Merge pull request #138 from pyplati/nnunet-segmentation

Open Atlas

docs/cardiac.rst

@@ -0,0 +1,2 @@
+
+.. mdinclude:: ../platipy/imaging/projects/cardiac/README.md

docs/index.rst

@@ -12,6 +12,13 @@ PlatiPy documentation
     getting_started
     contributing
 
+.. toctree::
+    :caption: Projects
+    :maxdepth: 2
+    :hidden:
+
+    cardiac
+
 .. toctree::
     :caption: Examples
     :maxdepth: 2

examples/cardiac_segmentation.ipynb

@@ -4,7 +4,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "# Cardiac Multi Atlas-Based Segmentation Example\n",
+    "# Cardiac Sub-Structure Segmentation Example\n",
     "\n",
     "## Import Modules"
    ]
@@ -21,12 +21,14 @@
     "    !pip install git+https://github.com/pyplati/platipy.git\n",
     "    import platipy\n",
     "\n",
+    "from matplotlib import pyplot as plt\n",
+    "\n",
     "import SimpleITK as sitk\n",
     "\n",
     "%matplotlib inline\n",
     "\n",
     "from platipy.imaging.tests.data import get_lung_nifti\n",
-    "from platipy.imaging.projects.cardiac.run import run_cardiac_segmentation\n",
+    "from platipy.imaging.projects.cardiac.run import run_hybrid_segmentation\n",
     "from platipy.imaging import ImageVisualiser\n",
     "from platipy.imaging.label.utils import get_com"
    ]
@@ -52,101 +54,6 @@
     "data_path = get_lung_nifti()"
    ]
   },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## Configure Settings\n",
-    "\n",
-    "We define the settings used for the end-to-end segmentation\n",
-    "Check out the guide in \"PlatiPy-GettingStarted.pdf\" for a description of these settings"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "settings = {\n",
-    "    \"output_format\": \"Auto_{0}.nii.gz\",\n",
-    "    \"atlas_settings\": {\n",
-    "        \"atlas_id_list\": [\"101\",\"102\",\"103\",\"104\"],\n",
-    "        \"atlas_structure_list\": [\"HEART\",\"ESOPHAGUS\",\"LUNG_L\",\"LUNG_R\",\"SPINALCORD\"],\n",
-    "        \"atlas_path\": str(data_path),\n",
-    "        \"atlas_image_format\": \"LCTSC-Test-S1-{0}/IMAGES/LCTSC_TEST_S1_{0}_0_CT_0.nii.gz\",\n",
-    "        \"atlas_label_format\": \"LCTSC-Test-S1-{0}/STRUCTURES/LCTSC_TEST_S1_{0}_0_RTSTRUCT_{1}.nii.gz\",\n",
-    "        \"crop_atlas_to_structures\": False,\n",
-    "        \"crop_atlas_expansion_mm\": (10, 10, 10),\n",
-    "        \"guide_structure_name\": \"HEART\",\n",
-    "    },\n",
-    "    \"auto_crop_target_image_settings\": {\n",
-    "        \"expansion_mm\": [2, 2, 2],\n",
-    "    },\n",
-    "    \"linear_registration_settings\": {\n",
-    "        \"reg_method\": \"similarity\",\n",
-    "        \"shrink_factors\": [16, 8, 4],\n",
-    "        \"smooth_sigmas\": [0, 0, 0],\n",
-    "        \"sampling_rate\": 0.75,\n",
-    "        \"default_value\": -1024,\n",
-    "        \"number_of_iterations\": 50,\n",
-    "        \"metric\": \"mean_squares\",\n",
-    "        \"optimiser\": \"gradient_descent_line_search\",\n",
-    "        \"verbose\": False,\n",
-    "    },\n",
-    "    \"deformable_registration_settings\": {\n",
-    "        \"isotropic_resample\": True,\n",
-    "        \"resolution_staging\": [\n",
-    "            16,\n",
-    "            8,\n",
-    "            4,\n",
-    "        ],  # specify voxel size (mm) since isotropic_resample is set\n",
-    "        \"iteration_staging\": [15, 15, 15],\n",
-    "        \"smoothing_sigmas\": [4, 2, 1],\n",
-    "        \"ncores\": 8,\n",
-    "        \"default_value\": -1000,\n",
-    "        \"verbose\": False,\n",
-    "    },\n",
-    "    \"structure_guided_registration_settings\": {\n",
-    "        \"isotropic_resample\": True,\n",
-    "        \"resolution_staging\": [\n",
-    "            16,\n",
-    "            8,\n",
-    "            2,\n",
-    "        ],  # specify voxel size (mm) since isotropic_resample is set\n",
-    "        \"iteration_staging\": [25, 25, 25],\n",
-    "        \"smoothing_sigmas\": [0, 0, 0],\n",
-    "        \"ncores\": 8,\n",
-    "        \"default_value\": 0,\n",
-    "        \"verbose\": False,\n",
-    "    },\n",
-    "    \"iar_settings\": {\n",
-    "        \"reference_structure\": \"HEART\",\n",
-    "        \"smooth_distance_maps\": True,\n",
-    "        \"smooth_sigma\": 1,\n",
-    "        \"z_score_statistic\": \"mad\",\n",
-    "        \"outlier_method\": \"iqr\",\n",
-    "        \"outlier_factor\": 1.5,\n",
-    "        \"min_best_atlases\": 5,\n",
-    "        \"project_on_sphere\": False,\n",
-    "    },\n",
-    "    \"label_fusion_settings\": {\n",
-    "        \"vote_type\": \"unweighted\",\n",
-    "        \"vote_params\": {},  # No parameters needed for majority voting\n",
-    "        \"optimal_threshold\": {\"HEART\": 0.5, \"ESOPHAGUS\": 0.5, \"LUNG_L\": 0.5, \"LUNG_R\": 0.5, \"SPINALCORD\": 0.5},\n",
-    "    },\n",
-    "    \"vessel_spline_settings\": {\n",
-    "        \"vessel_name_list\": [\"SPINALCORD\"],\n",
-    "        \"vessel_radius_mm_dict\": {\"SPINALCORD\": 6},\n",
-    "        \"scan_direction_dict\": {\"SPINALCORD\": \"z\"},\n",
-    "        \"stop_condition_type_dict\": {\"SPINALCORD\": \"count\"},\n",
-    "        \"stop_condition_value_dict\": {\"SPINALCORD\": 1},\n",
-    "    },\n",
-    "    \"return_as_cropped\": False,\n",
-    "    'returnAsCropped': False\n",
-    "}"
-   ]
-  },
   {
    "cell_type": "markdown",
    "metadata": {},
@@ -183,7 +90,7 @@
    },
    "outputs": [],
    "source": [
-    "auto_structures = run_cardiac_segmentation(test_image, settings=settings)"
+    "auto_structures, _ = run_hybrid_segmentation(test_image)"
    ]
   },
   {
@@ -200,12 +107,11 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "output_name = settings[\"output_format\"]\n",
-    "output_directory = test_pat_path.joinpath(\"SEGMENTATIONS\")\n",
+    "output_directory = test_pat_path.joinpath(\"substructures\")\n",
     "output_directory.mkdir(exist_ok=True)\n",
     "\n",
     "for struct_name in list(auto_structures.keys()):\n",
-    "    sitk.WriteImage(auto_structures[struct_name], str(output_directory.joinpath(output_name.format(struct_name))))\n",
+    "    sitk.WriteImage(auto_structures[struct_name], str(output_directory.joinpath(f\"{struct_name}.nii.gz\")))\n",
     "\n",
     "print(f\"Segmentations saved to: {output_directory}\")\n",
     "\n"
@@ -226,11 +132,19 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "vis = ImageVisualiser(test_image, cut=get_com(auto_structures[\"HEART\"]))\n",
-    "\n",
+    "vis = ImageVisualiser(test_image, cut=get_com(auto_structures[\"Heart\"]))\n",
     "vis.add_contour({struct: auto_structures[struct] for struct in auto_structures.keys()})\n",
-    "\n",
-    "fig = vis.show()"
+    "fig = vis.show()\n",
+    "plt.savefig(output_directory.joinpath(f\"snapshot.png\"))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "fig"
    ]
   },
   {
@@ -257,7 +171,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.6.9"
+   "version": "3.8.10"
   }
  },
  "nbformat": 4,

platipy/backend/models.py

@@ -117,6 +117,7 @@ class Dataset(db.Model):
     output_data_objects = relationship(
         "DataObject",
         primaryjoin="and_(DataObject.dataset_id == Dataset.id, DataObject.is_input == False)",
+        overlaps="input_data_objects"
     )
 
     # The Dicom location from which to retrieve data
@@ -143,7 +144,7 @@ class DataObject(db.Model):
 
     id = db.Column(db.Integer, primary_key=True)
     dataset_id = db.Column(db.Integer, db.ForeignKey("Dataset.id"), nullable=False)
-    dataset = relationship("Dataset")
+    dataset = relationship("Dataset", overlaps="input_data_objects,output_data_objects")
     is_input = db.Column(db.Boolean, default=False)
 
     path = db.Column(db.String(256))
@@ -163,7 +164,7 @@ class DataObject(db.Model):
     parent_id = db.Column(db.Integer, db.ForeignKey("DataObject.id"), index=True)
     parent = relationship("DataObject", remote_side=[id])
     children = relationship(
-        "DataObject", primaryjoin="and_(DataObject.parent_id == DataObject.id)"
+        "DataObject", primaryjoin="and_(DataObject.parent_id == DataObject.id)", overlaps="parent"
     )
 
     def __repr__(self):

platipy/backend/views.py

@@ -98,7 +98,8 @@ def fetch_status():
 
     status_context["applications"] = []
     for ak in APIKey.query.all():
-        status_context["applications"].append({"name": ak.name, "key": ak.key})
+        # Hide Key from the dashboard
+        status_context["applications"].append({"name": ak.name, "key": "XXX"})
 
     return jsonify(status_context)
 

platipy/cli/segmentation.py

@@ -28,14 +28,14 @@
     BRONCHUS_SETTINGS_DEFAULTS,
 )
 from platipy.imaging.projects.cardiac.run import (
-    run_cardiac_segmentation,
-    CARDIAC_SETTINGS_DEFAULTS,
+    run_hybrid_segmentation,
+    HYBRID_SETTINGS_DEFAULTS,
 )
 
 segmentation_algorithms = {
     "cardiac": {
-        "algorithm": run_cardiac_segmentation,
-        "default_settings": CARDIAC_SETTINGS_DEFAULTS,
+        "algorithm": run_hybrid_segmentation,
+        "default_settings": HYBRID_SETTINGS_DEFAULTS
     },
     "bronchus": {
         "algorithm": run_bronchus_segmentation,
@@ -105,6 +105,11 @@ def click_command(algorithm, input_path, config, default, output):
     # Run the algorithm
     results = segmentation_algorithms[algorithm]["algorithm"](image, algorithm_config)
 
+    # If results is a tuple, the algorithm is returning other items as well as the semgmentation.
+    # We are only interested in the segmentation here (the first element of the tuple)
+    if isinstance(results, tuple):
+        results = results[0]
+
     # Save the output to the output directory
     if not output:
         output = "."