)
-            if pydicom.misc.is_dicom(os.path.join(dicom_path, file))
+            os.path.join(dicom_path, item)
+            for item in os.listdir(dicom_path)
+            if (
+                not os.path.isdir(os.path.join(dicom_path, item))
+                and pydicom.misc.is_dicom(os.path.join(dicom_path, item))
+            )
         ]
     elif pydicom.misc.is_dicom(dicom_path):
         files = [dicom_path]
@@ -173,25 +181,24 @@ def denoise_dicom(
 
     # Iterate over filepaths
     for file in tqdm(files, desc="Denoise DICOMs", disable=disable_progress):
-        ds = pydicom.read_file(file)
+        ds = pydicom.filereader.dcmread(file)
 
-        # Training data had RescaleIntercept: -1024 HU and RescaleSlope: 1.0. raise a warning if given dicom has different values
+        # Training data had RescaleIntercept: -1024 HU and RescaleSlope: 1.0. Raise a warning if given dicom has different values
         intercept = getattr(ds, "RescaleIntercept", None)
         slope = getattr(ds, "RescaleSlope", None)
-        if not intercept or int(intercept) != -1024:
+        if intercept is None or int(intercept) != -1024:
             warnings.warn(
                 f"Expected DICOM data to have intercept -1024 but got {intercept} instead!"
             )
-        if not intercept or float(slope) != 1.0:
+        if slope is None or float(slope) != 1.0:
             warnings.warn(
                 f"Expected DICOM data to have slope 1.0 but got {slope} instead!"
             )
 
-        # Might be that this function doesn't properly handle DICOM files with compressed pixel data. Raise a warning.
-        if not ds.file_meta.TransferSyntaxUID == pydicom.uid.ExplicitVRLittleEndian:
-            ds.decompress()
-            warnings.warn(
-                f"TransferSyntaxUID (0002, 0010) must be {pydicom.uid.ExplicitVRLittleEndian} ({pydicom.uid.ExplicitVRLittleEndian.name}) but got {ds.file_meta.TransferSyntaxUID} ({ds.file_meta.TransferSyntaxUID.name}) instead! Used pydicoms .decompress()"
+        # Raise error if TransferSyntaxUID is not supported
+        if not ds.file_meta.TransferSyntaxUID in SUPPORTED_TRANSFER_SYNTAX_UIDS:
+            raise ValueError(
+                f"TransferSyntaxUID (0002, 0010) must be one of {', '.join([f'{item} ({item.name})' for item in SUPPORTED_TRANSFER_SYNTAX_UIDS])}) but got {ds.file_meta.TransferSyntaxUID} ({ds.file_meta.TransferSyntaxUID.name}) instead!"
             )
 
         # Denoise image data
@@ -208,8 +215,16 @@ def denoise_dicom(
             out=x_denoised,
         )
 
-        # Overwrite existing PixelData
-        ds.PixelData = x_denoised.astype(x.dtype).tobytes()
+        # Update PixelData, compressed if needed
+        if ds.file_meta.TransferSyntaxUID.is_compressed:
+            ds.compress(
+                ds.file_meta.TransferSyntaxUID,
+                x_denoised.astype(x.dtype),
+                generate_instance_uid=False,
+            )
+            print(f"Compressed again using {ds.file_meta.TransferSyntaxUID.name}")
+        else:
+            ds.PixelData = x_denoised.astype(x.dtype).tobytes()
 
         # Save as a new file
         new_filepath = os.path.join(savedir, os.path.basename(file))

diff --git a/ldctbench/utils/auxiliaries.py b/ldctbench/utils/auxiliaries.py
@@ -26,6 +26,12 @@ def load_json(path: str):
         return d
 
 
+def save_json(content, path: str):
+    """Save python object to json file"""
+    with open(path, "w") as f:
+        json.dump(content, f)
+
+
 def dump_config(args: Namespace, path: str):
     """Save argparse.Namespace to yaml file"""
     with open(os.path.join(path, "args.yaml"), "w") as outfile: