{ "id": 527759, "title": "Breaking down 2d slices of Axial Images into their respective levels", "url": "/competitions/rsna-2024-lumbar-spine-degenerative-classification/discussion/527759", "author_name": "", "post_date": "2024-08-13T15:50:23.010282800Z", "votes": 12, "comment_count": 1, "views": 0, "content": "

`import os
\nimport numpy as np
\nimport pydicom
\nimport matplotlib.pyplot as plt
\nfrom mpl_toolkits.mplot3d.art3d import Poly3DCollection
\nfrom sklearn.cluster import KMeans

\n

Function to load DICOM files and extract IOP and positions

\n

def load_dicom_slices(directory):
\n slices = []
\n filepaths = []
\n for filename in os.listdir(directory):
\n if filename.endswith(\".dcm\"):
\n filepath = os.path.join(directory, filename)
\n ds = pydicom.dcmread(filepath)
\n slices.append(ds)
\n filepaths.append(filepath)
\n return slices, filepaths

\n

def extract_iop_and_position(slices):
\n iop_vectors = []
\n positions = []
\n for slice in slices:
\n iop = np.array(slice.ImageOrientationPatient).reshape(2, 3)
\n iop_normal = np.cross(iop[0], iop[1]) # Calculate normal vector
\n position = np.array(slice.ImagePositionPatient)
\n iop_vectors.append(iop_normal)
\n positions.append(position)
\n return np.array(iop_vectors), np.array(positions)

\n

def cluster_slices_by_iop(iop_vectors, n_clusters=5):
\n kmeans = KMeans(n_clusters=n_clusters)
\n labels = kmeans.fit_predict(iop_vectors)
\n return labels

\n

def sort_slices_within_clusters(labels, positions, filepaths):
\n sorted_slices = {}
\n for cluster_label in np.unique(labels):
\n cluster_indices = np.where(labels == cluster_label)[0]
\n cluster_positions = positions[cluster_indices]
\n # Sort based on the z-component of the position vector
\n sorted_indices = np.argsort(cluster_positions[:, 2])
\n sorted_slices[cluster_label] = [filepaths[idx] for idx in cluster_indices[sorted_indices]]
\n return sorted_slices

\n

def print_cluster_file_paths(sorted_slices):
\n for cluster_label, filepaths in sorted_slices.items():
\n print(f\"Cluster {cluster_label}:\")
\n for filepath in filepaths:
\n print(f\" {filepath}\")

\n

def plot_clusters(slices, labels, positions, sorted_slices):
\n fig = plt.figure(figsize=(12, 6))

\n
\nax1 = fig.add_subplot(121, =)\nax2 = fig.add_subplot(122, =)\n\n\ncolors = plt.cm.rainbow(np.linspace(0, 1, len(sorted_slices)))\n\n cluster_label, color  zip(sorted_slices.keys(), colors):\n     idx  range(len(sorted_slices[cluster_label])):\n        #  the DICOM slice  this index\n        filepath = sorted_slices[cluster_label][idx]\n        ds = pydicom.dcmread(filepath)\n\n        #  the 3 points that define the plane (corner points of the slice)\n        iop = np.array(ds.ImageOrientationPatient).reshape(2, 3)\n        ipp = np.array(ds.ImagePositionPatient)\n        row_dir = iop[0] * ds.Rows\n        col_dir = iop[1] * ds.Columns\n\n        p1 = ipp\n        p2 = ipp + row_dir\n        p3 = ipp + col_dir\n\n        # Plane defined by three points\n        vertices = [p1, p2, p2 + col_dir, p3]\n\n        # Create a new Poly3DCollection  each axis\n        plane1 = Poly3DCollection([vertices], =color, =0.3)\n        plane2 = Poly3DCollection([vertices], =color, =0.3)\n\n        ax1.add_collection3d(plane1)\n        ax2.add_collection3d(plane2)\n\n        # Plot the points\n        ax1.scatter(*p1, =)\n        ax2.scatter(*p1, =)\n\n\nax1.set_xlabel()\nax1.set_ylabel()\nax1.set_zlabel()\n\nax2.set_xlabel()\nax2.set_ylabel()\nax2.set_zlabel()\n\n\nax1.set_box_aspect([1, 1, 1])\nax2.set_box_aspect([1, 1, 1])\n\nplt.show()\n
\n

Main function to process the DICOM slices and visualize them

\n

def process_and_plot_dicom_slices(directory, n_clusters=5):
\n slices, filepaths = load_dicom_slices(directory)
\n iop_vectors, positions = extract_iop_and_position(slices)
\n labels = cluster_slices_by_iop(iop_vectors, n_clusters)
\n sorted_slices = sort_slices_within_clusters(labels, positions, filepaths)
\n print_cluster_file_paths(sorted_slices) # Print file paths for each cluster
\n plot_clusters(slices, labels, positions, sorted_slices)

\n

Example usage

\n

dicom_directory = \"rsna_clipped/1008446160/3775545364\"
\nprocess_and_plot_dicom_slices(dicom_directory, n_clusters=5)
\n`
\nThis code outputs the images grouped by their respective levels, organized into clusters. The rest of the process is straightforward.

\n

Cluster 0:\n rsna_clipped/1008446160/3775545364\\13.dcm\n rsna_clipped/1008446160/3775545364\\12.dcm\n rsna_clipped/1008446160/3775545364\\11.dcm\n rsna_clipped/1008446160/3775545364\\10.dcm\n rsna_clipped/1008446160/3775545364\\9.dcm\nCluster 1:\n rsna_clipped/1008446160/3775545364\\24.dcm\n rsna_clipped/1008446160/3775545364\\23.dcm\n rsna_clipped/1008446160/3775545364\\22.dcm\n rsna_clipped/1008446160/3775545364\\21.dcm\n rsna_clipped/1008446160/3775545364\\20.dcm\n rsna_clipped/1008446160/3775545364\\19.dcm\nCluster 2:\n rsna_clipped/1008446160/3775545364\\18.dcm\n rsna_clipped/1008446160/3775545364\\17.dcm\n rsna_clipped/1008446160/3775545364\\16.dcm\n rsna_clipped/1008446160/3775545364\\15.dcm\n rsna_clipped/1008446160/3775545364\\14.dcm\nCluster 3:\n rsna_clipped/1008446160/3775545364\\4.dcm\n rsna_clipped/1008446160/3775545364\\3.dcm\n rsna_clipped/1008446160/3775545364\\2.dcm\n rsna_clipped/1008446160/3775545364\\1.dcm\nCluster 4:\n rsna_clipped/1008446160/3775545364\\8.dcm\n rsna_clipped/1008446160/3775545364\\7.dcm\n rsna_clipped/1008446160/3775545364\\6.dcm\n rsna_clipped/1008446160/3775545364\\5.dcm

\n

\"\"

\n

support by upvoting -- I'm 16yrs I've tried😊 tnx

", "messages": [ { "id": "2958025", "postDate": "08/13/2024 15:50:23", "content": "

`import os
\nimport numpy as np
\nimport pydicom
\nimport matplotlib.pyplot as plt
\nfrom mpl_toolkits.mplot3d.art3d import Poly3DCollection
\nfrom sklearn.cluster import KMeans

\n

Function to load DICOM files and extract IOP and positions

\n

def load_dicom_slices(directory):
\n slices = []
\n filepaths = []
\n for filename in os.listdir(directory):
\n if filename.endswith(\".dcm\"):
\n filepath = os.path.join(directory, filename)
\n ds = pydicom.dcmread(filepath)
\n slices.append(ds)
\n filepaths.append(filepath)
\n return slices, filepaths

\n

def extract_iop_and_position(slices):
\n iop_vectors = []
\n positions = []
\n for slice in slices:
\n iop = np.array(slice.ImageOrientationPatient).reshape(2, 3)
\n iop_normal = np.cross(iop[0], iop[1]) # Calculate normal vector
\n position = np.array(slice.ImagePositionPatient)
\n iop_vectors.append(iop_normal)
\n positions.append(position)
\n return np.array(iop_vectors), np.array(positions)

\n

def cluster_slices_by_iop(iop_vectors, n_clusters=5):
\n kmeans = KMeans(n_clusters=n_clusters)
\n labels = kmeans.fit_predict(iop_vectors)
\n return labels

\n

def sort_slices_within_clusters(labels, positions, filepaths):
\n sorted_slices = {}
\n for cluster_label in np.unique(labels):
\n cluster_indices = np.where(labels == cluster_label)[0]
\n cluster_positions = positions[cluster_indices]
\n # Sort based on the z-component of the position vector
\n sorted_indices = np.argsort(cluster_positions[:, 2])
\n sorted_slices[cluster_label] = [filepaths[idx] for idx in cluster_indices[sorted_indices]]
\n return sorted_slices

\n

def print_cluster_file_paths(sorted_slices):
\n for cluster_label, filepaths in sorted_slices.items():
\n print(f\"Cluster {cluster_label}:\")
\n for filepath in filepaths:
\n print(f\" {filepath}\")

\n

def plot_clusters(slices, labels, positions, sorted_slices):
\n fig = plt.figure(figsize=(12, 6))

\n
\nax1 = fig.add_subplot(121, =)\nax2 = fig.add_subplot(122, =)\n\n\ncolors = plt.cm.rainbow(np.linspace(0, 1, len(sorted_slices)))\n\n cluster_label, color  zip(sorted_slices.keys(), colors):\n     idx  range(len(sorted_slices[cluster_label])):\n        #  the DICOM slice  this index\n        filepath = sorted_slices[cluster_label][idx]\n        ds = pydicom.dcmread(filepath)\n\n        #  the 3 points that define the plane (corner points of the slice)\n        iop = np.array(ds.ImageOrientationPatient).reshape(2, 3)\n        ipp = np.array(ds.ImagePositionPatient)\n        row_dir = iop[0] * ds.Rows\n        col_dir = iop[1] * ds.Columns\n\n        p1 = ipp\n        p2 = ipp + row_dir\n        p3 = ipp + col_dir\n\n        # Plane defined by three points\n        vertices = [p1, p2, p2 + col_dir, p3]\n\n        # Create a new Poly3DCollection  each axis\n        plane1 = Poly3DCollection([vertices], =color, =0.3)\n        plane2 = Poly3DCollection([vertices], =color, =0.3)\n\n        ax1.add_collection3d(plane1)\n        ax2.add_collection3d(plane2)\n\n        # Plot the points\n        ax1.scatter(*p1, =)\n        ax2.scatter(*p1, =)\n\n\nax1.set_xlabel()\nax1.set_ylabel()\nax1.set_zlabel()\n\nax2.set_xlabel()\nax2.set_ylabel()\nax2.set_zlabel()\n\n\nax1.set_box_aspect([1, 1, 1])\nax2.set_box_aspect([1, 1, 1])\n\nplt.show()\n
\n

Main function to process the DICOM slices and visualize them

\n

def process_and_plot_dicom_slices(directory, n_clusters=5):
\n slices, filepaths = load_dicom_slices(directory)
\n iop_vectors, positions = extract_iop_and_position(slices)
\n labels = cluster_slices_by_iop(iop_vectors, n_clusters)
\n sorted_slices = sort_slices_within_clusters(labels, positions, filepaths)
\n print_cluster_file_paths(sorted_slices) # Print file paths for each cluster
\n plot_clusters(slices, labels, positions, sorted_slices)

\n

Example usage

\n

dicom_directory = \"rsna_clipped/1008446160/3775545364\"
\nprocess_and_plot_dicom_slices(dicom_directory, n_clusters=5)
\n`
\nThis code outputs the images grouped by their respective levels, organized into clusters. The rest of the process is straightforward.

\n

Cluster 0:\n rsna_clipped/1008446160/3775545364\\13.dcm\n rsna_clipped/1008446160/3775545364\\12.dcm\n rsna_clipped/1008446160/3775545364\\11.dcm\n rsna_clipped/1008446160/3775545364\\10.dcm\n rsna_clipped/1008446160/3775545364\\9.dcm\nCluster 1:\n rsna_clipped/1008446160/3775545364\\24.dcm\n rsna_clipped/1008446160/3775545364\\23.dcm\n rsna_clipped/1008446160/3775545364\\22.dcm\n rsna_clipped/1008446160/3775545364\\21.dcm\n rsna_clipped/1008446160/3775545364\\20.dcm\n rsna_clipped/1008446160/3775545364\\19.dcm\nCluster 2:\n rsna_clipped/1008446160/3775545364\\18.dcm\n rsna_clipped/1008446160/3775545364\\17.dcm\n rsna_clipped/1008446160/3775545364\\16.dcm\n rsna_clipped/1008446160/3775545364\\15.dcm\n rsna_clipped/1008446160/3775545364\\14.dcm\nCluster 3:\n rsna_clipped/1008446160/3775545364\\4.dcm\n rsna_clipped/1008446160/3775545364\\3.dcm\n rsna_clipped/1008446160/3775545364\\2.dcm\n rsna_clipped/1008446160/3775545364\\1.dcm\nCluster 4:\n rsna_clipped/1008446160/3775545364\\8.dcm\n rsna_clipped/1008446160/3775545364\\7.dcm\n rsna_clipped/1008446160/3775545364\\6.dcm\n rsna_clipped/1008446160/3775545364\\5.dcm

\n

\"\"

\n

support by upvoting -- I'm 16yrs I've tried😊 tnx

", "rawMarkdown": "`import os\nimport numpy as np\nimport pydicom\nimport matplotlib.pyplot as plt\nfrom mpl_toolkits.mplot3d.art3d import Poly3DCollection\nfrom sklearn.cluster import KMeans\n\n# Function to load DICOM files and extract IOP and positions\ndef load_dicom_slices(directory):\n slices = []\n filepaths = []\n for filename in os.listdir(directory):\n if filename.endswith(\".dcm\"):\n filepath = os.path.join(directory, filename)\n ds = pydicom.dcmread(filepath)\n slices.append(ds)\n filepaths.append(filepath)\n return slices, filepaths\n\ndef extract_iop_and_position(slices):\n iop_vectors = []\n positions = []\n for slice in slices:\n iop = np.array(slice.ImageOrientationPatient).reshape(2, 3)\n iop_normal = np.cross(iop[0], iop[1]) # Calculate normal vector\n position = np.array(slice.ImagePositionPatient)\n iop_vectors.append(iop_normal)\n positions.append(position)\n return np.array(iop_vectors), np.array(positions)\n\ndef cluster_slices_by_iop(iop_vectors, n_clusters=5):\n kmeans = KMeans(n_clusters=n_clusters)\n labels = kmeans.fit_predict(iop_vectors)\n return labels\n\ndef sort_slices_within_clusters(labels, positions, filepaths):\n sorted_slices = {}\n for cluster_label in np.unique(labels):\n cluster_indices = np.where(labels == cluster_label)[0]\n cluster_positions = positions[cluster_indices]\n # Sort based on the z-component of the position vector\n sorted_indices = np.argsort(cluster_positions[:, 2])\n sorted_slices[cluster_label] = [filepaths[idx] for idx in cluster_indices[sorted_indices]]\n return sorted_slices\n\ndef print_cluster_file_paths(sorted_slices):\n for cluster_label, filepaths in sorted_slices.items():\n print(f\"Cluster {cluster_label}:\")\n for filepath in filepaths:\n print(f\" {filepath}\")\n\ndef plot_clusters(slices, labels, positions, sorted_slices):\n fig = plt.figure(figsize=(12, 6))\n\n # Create subplots for different views\n ax1 = fig.add_subplot(121, projection='3d')\n ax2 = fig.add_subplot(122, projection='3d')\n\n # Define colors for clusters\n colors = plt.cm.rainbow(np.linspace(0, 1, len(sorted_slices)))\n\n for cluster_label, color in zip(sorted_slices.keys(), colors):\n for idx in range(len(sorted_slices[cluster_label])):\n # Get the DICOM slice for this index\n filepath = sorted_slices[cluster_label][idx]\n ds = pydicom.dcmread(filepath)\n\n # Get the 3 points that define the plane (corner points of the slice)\n iop = np.array(ds.ImageOrientationPatient).reshape(2, 3)\n ipp = np.array(ds.ImagePositionPatient)\n row_dir = iop[0] * ds.Rows\n col_dir = iop[1] * ds.Columns\n\n p1 = ipp\n p2 = ipp + row_dir\n p3 = ipp + col_dir\n\n # Plane defined by three points\n vertices = [p1, p2, p2 + col_dir, p3]\n\n # Create a new Poly3DCollection for each axis\n plane1 = Poly3DCollection([vertices], color=color, alpha=0.3)\n plane2 = Poly3DCollection([vertices], color=color, alpha=0.3)\n\n ax1.add_collection3d(plane1)\n ax2.add_collection3d(plane2)\n\n # Plot the points\n ax1.scatter(*p1, color='k')\n ax2.scatter(*p1, color='k')\n\n # Set axis labels\n ax1.set_xlabel('X')\n ax1.set_ylabel('Y')\n ax1.set_zlabel('Z')\n\n ax2.set_xlabel('X')\n ax2.set_ylabel('Y')\n ax2.set_zlabel('Z')\n\n # Set the aspect ratio of the plot to be equal\n ax1.set_box_aspect([1, 1, 1])\n ax2.set_box_aspect([1, 1, 1])\n\n plt.show()\n\n# Main function to process the DICOM slices and visualize them\ndef process_and_plot_dicom_slices(directory, n_clusters=5):\n slices, filepaths = load_dicom_slices(directory)\n iop_vectors, positions = extract_iop_and_position(slices)\n labels = cluster_slices_by_iop(iop_vectors, n_clusters)\n sorted_slices = sort_slices_within_clusters(labels, positions, filepaths)\n print_cluster_file_paths(sorted_slices) # Print file paths for each cluster\n plot_clusters(slices, labels, positions, sorted_slices)\n\n# Example usage\ndicom_directory = \"rsna_clipped/1008446160/3775545364\"\nprocess_and_plot_dicom_slices(dicom_directory, n_clusters=5)\n`\nThis code outputs the images grouped by their respective levels, organized into clusters. The rest of the process is straightforward.\n\n*Cluster 0:\n rsna_clipped/1008446160/3775545364\\13.dcm\n rsna_clipped/1008446160/3775545364\\12.dcm\n rsna_clipped/1008446160/3775545364\\11.dcm\n rsna_clipped/1008446160/3775545364\\10.dcm\n rsna_clipped/1008446160/3775545364\\9.dcm\nCluster 1:\n rsna_clipped/1008446160/3775545364\\24.dcm\n rsna_clipped/1008446160/3775545364\\23.dcm\n rsna_clipped/1008446160/3775545364\\22.dcm\n rsna_clipped/1008446160/3775545364\\21.dcm\n rsna_clipped/1008446160/3775545364\\20.dcm\n rsna_clipped/1008446160/3775545364\\19.dcm\nCluster 2:\n rsna_clipped/1008446160/3775545364\\18.dcm\n rsna_clipped/1008446160/3775545364\\17.dcm\n rsna_clipped/1008446160/3775545364\\16.dcm\n rsna_clipped/1008446160/3775545364\\15.dcm\n rsna_clipped/1008446160/3775545364\\14.dcm\nCluster 3:\n rsna_clipped/1008446160/3775545364\\4.dcm\n rsna_clipped/1008446160/3775545364\\3.dcm\n rsna_clipped/1008446160/3775545364\\2.dcm\n rsna_clipped/1008446160/3775545364\\1.dcm\nCluster 4:\n rsna_clipped/1008446160/3775545364\\8.dcm\n rsna_clipped/1008446160/3775545364\\7.dcm\n rsna_clipped/1008446160/3775545364\\6.dcm\n rsna_clipped/1008446160/3775545364\\5.dcm*\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F21070744%2F5cd4617727b06d4cf4c2a68d8dfcdfe3%2Fqq.png?generation=1723563844340495&alt=media)\n\nsupport by upvoting -- I'm 16yrs I've tried😊 tnx", "votes": null }, { "id": "2974081", "postDate": "08/30/2024 09:18:58", "content": "

Hi,

\n

How did you handle study_ids with multiple axial folders ?

\n

Did you merge them or cluster them separately?

", "rawMarkdown": "Hi,\n\nHow did you handle study_ids with multiple axial folders ?\n\nDid you merge them or cluster them separately?", "votes": null } ], "comments": [ { "id": 2974081, "author_name": "rohitchaudhari25", "author_url": "", "post_date": "08/30/2024 09:18:58", "content": "

Hi,

\n

How did you handle study_ids with multiple axial folders ?

\n

Did you merge them or cluster them separately?

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2958025": "`import os\nimport numpy as np\nimport pydicom\nimport matplotlib.pyplot as plt\nfrom mpl_toolkits.mplot3d.art3d import Poly3DCollection\nfrom sklearn.cluster import KMeans\n\n# Function to load DICOM files and extract IOP and positions\ndef load_dicom_slices(directory):\n slices = []\n filepaths = []\n for filename in os.listdir(directory):\n if filename.endswith(\".dcm\"):\n filepath = os.path.join(directory, filename)\n ds = pydicom.dcmread(filepath)\n slices.append(ds)\n filepaths.append(filepath)\n return slices, filepaths\n\ndef extract_iop_and_position(slices):\n iop_vectors = []\n positions = []\n for slice in slices:\n iop = np.array(slice.ImageOrientationPatient).reshape(2, 3)\n iop_normal = np.cross(iop[0], iop[1]) # Calculate normal vector\n position = np.array(slice.ImagePositionPatient)\n iop_vectors.append(iop_normal)\n positions.append(position)\n return np.array(iop_vectors), np.array(positions)\n\ndef cluster_slices_by_iop(iop_vectors, n_clusters=5):\n kmeans = KMeans(n_clusters=n_clusters)\n labels = kmeans.fit_predict(iop_vectors)\n return labels\n\ndef sort_slices_within_clusters(labels, positions, filepaths):\n sorted_slices = {}\n for cluster_label in np.unique(labels):\n cluster_indices = np.where(labels == cluster_label)[0]\n cluster_positions = positions[cluster_indices]\n # Sort based on the z-component of the position vector\n sorted_indices = np.argsort(cluster_positions[:, 2])\n sorted_slices[cluster_label] = [filepaths[idx] for idx in cluster_indices[sorted_indices]]\n return sorted_slices\n\ndef print_cluster_file_paths(sorted_slices):\n for cluster_label, filepaths in sorted_slices.items():\n print(f\"Cluster {cluster_label}:\")\n for filepath in filepaths:\n print(f\" {filepath}\")\n\ndef plot_clusters(slices, labels, positions, sorted_slices):\n fig = plt.figure(figsize=(12, 6))\n\n # Create subplots for different views\n ax1 = fig.add_subplot(121, projection='3d')\n ax2 = fig.add_subplot(122, projection='3d')\n\n # Define colors for clusters\n colors = plt.cm.rainbow(np.linspace(0, 1, len(sorted_slices)))\n\n for cluster_label, color in zip(sorted_slices.keys(), colors):\n for idx in range(len(sorted_slices[cluster_label])):\n # Get the DICOM slice for this index\n filepath = sorted_slices[cluster_label][idx]\n ds = pydicom.dcmread(filepath)\n\n # Get the 3 points that define the plane (corner points of the slice)\n iop = np.array(ds.ImageOrientationPatient).reshape(2, 3)\n ipp = np.array(ds.ImagePositionPatient)\n row_dir = iop[0] * ds.Rows\n col_dir = iop[1] * ds.Columns\n\n p1 = ipp\n p2 = ipp + row_dir\n p3 = ipp + col_dir\n\n # Plane defined by three points\n vertices = [p1, p2, p2 + col_dir, p3]\n\n # Create a new Poly3DCollection for each axis\n plane1 = Poly3DCollection([vertices], color=color, alpha=0.3)\n plane2 = Poly3DCollection([vertices], color=color, alpha=0.3)\n\n ax1.add_collection3d(plane1)\n ax2.add_collection3d(plane2)\n\n # Plot the points\n ax1.scatter(*p1, color='k')\n ax2.scatter(*p1, color='k')\n\n # Set axis labels\n ax1.set_xlabel('X')\n ax1.set_ylabel('Y')\n ax1.set_zlabel('Z')\n\n ax2.set_xlabel('X')\n ax2.set_ylabel('Y')\n ax2.set_zlabel('Z')\n\n # Set the aspect ratio of the plot to be equal\n ax1.set_box_aspect([1, 1, 1])\n ax2.set_box_aspect([1, 1, 1])\n\n plt.show()\n\n# Main function to process the DICOM slices and visualize them\ndef process_and_plot_dicom_slices(directory, n_clusters=5):\n slices, filepaths = load_dicom_slices(directory)\n iop_vectors, positions = extract_iop_and_position(slices)\n labels = cluster_slices_by_iop(iop_vectors, n_clusters)\n sorted_slices = sort_slices_within_clusters(labels, positions, filepaths)\n print_cluster_file_paths(sorted_slices) # Print file paths for each cluster\n plot_clusters(slices, labels, positions, sorted_slices)\n\n# Example usage\ndicom_directory = \"rsna_clipped/1008446160/3775545364\"\nprocess_and_plot_dicom_slices(dicom_directory, n_clusters=5)\n`\nThis code outputs the images grouped by their respective levels, organized into clusters. The rest of the process is straightforward.\n\n*Cluster 0:\n rsna_clipped/1008446160/3775545364\\13.dcm\n rsna_clipped/1008446160/3775545364\\12.dcm\n rsna_clipped/1008446160/3775545364\\11.dcm\n rsna_clipped/1008446160/3775545364\\10.dcm\n rsna_clipped/1008446160/3775545364\\9.dcm\nCluster 1:\n rsna_clipped/1008446160/3775545364\\24.dcm\n rsna_clipped/1008446160/3775545364\\23.dcm\n rsna_clipped/1008446160/3775545364\\22.dcm\n rsna_clipped/1008446160/3775545364\\21.dcm\n rsna_clipped/1008446160/3775545364\\20.dcm\n rsna_clipped/1008446160/3775545364\\19.dcm\nCluster 2:\n rsna_clipped/1008446160/3775545364\\18.dcm\n rsna_clipped/1008446160/3775545364\\17.dcm\n rsna_clipped/1008446160/3775545364\\16.dcm\n rsna_clipped/1008446160/3775545364\\15.dcm\n rsna_clipped/1008446160/3775545364\\14.dcm\nCluster 3:\n rsna_clipped/1008446160/3775545364\\4.dcm\n rsna_clipped/1008446160/3775545364\\3.dcm\n rsna_clipped/1008446160/3775545364\\2.dcm\n rsna_clipped/1008446160/3775545364\\1.dcm\nCluster 4:\n rsna_clipped/1008446160/3775545364\\8.dcm\n rsna_clipped/1008446160/3775545364\\7.dcm\n rsna_clipped/1008446160/3775545364\\6.dcm\n rsna_clipped/1008446160/3775545364\\5.dcm*\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F21070744%2F5cd4617727b06d4cf4c2a68d8dfcdfe3%2Fqq.png?generation=1723563844340495&alt=media)\n\nsupport by upvoting -- I'm 16yrs I've tried😊 tnx", "2974081": "Hi,\n\nHow did you handle study_ids with multiple axial folders ?\n\nDid you merge them or cluster them separately?" }, "source": "meta" }