updates

Author: msneubauer

_sources/projects/Project_02_Radiotherapy.ipynb

@@ -53,14 +53,6 @@
     "* https://github.com/ababier/open-kbp"
    ]
   },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "Link to data\n",
-    "* https://github.com/ababier/open-kbp"
-   ]
-  },
   {
    "cell_type": "markdown",
    "metadata": {},
@@ -94,7 +86,7 @@
     "\\end{aligned}\n",
     "$$\n",
     "\n",
-    "What physical phenomena govern this loss of intensity (state at least 3 specific processes)? Which one dominates radiotherapy (6-15 MV energies)? It's easy to conclude that the highest dose of radiation is delivered at the skin, why is this untrue?"
+    "What physical phenomena govern this loss of intensity (state at least 3 specific processes)? Which one dominates radiotherapy (which uses ~6 MeV photons)? It's easy to conclude that the highest dose of radiation is delivered at the skin, why is this untrue?"
    ]
   },
   {
@@ -147,14 +139,24 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "Run the cell below and unzip the contents of `openkbp_patient_data` into a folder. It should be present in Colab's (`/content/..`) or local directory. Note that you will need to adjust `base_path` to lead to this folder.\n",
+    "Run the cell below and unzip the contents of `openkbp_patient_data` into a folder. It should be present in Colab's (`/content/..`) or local directory. Note that you will need to adjust `base_path` to lead to this folder.     \n",
     "\n",
     "Each patient hosts the following three files:  \n",
     "`ct.csv` - 3D grayscale CT scan of patient anatomy composed of 2D slices. Each voxel represents how much X-ray is absorbed by the tissue in Hounsfield Units (HU)  \n",
-    "`PTV63.csv` - Planning Target Volume (PTV) is a binary mask that maps the location of the tumor (tumor is 1, everything else is 0)  \n",
-    "`SpinalCord.csv` - This is a similar binary mask of our Organ at Risk (OAR), representing what we would like our beam to avoid.  \n",
+    "`PTV63.csv` - Planning Target Volume (PTV) is a sparse mask that maps the location of the tumor. We'd like binary mask so that 1 is tumor, 0 is no-tumor.\n",
+    "\n",
+    "The following are sparse masks of Organs at Risk (OAR), representing what we'd like our beams to avoid\n",
+    "- `SpinalCord.csv` - Irradiating the spine can cause a serious condition known as radiation myelopathy.\n",
+    "- `Brainstem.csv` - Radiation causes a vast myriad of devastating effects like RIBN, Ataxia, or even comas.\n",
+    "- `LeftParotid.csv`, `RightParotid.csv` - Irradiating salivary glands can cause xerostomia (dry mouth) or worse\n",
+    "- `Mandible.csv` - The mandible has a limited blood supply, and osteoradionecrosis can occur YEARS after treatment.\n",
+    "\n",
+    "These are flattened volumes with voxel indices. The cell below converts the csv files into binary 3D volumes [128, 128, 128] where each point represents an intensity (for the ct scan) or binary value (for ptv & spinal cord). You will have an optional dictionary of `patient_data` to access for your convenience.\n",
+    "\n",
+    "Your tasks:\n",
+    "1) Add a line to the for loop that stacks `ct`, `ptv`, and `oars` (Organs at Risk) into a single tensor. Note that oars is a dictionary you may need to unpack. What's the final shape of this tensor? Draw an analogy to an RGB image. \n",
     "\n",
-    "These are currently flattened 1D rows. The cell below converts the csv files into 3D volumes [64, 128, 128] where each point represents an intensity (for the ct scan) or binary value (for ptv & spinal cord). Add a line to the for loop that stacks `ct`, `ptv`, and `spine` into a single tensor. What's the final shape of this tensor? Draw an analogy to an RGB image."
+    "2) Visualize a single 2D axial slice (z-axis slice) of a patient's CT scan with the tumor(ptv) overlaid...as a sanity check."
    ]
   },
   {
@@ -163,8 +165,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "!wget https://github.com/florilegium7/Physics-informed-DQN-Radiotherapy/releases/download/v1.0/openkbp_patient_data.zip\n",
-    "!unzip patient_data.zip"
+    "!wget https://github.com/florilegium7/Physics-informed-DQN-Radiotherapy/releases/download/v1/openkbp_patient_data.zip\n",
+    "!unzip openkbp_patient_data.zip -d openkbp_patient_data"
    ]
   },
   {
@@ -179,20 +181,60 @@
     "\n",
     "base_path = 'openkbp_patient_data' #you may adjust this (e.g. /content/openkbp_patient_data)\n",
     "\n",
-    "patient_ids = [\"patient_1\", \"patient_2\", \"patient_3\",\"patient_5\",\"patient_7\",\"patient_9\", \"patient_10\", \"patient_12\" , \"patient_14\", \"patient_16\"]\n",
+    "patient_ids = [\"pt_1\", \"pt_2\", \"pt_9\",\"pt_40\",\"pt_68\",\"pt_70\", \"pt_90\", \"pt_91\" , \"pt_99\", \"pt_187\"]\n",
     "patient_tensors = []\n",
     "\n",
-    "#64 slices of 128 x 128 pixels : (depth, height, width)\n",
+    "def mask_from_sparse(path,shape=(128, 128,128)): #converts voxel indices to 3D binary masks!\n",
+    "    indices = pd.read_csv(path, header=None)[0].dropna().astype(int).values\n",
+    "    mask_flat = np.zeros(np.prod(shape), dtype=np.uint8)\n",
+    "    mask_flat[indices] = 1\n",
+    "    return mask_flat.reshape(shape)\n",
+    "\n",
+    "def sparse_to_ct_volume(path, shape=(128, 128, 128)): #turns the csv into full 3D volume CT scans\n",
+    "    df = pd.read_csv(path, header=None).dropna()\n",
+    "    indices = df[0].astype(int).values\n",
+    "    values = df[1].astype(float).values\n",
+    "    ct_flat = np.zeros(np.prod(shape), dtype=np.float32)\n",
+    "    ct_flat[indices] = values\n",
+    "    return ct_flat.reshape(shape)\n",
+    "\n",
+    "\n",
+    "#128 x 128 x 128 pixels : (depth, height, width)\n",
     "for pt_id in patient_ids:\n",
-    "    ct = pd.read_csv(os.path.join(base_path, pt_id, \"ct.csv\"), header=None).values.reshape((64, 128, 128))\n",
-    "    ptv = pd.read_csv(os.path.join(base_path, pt_id, \"PTV63.csv\"), header=None).values.reshape((64, 128, 128))\n",
-    "    spine = pd.read_csv(os.path.join(base_path, pt_id, \"SpinalCord.csv\"), header=None).values.reshape((64, 128, 128))\n",
     "\n",
-    "    pt_tensor = #Your Code Here \n",
-    "    patient_tensors.append(pt_tensor)\n",
+    "    pt_dir = os.path.join(base_path, pt_id)\n",
+    "\n",
+    "    ct = sparse_to_ct_volume(os.path.join(pt_dir, \"ct.csv\"))\n",
+    "    ptv = mask_from_sparse(os.path.join(pt_dir, \"PTV63.csv\"))\n",
+    "\n",
+    "    oars = {} #dictionary\n",
+    "    organs = [\"SpinalCord\", \"Brainstem\", \"LeftParotid\", \"RightParotid\", \"Mandible\"]\n",
+    "    for organ in organs:\n",
+    "        organ_path = os.path.join(pt_dir, f\"{organ}.csv\")\n",
+    "        if os.path.exists(organ_path):\n",
+    "            oars[organ] = mask_from_sparse(organ_path)\n",
+    "    \n",
+    "    patient_data = {\n",
+    "        \"id\": pt_id,\n",
+    "        \"ct\": ct, #ct scan -> 3D volume with intensities\n",
+    "        \"ptv\": ptv, #tumor -> 3D binary mask (1 means tumor!)\n",
+    "        \"oars\": oars #dictionary of organs at risk -> each is a 3D binary mask\n",
+    "    }\n",
     "    \n",
     "\n",
-    "#patient_tensors[0] --> patient 1\n"
+    "    pt_tensor = #Your Code Here \n",
+    "    patient_tensors.append(pt_tensor)\n",
+    "\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#Your slice visualization here\n",
+    "import matplotlib.pyplot as plt"
    ]
   },
   {
@@ -225,6 +267,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
+    "import random\n",
+    "\n",
     "beam_angles = np.arange(0, 360, 10)  #[0, 10, 20, ..., 350]\n",
     "max_beams = 5 #ensures beams are chosen strategically\n",
     "episodes = 500\n",
@@ -259,7 +303,7 @@
     "    mid = ct.shape[2] // 2 \n",
     "    ct_slice   = ct[:, :, mid]\n",
     "    ptv_slice  = ptv[:, :, mid]\n",
-    "    spine_slice = spine[:, :, mid]\n",
+    "    #extract organ slices HERE\n",
     "\n",
     "    dose = np.zeros(slice_shape, dtype=np.float32)\n",
     "    selected_angles = []\n",
@@ -269,18 +313,19 @@
     "\n",
     "        beam = generate_beam(angle, slice_shape)\n",
     "        dose += beam.astype(np.float32) #adds 'radiation' to the pixels \n",
-    "        chosen_angles.append(angle)\n",
+    "        selected_angles.append(angle)\n",
+    "\n",
     "    \n",
-    "    reward_score = reward(dose, ptv, spine)\n",
+    "    reward_score = reward(dose, ptv, organs)\n",
     "\n",
-    "    for angle in chosen_angles:\n",
+    "    for angle in selected_angles:\n",
     "        #Q-learning update here\n",
     "\n",
     "\n",
     "\n",
-    "def reward(dose, ptv, spine): #your reward function HERE\n",
+    "def reward(dose, ptv, organs): #your reward function HERE\n",
     "    raise NotImplementedError()\n",
-    "\n",
+    " \n",
     "\n",
     "Q = np.zeros(len(beam_angles)) #Q-table\n"
    ]
@@ -300,6 +345,7 @@
    "source": [
     "import matplotlib as plt\n",
     "\n",
+    "\n",
     "patient = random.choice(patient_tensors)\n",
     "mid = patient.shape[3] // 2\n",
     "ct, ptv, cord = patient[0, :, :, mid], patient[1, :, :, mid], patient[2, :, :, mid]\n",
@@ -375,8 +421,7 @@
     "                    beam_mask[r_spread, c_spread] += decay * spread_value \n",
     "\n",
     "    beam_mask = np.clip(beam_mask, 0, 1.0)  \n",
-    "    return beam_mask\n",
-    "\n"
+    "    return beam_mask"
    ]
   },
   {
@@ -437,10 +482,6 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "import numpy as np\n",
-    "import matplotlib.pyplot as plt\n",
-    "from skimage.draw import line_nd\n",
-    "\n",
     "#your conclusions\n",
     "selected_angles = \n",
     "\n",
@@ -501,7 +542,7 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "Python 3 (ipykernel)",
+   "display_name": "Python 3",
    "language": "python",
    "name": "python3"
   },
@@ -515,7 +556,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.9.7"
+   "version": "3.9.6"
   }
  },
  "nbformat": 4,