{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.10.14","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"none","dataSources":[{"sourceId":71549,"databundleVersionId":8561470,"sourceType":"competition"}],"dockerImageVersionId":30761,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"# This Python 3 environment comes with many helpful analytics libraries installed\n# It is defined by the kaggle/python Docker image: https://github.com/kaggle/docker-python\n# For example, here's several helpful packages to load\n\nimport numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\n\n# Input data files are available in the read-only \"../input/\" directory\n# For example, running this (by clicking run or pressing Shift+Enter) will list all files under the input directory\n\nimport gc\nimport sys\nfrom PIL import Image\nimport cv2\nimport math, random\nimport numpy as np\nimport pandas as pd\nimport glob\nfrom glob import glob\nfrom tqdm import tqdm\nimport matplotlib.pyplot as plt\nfrom sklearn.model_selection import KFold\nfrom collections import OrderedDict\nimport tensorflow as tf\nfrom tensorflow.keras import layers, models, optimizers\nfrom tensorflow.keras.models import load_model\nimport albumentations as A\nfrom sklearn.model_selection import KFold\nimport re\nimport pydicom\nfrom typing import Optional\nimport torch\nfrom torch.utils.data import Dataset, DataLoader # Import Dataset here\nimport torch.nn as nn\nimport torch.optim as optim\nimport glob\nimport torch\nfrom torch.utils.data import DataLoader, Dataset\nfrom tqdm import tqdm\nimport timm\n\n# You can write up to 20GB to the current directory (/kaggle/working/) that gets preserved as output when you create a version using \"Save & Run All\" \n# You can also write temporary files to /kaggle/temp/, but they won't be saved outside of the current session","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2024-10-02T13:38:27.793333Z","iopub.execute_input":"2024-10-02T13:38:27.793852Z","iopub.status.idle":"2024-10-02T13:38:53.311284Z","shell.execute_reply.started":"2024-10-02T13:38:27.793784Z","shell.execute_reply":"2024-10-02T13:38:53.310014Z"},"trusted":true},"execution_count":1,"outputs":[{"name":"stderr","text":"/opt/conda/lib/python3.10/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 1.4.17 (you have 1.4.14). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.\n check_for_updates()\n","output_type":"stream"}]},{"cell_type":"code","source":"import os\nimport numpy as np\nfrom tensorflow.keras.preprocessing.image import load_img, img_to_array\nfrom tensorflow.keras.utils import to_categorical\nfrom tensorflow.keras.models import Sequential\nfrom tensorflow.keras.layers import Conv2D, MaxPooling2D, Flatten, Dense, Dropout\nfrom sklearn.model_selection import train_test_split\n\n# Directory where your 2D images are stored","metadata":{"execution":{"iopub.status.busy":"2024-10-02T13:38:53.313648Z","iopub.execute_input":"2024-10-02T13:38:53.314149Z","iopub.status.idle":"2024-10-02T13:38:53.3274Z","shell.execute_reply.started":"2024-10-02T13:38:53.314092Z","shell.execute_reply":"2024-10-02T13:38:53.326073Z"},"trusted":true},"execution_count":2,"outputs":[]},{"cell_type":"code","source":"# import pandas as pd\n\n# # Load the data\n# df = pd.read_csv('/kaggle/input/rsna-2024-lumbar-spine-degenerative-classification/train_series_descriptions.csv')\n# train = pd.read_csv('/kaggle/input/rsna-2024-lumbar-spine-degenerative-classification/train.csv')\n\n# # Columns to check for 'Moderate' or 'Severe' conditions\n# columns_to_check = train.columns[1:6]\n\n# # Create a condition based on 'Moderate' or 'Severe'\n# condition = train[columns_to_check].apply(lambda row: any(val == 'Moderate' or val == 'Severe' for val in row), axis=1)\n\n# # Filter the train DataFrame\n# df_train = train[condition]\n\n# # Filter the df DataFrame to include only study_id values that are present in df_train\n# df_filtered = df[df['study_id'].isin(df_train['study_id'])]\n\n# # Display the filtered DataFrame\n# df_filtered","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# # Define the order for series_description\n# order = [\"Axial T2\", \"Sagittal T1\", \"Sagittal T2/STIR\"]\n\n# # Create a categorical type for series_description based on the order\n# df_filtered['series_description'] = pd.Categorical(df_filtered['series_description'], categories=order, ordered=True)\n\n# # Sort the DataFrame by study_id and series_description to ensure correct order\n# df_filtered = df_filtered.sort_values(by=['study_id', 'series_description'])\n\n# # For each study_id, drop extra duplicates, keeping the first occurrence for each series_description\n# df_filtered = df_filtered.drop_duplicates(subset=['study_id', 'series_description'], keep='first')\n\n# # Reset the index if needed\n# df_filtered.reset_index(drop=True, inplace=True)\n\n# # Display the rearranged DataFrame\n# df_filtered\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import pandas as pd\n\n# # Load the data\n# df = pd.read_csv('/kaggle/input/rsna-2024-lumbar-spine-degenerative-classification/train_series_descriptions.csv')\n# train = pd.read_csv('/kaggle/input/rsna-2024-lumbar-spine-degenerative-classification/train.csv')\n\n# # Columns to check for 'Moderate' or 'Severe' conditions\n# columns_to_check = train.columns[1:6]\n\n# # Create a condition based on 'Moderate' or 'Severe'\n# condition = train[columns_to_check].apply(lambda row: any(val == 'Moderate' or val == 'Severe' for val in row), axis=1)\n\n# # Filter the train DataFrame\n# df_train = train[condition]\n\n# # Filter the df DataFrame to include only study_id values that are present in df_train\n# df_filtered = df[df['study_id'].isin(df_train['study_id'])]\n\n# # Define the order for series_description\n# order = [\"Axial T2\", \"Sagittal T1\", \"Sagittal T2/STIR\"]\n\n# # Create a categorical type for series_description based on the order\n# df_filtered['series_description'] = pd.Categorical(df_filtered['series_description'], categories=order, ordered=True)\n\n# # Sort the DataFrame by study_id and series_description to ensure correct order\n# df_filtered = df_filtered.sort_values(by=['study_id', 'series_description'])\n\n# # For each study_id, drop extra duplicates, keeping the first occurrence for each series_description\n# df_filtered = df_filtered.drop_duplicates(subset=['study_id', 'series_description'], keep='first')\n\n# # Reset the index if needed\n# df_filtered.reset_index(drop=True, inplace=True)\n\n# # Ensure study_id in df_train is in the same order as in df_filtered\n# ordered_study_ids = df_filtered['study_id'].unique()\n# df_train = df_train[df_train['study_id'].isin(ordered_study_ids)]\n# df_train = df_train.set_index('study_id').reindex(ordered_study_ids).reset_index()\n\n# # Display the rearranged DataFrame\n# df_filtered, df_train\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import pandas as pd\n\n# Load the data\ndf = pd.read_csv('/kaggle/input/rsna-2024-lumbar-spine-degenerative-classification/train_series_descriptions.csv')\ntrain = pd.read_csv('/kaggle/input/rsna-2024-lumbar-spine-degenerative-classification/train.csv')\n\n# Columns to check for 'Moderate' or 'Severe' conditions\ncolumns_to_check = train.columns[1:6]\n\n# Create a condition based on 'Moderate' or 'Severe'\ncondition = train[columns_to_check].apply(lambda row: any(val == 'Moderate' or val == 'Severe' for val in row), axis=1)\n\n# Filter the train DataFrame\ndf_train = train[condition]\n\n# Select the first 30 study_id from df_train\nfirst_30_study_ids = df_train['study_id'].unique()[:540]\n\n# Filter df_train to include only the first 30 study_id\ndf_train = df_train[df_train['study_id'].isin(first_30_study_ids)]\n\n# Filter df DataFrame to include only study_id values that are present in df_train\ndf_filtered = df[df['study_id'].isin(first_30_study_ids)]\n\n# Define the order for series_description\norder = [\"Axial T2\", \"Sagittal T1\", \"Sagittal T2/STIR\"]\n\n# Create a categorical type for series_description based on the order\ndf_filtered['series_description'] = pd.Categorical(df_filtered['series_description'], categories=order, ordered=True)\n\n# Sort the DataFrame by study_id and series_description to ensure correct order\ndf_filtered = df_filtered.sort_values(by=['study_id', 'series_description'])\n\n# For each study_id, drop extra duplicates, keeping the first occurrence for each series_description\ndf_filtered = df_filtered.drop_duplicates(subset=['study_id', 'series_description'], keep='first')\n\n# Reset the index if needed\ndf_filtered.reset_index(drop=True, inplace=True)\n\n# Ensure study_id in df_train is in the same order as in df_filtered\nordered_study_ids = df_filtered['study_id'].unique()\ndf_train = df_train[df_train['study_id'].isin(ordered_study_ids)]\ndf_train = df_train.set_index('study_id').reindex(ordered_study_ids).reset_index()\n","metadata":{"execution":{"iopub.status.busy":"2024-10-02T13:39:05.014601Z","iopub.execute_input":"2024-10-02T13:39:05.015103Z","iopub.status.idle":"2024-10-02T13:39:05.161511Z","shell.execute_reply.started":"2024-10-02T13:39:05.015054Z","shell.execute_reply":"2024-10-02T13:39:05.16017Z"},"trusted":true},"execution_count":3,"outputs":[{"name":"stderr","text":"/tmp/ipykernel_36/2173964938.py:29: SettingWithCopyWarning: \nA value is trying to be set on a copy of a slice from a DataFrame.\nTry using .loc[row_indexer,col_indexer] = value instead\n\nSee the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n df_filtered['series_description'] = pd.Categorical(df_filtered['series_description'], categories=order, ordered=True)\n","output_type":"stream"}]},{"cell_type":"code","source":"# import pandas as pd\n\n# # Load the data\n# df = pd.read_csv('/kaggle/input/rsna-2024-lumbar-spine-degenerative-classification/train_series_descriptions.csv')\n# train = pd.read_csv('/kaggle/input/rsna-2024-lumbar-spine-degenerative-classification/train.csv')\n\n# # Columns to check for 'Moderate' or 'Severe' conditions\n# columns_to_check = train.columns[1:6]\n\n# # Create a condition based on 'Moderate' or 'Severe'\n# condition = train[columns_to_check].apply(lambda row: any(val == 'Moderate' or val == 'Severe' for val in row), axis=1)\n\n# # Filter the train DataFrame based on the condition\n# df_train = train[condition]\n\n# # Select study_id from the range 180:270 (90 study_ids)\n# study_ids_180_270 = df_train['study_id'].unique()[180:270]\n\n# # Filter df_train to include only the study_ids in the range 180:270\n# df_train = df_train[df_train['study_id'].isin(study_ids_180_270)]\n\n# # Filter df DataFrame to include only study_id values that are present in df_train\n# df_filtered = df[df['study_id'].isin(study_ids_180_270)]\n\n# # Define the order for series_description\n# order = [\"Axial T2\", \"Sagittal T1\", \"Sagittal T2/STIR\"]\n\n# # Create a categorical type for series_description based on the order\n# df_filtered['series_description'] = pd.Categorical(df_filtered['series_description'], categories=order, ordered=True)\n\n# # Sort the DataFrame by study_id and series_description to ensure correct order\n# df_filtered = df_filtered.sort_values(by=['study_id', 'series_description'])\n\n# # For each study_id, drop extra duplicates, keeping the first occurrence for each series_description\n# df_filtered = df_filtered.drop_duplicates(subset=['study_id', 'series_description'], keep='first')\n\n# # Reset the index if needed\n# df_filtered.reset_index(drop=True, inplace=True)\n\n# # Ensure study_id in df_train is in the same order as in df_filtered\n# ordered_study_ids = df_filtered['study_id'].unique()\n# df_train = df_train[df_train['study_id'].isin(ordered_study_ids)]\n# df_train = df_train.set_index('study_id').reindex(ordered_study_ids).reset_index()\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_train","metadata":{"execution":{"iopub.status.busy":"2024-10-02T13:39:14.067603Z","iopub.execute_input":"2024-10-02T13:39:14.068818Z","iopub.status.idle":"2024-10-02T13:39:14.109819Z","shell.execute_reply.started":"2024-10-02T13:39:14.068762Z","shell.execute_reply":"2024-10-02T13:39:14.108416Z"},"trusted":true},"execution_count":4,"outputs":[{"execution_count":4,"output_type":"execute_result","data":{"text/plain":" study_id spinal_canal_stenosis_l1_l2 spinal_canal_stenosis_l2_l3 \\\n0 4646740 Normal/Mild Normal/Mild \n1 22191399 Normal/Mild Normal/Mild \n2 29931867 Normal/Mild Moderate \n3 38281420 Normal/Mild Normal/Mild \n4 40745534 Normal/Mild Normal/Mild \n.. ... ... ... \n535 3158185203 Normal/Mild Moderate \n536 3160528641 Normal/Mild Normal/Mild \n537 3163594538 Severe Severe \n538 3168755174 Normal/Mild Normal/Mild \n539 3172703788 Moderate Normal/Mild \n\n spinal_canal_stenosis_l3_l4 spinal_canal_stenosis_l4_l5 \\\n0 Moderate Severe \n1 Normal/Mild Severe \n2 Moderate Normal/Mild \n3 Normal/Mild Moderate \n4 Moderate Severe \n.. ... ... \n535 Normal/Mild Normal/Mild \n536 Normal/Mild Moderate \n537 Severe Severe \n538 Severe Normal/Mild \n539 Normal/Mild Normal/Mild \n\n spinal_canal_stenosis_l5_s1 left_neural_foraminal_narrowing_l1_l2 \\\n0 Normal/Mild Normal/Mild \n1 Normal/Mild Normal/Mild \n2 Normal/Mild Normal/Mild \n3 Normal/Mild Normal/Mild \n4 Normal/Mild Normal/Mild \n.. ... ... \n535 Normal/Mild Normal/Mild \n536 Normal/Mild Normal/Mild \n537 Normal/Mild Moderate \n538 Normal/Mild Normal/Mild \n539 Normal/Mild Normal/Mild \n\n left_neural_foraminal_narrowing_l2_l3 \\\n0 Normal/Mild \n1 Normal/Mild \n2 Normal/Mild \n3 Normal/Mild \n4 Normal/Mild \n.. ... \n535 Moderate \n536 Normal/Mild \n537 Moderate \n538 Normal/Mild \n539 Normal/Mild \n\n left_neural_foraminal_narrowing_l3_l4 \\\n0 Normal/Mild \n1 Normal/Mild \n2 Normal/Mild \n3 Normal/Mild \n4 Normal/Mild \n.. ... \n535 Moderate \n536 Moderate \n537 Moderate \n538 Normal/Mild \n539 Normal/Mild \n\n left_neural_foraminal_narrowing_l4_l5 ... \\\n0 Moderate ... \n1 Normal/Mild ... \n2 Moderate ... \n3 Moderate ... \n4 Moderate ... \n.. ... ... \n535 Severe ... \n536 Moderate ... \n537 Moderate ... \n538 Moderate ... \n539 Normal/Mild ... \n\n left_subarticular_stenosis_l1_l2 left_subarticular_stenosis_l2_l3 \\\n0 Normal/Mild Normal/Mild \n1 Normal/Mild Normal/Mild \n2 Normal/Mild Normal/Mild \n3 Normal/Mild Normal/Mild \n4 Normal/Mild Normal/Mild \n.. ... ... \n535 Normal/Mild Moderate \n536 Normal/Mild Normal/Mild \n537 Severe Severe \n538 NaN NaN \n539 Normal/Mild Normal/Mild \n\n left_subarticular_stenosis_l3_l4 left_subarticular_stenosis_l4_l5 \\\n0 Normal/Mild Severe \n1 Normal/Mild Moderate \n2 Moderate Moderate \n3 Normal/Mild Normal/Mild \n4 Moderate Severe \n.. ... ... \n535 Moderate Severe \n536 Moderate Moderate \n537 Severe Severe \n538 Severe Normal/Mild \n539 Normal/Mild Normal/Mild \n\n left_subarticular_stenosis_l5_s1 right_subarticular_stenosis_l1_l2 \\\n0 Normal/Mild Normal/Mild \n1 Moderate Normal/Mild \n2 Moderate Moderate \n3 Normal/Mild Normal/Mild \n4 Moderate Normal/Mild \n.. ... ... \n535 Normal/Mild Normal/Mild \n536 Normal/Mild Normal/Mild \n537 Normal/Mild Moderate \n538 Normal/Mild NaN \n539 Normal/Mild Normal/Mild \n\n right_subarticular_stenosis_l2_l3 right_subarticular_stenosis_l3_l4 \\\n0 Moderate Moderate \n1 Normal/Mild Normal/Mild \n2 Severe Severe \n3 Moderate Moderate \n4 Normal/Mild Moderate \n.. ... ... \n535 Moderate Normal/Mild \n536 Normal/Mild Normal/Mild \n537 Severe Severe \n538 NaN Severe \n539 Normal/Mild Normal/Mild \n\n right_subarticular_stenosis_l4_l5 right_subarticular_stenosis_l5_s1 \n0 Moderate Normal/Mild \n1 Severe Moderate \n2 Moderate Severe \n3 Severe Severe \n4 Severe Normal/Mild \n.. ... ... \n535 Moderate Normal/Mild \n536 Severe Normal/Mild \n537 Severe Moderate \n538 Severe Severe \n539 Moderate Moderate \n\n[540 rows x 26 columns]","text/html":"
\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n
study_idspinal_canal_stenosis_l1_l2spinal_canal_stenosis_l2_l3spinal_canal_stenosis_l3_l4spinal_canal_stenosis_l4_l5spinal_canal_stenosis_l5_s1left_neural_foraminal_narrowing_l1_l2left_neural_foraminal_narrowing_l2_l3left_neural_foraminal_narrowing_l3_l4left_neural_foraminal_narrowing_l4_l5...left_subarticular_stenosis_l1_l2left_subarticular_stenosis_l2_l3left_subarticular_stenosis_l3_l4left_subarticular_stenosis_l4_l5left_subarticular_stenosis_l5_s1right_subarticular_stenosis_l1_l2right_subarticular_stenosis_l2_l3right_subarticular_stenosis_l3_l4right_subarticular_stenosis_l4_l5right_subarticular_stenosis_l5_s1
04646740Normal/MildNormal/MildModerateSevereNormal/MildNormal/MildNormal/MildNormal/MildModerate...Normal/MildNormal/MildNormal/MildSevereNormal/MildNormal/MildModerateModerateModerateNormal/Mild
122191399Normal/MildNormal/MildNormal/MildSevereNormal/MildNormal/MildNormal/MildNormal/MildNormal/Mild...Normal/MildNormal/MildNormal/MildModerateModerateNormal/MildNormal/MildNormal/MildSevereModerate
229931867Normal/MildModerateModerateNormal/MildNormal/MildNormal/MildNormal/MildNormal/MildModerate...Normal/MildNormal/MildModerateModerateModerateModerateSevereSevereModerateSevere
338281420Normal/MildNormal/MildNormal/MildModerateNormal/MildNormal/MildNormal/MildNormal/MildModerate...Normal/MildNormal/MildNormal/MildNormal/MildNormal/MildNormal/MildModerateModerateSevereSevere
440745534Normal/MildNormal/MildModerateSevereNormal/MildNormal/MildNormal/MildNormal/MildModerate...Normal/MildNormal/MildModerateSevereModerateNormal/MildNormal/MildModerateSevereNormal/Mild
..................................................................
5353158185203Normal/MildModerateNormal/MildNormal/MildNormal/MildNormal/MildModerateModerateSevere...Normal/MildModerateModerateSevereNormal/MildNormal/MildModerateNormal/MildModerateNormal/Mild
5363160528641Normal/MildNormal/MildNormal/MildModerateNormal/MildNormal/MildNormal/MildModerateModerate...Normal/MildNormal/MildModerateModerateNormal/MildNormal/MildNormal/MildNormal/MildSevereNormal/Mild
5373163594538SevereSevereSevereSevereNormal/MildModerateModerateModerateModerate...SevereSevereSevereSevereNormal/MildModerateSevereSevereSevereModerate
5383168755174Normal/MildNormal/MildSevereNormal/MildNormal/MildNormal/MildNormal/MildNormal/MildModerate...NaNNaNSevereNormal/MildNormal/MildNaNNaNSevereSevereSevere
5393172703788ModerateNormal/MildNormal/MildNormal/MildNormal/MildNormal/MildNormal/MildNormal/MildNormal/Mild...Normal/MildNormal/MildNormal/MildNormal/MildNormal/MildNormal/MildNormal/MildNormal/MildModerateModerate
\n

540 rows × 26 columns

\n
"},"metadata":{}}]},{"cell_type":"code","source":"df_filtered","metadata":{"execution":{"iopub.status.busy":"2024-10-02T13:39:16.985145Z","iopub.execute_input":"2024-10-02T13:39:16.985654Z","iopub.status.idle":"2024-10-02T13:39:17.000769Z","shell.execute_reply.started":"2024-10-02T13:39:16.985605Z","shell.execute_reply":"2024-10-02T13:39:16.999355Z"},"trusted":true},"execution_count":5,"outputs":[{"execution_count":5,"output_type":"execute_result","data":{"text/plain":" study_id series_id series_description\n0 4646740 3201256954 Axial T2\n1 4646740 3486248476 Sagittal T1\n2 4646740 3666319702 Sagittal T2/STIR\n3 22191399 3687121182 Axial T2\n4 22191399 434280813 Sagittal T1\n... ... ... ...\n1614 3168755174 3335509714 Sagittal T1\n1615 3168755174 3299192635 Sagittal T2/STIR\n1616 3172703788 3357340646 Axial T2\n1617 3172703788 1681952265 Sagittal T1\n1618 3172703788 619632020 Sagittal T2/STIR\n\n[1619 rows x 3 columns]","text/html":"
\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n
study_idseries_idseries_description
046467403201256954Axial T2
146467403486248476Sagittal T1
246467403666319702Sagittal T2/STIR
3221913993687121182Axial T2
422191399434280813Sagittal T1
............
161431687551743335509714Sagittal T1
161531687551743299192635Sagittal T2/STIR
161631727037883357340646Axial T2
161731727037881681952265Sagittal T1
16183172703788619632020Sagittal T2/STIR
\n

1619 rows × 3 columns

\n
"},"metadata":{}}]},{"cell_type":"code","source":"df_filtered.head(30)","metadata":{"execution":{"iopub.status.busy":"2024-10-02T13:39:19.308369Z","iopub.execute_input":"2024-10-02T13:39:19.308875Z","iopub.status.idle":"2024-10-02T13:39:19.326062Z","shell.execute_reply.started":"2024-10-02T13:39:19.308829Z","shell.execute_reply":"2024-10-02T13:39:19.324664Z"},"trusted":true},"execution_count":6,"outputs":[{"execution_count":6,"output_type":"execute_result","data":{"text/plain":" study_id series_id series_description\n0 4646740 3201256954 Axial T2\n1 4646740 3486248476 Sagittal T1\n2 4646740 3666319702 Sagittal T2/STIR\n3 22191399 3687121182 Axial T2\n4 22191399 434280813 Sagittal T1\n5 22191399 3753885158 Sagittal T2/STIR\n6 29931867 231278500 Axial T2\n7 29931867 1152175603 Sagittal T1\n8 29931867 1676821058 Sagittal T2/STIR\n9 38281420 1178941473 Axial T2\n10 38281420 2565838687 Sagittal T1\n11 38281420 880361156 Sagittal T2/STIR\n12 40745534 1395478236 Axial T2\n13 40745534 2948359731 Sagittal T1\n14 40745534 3918681171 Sagittal T2/STIR\n15 41477684 3749739759 Axial T2\n16 41477684 3009399271 Sagittal T1\n17 41477684 2595734107 Sagittal T2/STIR\n18 46494080 1543341132 Axial T2\n19 46494080 4061588226 Sagittal T1\n20 46494080 1763376930 Sagittal T2/STIR\n21 52397721 3220885418 Axial T2\n22 52397721 2770638094 Sagittal T1\n23 52397721 2452297573 Sagittal T2/STIR\n24 52695609 3778925627 Axial T2\n25 52695609 3284178089 Sagittal T1\n26 52695609 32135551 Sagittal T2/STIR\n27 58813022 302608236 Axial T2\n28 58813022 692718517 Sagittal T1\n29 58813022 3875372861 Sagittal T2/STIR","text/html":"
\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n
study_idseries_idseries_description
046467403201256954Axial T2
146467403486248476Sagittal T1
246467403666319702Sagittal T2/STIR
3221913993687121182Axial T2
422191399434280813Sagittal T1
5221913993753885158Sagittal T2/STIR
629931867231278500Axial T2
7299318671152175603Sagittal T1
8299318671676821058Sagittal T2/STIR
9382814201178941473Axial T2
10382814202565838687Sagittal T1
1138281420880361156Sagittal T2/STIR
12407455341395478236Axial T2
13407455342948359731Sagittal T1
14407455343918681171Sagittal T2/STIR
15414776843749739759Axial T2
16414776843009399271Sagittal T1
17414776842595734107Sagittal T2/STIR
18464940801543341132Axial T2
19464940804061588226Sagittal T1
20464940801763376930Sagittal T2/STIR
21523977213220885418Axial T2
22523977212770638094Sagittal T1
23523977212452297573Sagittal T2/STIR
24526956093778925627Axial T2
25526956093284178089Sagittal T1
265269560932135551Sagittal T2/STIR
2758813022302608236Axial T2
2858813022692718517Sagittal T1
29588130223875372861Sagittal T2/STIR
\n
"},"metadata":{}}]},{"cell_type":"code","source":"df_filtered.tail(30)","metadata":{"execution":{"iopub.status.busy":"2024-10-02T13:39:22.751474Z","iopub.execute_input":"2024-10-02T13:39:22.751936Z","iopub.status.idle":"2024-10-02T13:39:22.766903Z","shell.execute_reply.started":"2024-10-02T13:39:22.751892Z","shell.execute_reply":"2024-10-02T13:39:22.765497Z"},"trusted":true},"execution_count":7,"outputs":[{"execution_count":7,"output_type":"execute_result","data":{"text/plain":" study_id series_id series_description\n1589 3128795155 2985676628 Axial T2\n1590 3128795155 483456514 Sagittal T1\n1591 3128795155 246275868 Sagittal T2/STIR\n1592 3138242355 684398187 Axial T2\n1593 3138242355 3310553683 Sagittal T1\n1594 3138242355 992065261 Sagittal T2/STIR\n1595 3139274536 1038845969 Axial T2\n1596 3139274536 2171939378 Sagittal T1\n1597 3139274536 3906320584 Sagittal T2/STIR\n1598 3141331592 3267884718 Axial T2\n1599 3141331592 72737584 Sagittal T1\n1600 3141331592 1244841079 Sagittal T2/STIR\n1601 3156269631 501368228 Axial T2\n1602 3156269631 4030602643 Sagittal T1\n1603 3156269631 4059442027 Sagittal T2/STIR\n1604 3158185203 2163803231 Axial T2\n1605 3158185203 1054679659 Sagittal T1\n1606 3158185203 2946239560 Sagittal T2/STIR\n1607 3160528641 3183076389 Axial T2\n1608 3160528641 1845964212 Sagittal T1\n1609 3160528641 3180832116 Sagittal T2/STIR\n1610 3163594538 294771064 Axial T2\n1611 3163594538 15983577 Sagittal T1\n1612 3163594538 2752448967 Sagittal T2/STIR\n1613 3168755174 3241932744 Axial T2\n1614 3168755174 3335509714 Sagittal T1\n1615 3168755174 3299192635 Sagittal T2/STIR\n1616 3172703788 3357340646 Axial T2\n1617 3172703788 1681952265 Sagittal T1\n1618 3172703788 619632020 Sagittal T2/STIR","text/html":"
\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n
study_idseries_idseries_description
158931287951552985676628Axial T2
15903128795155483456514Sagittal T1
15913128795155246275868Sagittal T2/STIR
15923138242355684398187Axial T2
159331382423553310553683Sagittal T1
15943138242355992065261Sagittal T2/STIR
159531392745361038845969Axial T2
159631392745362171939378Sagittal T1
159731392745363906320584Sagittal T2/STIR
159831413315923267884718Axial T2
1599314133159272737584Sagittal T1
160031413315921244841079Sagittal T2/STIR
16013156269631501368228Axial T2
160231562696314030602643Sagittal T1
160331562696314059442027Sagittal T2/STIR
160431581852032163803231Axial T2
160531581852031054679659Sagittal T1
160631581852032946239560Sagittal T2/STIR
160731605286413183076389Axial T2
160831605286411845964212Sagittal T1
160931605286413180832116Sagittal T2/STIR
16103163594538294771064Axial T2
1611316359453815983577Sagittal T1
161231635945382752448967Sagittal T2/STIR
161331687551743241932744Axial T2
161431687551743335509714Sagittal T1
161531687551743299192635Sagittal T2/STIR
161631727037883357340646Axial T2
161731727037881681952265Sagittal T1
16183172703788619632020Sagittal T2/STIR
\n
"},"metadata":{}}]},{"cell_type":"code","source":"image_dir = \"/kaggle/input/rsna-2024-lumbar-spine-degenerative-classification/train_images\"\nstudy_id_dir = '/kaggle/input/rsna-2024-lumbar-spine-degenerative-classification/train_images/100206310'\nseries_id_dir = '/kaggle/input/rsna-2024-lumbar-spine-degenerative-classification/train_images/100206310/1012284084'\n\n# Example image size and number of stacked slices (depth for 3D)\nimg_width, img_height = 256, 256\ndepth = 50 # Number of 2D slices to stack","metadata":{"execution":{"iopub.status.busy":"2024-10-02T13:39:30.172139Z","iopub.execute_input":"2024-10-02T13:39:30.172607Z","iopub.status.idle":"2024-10-02T13:39:30.178722Z","shell.execute_reply.started":"2024-10-02T13:39:30.172562Z","shell.execute_reply":"2024-10-02T13:39:30.177408Z"},"trusted":true},"execution_count":8,"outputs":[]},{"cell_type":"code","source":"# import os\n# import numpy as np\n# import pandas as pd\n# import pydicom\n# from tensorflow.keras.preprocessing.image import img_to_array\n# from skimage.transform import resize\n\n# # Load 2D DICOM images and stack them into 3D arrays\n# def load_images(df_filtered, image_dir, img_width, img_height, depth):\n# images = []\n# labels = [] # Assuming you have some labels like 'series_description'\n\n# # Iterate over the rows in the filtered DataFrame\n# for _, row in df_filtered.iterrows():\n# study_id = str(row['study_id'])\n# series_id = str(row['series_id'])\n \n# # Construct the series directory path\n# series_id_dir = os.path.join(image_dir, study_id, series_id)\n \n# # Ensure the directory exists\n# if not os.path.exists(series_id_dir):\n# continue\n\n# # Collect DICOM files, sorted by their numeric order\n# dicom_files = sorted([f for f in os.listdir(series_id_dir) if f.endswith('.dcm')],\n# key=lambda x: int(x.split('.')[0])) # Sort by the number in the filename\n \n# stacked_images = []\n\n# # Load the first 'depth' number of DICOM files (or all if less than 'depth')\n# for i, dicom_file in enumerate(dicom_files[:depth]):\n# dicom_path = os.path.join(series_id_dir, dicom_file)\n# dicom_data = pydicom.dcmread(dicom_path)\n\n# # Convert the DICOM pixel array to a NumPy array and resize it\n# img_array = dicom_data.pixel_array\n# img_resized = resize(img_array, (img_width, img_height), preserve_range=True)\n# img_resized = img_to_array(img_resized)\n\n# stacked_images.append(img_resized)\n \n# # Stack the 2D slices to create a 3D array\n# if len(stacked_images) > 0:\n# stacked_images = np.stack(stacked_images, axis=-1) # Create 3D array\n \n# images.append(stacked_images)\n# labels.append(row['series_description']) # Add the label, assuming 'series_description' is the label\n \n# return np.array(images), np.array(labels)\n\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import os\nimport numpy as np\nimport pandas as pd\nimport pydicom\nfrom tensorflow.keras.preprocessing.image import img_to_array\nfrom skimage.transform import resize\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.preprocessing import OneHotEncoder\n","metadata":{"execution":{"iopub.status.busy":"2024-10-02T13:39:34.319554Z","iopub.execute_input":"2024-10-02T13:39:34.319986Z","iopub.status.idle":"2024-10-02T13:39:34.354407Z","shell.execute_reply.started":"2024-10-02T13:39:34.319946Z","shell.execute_reply":"2024-10-02T13:39:34.353244Z"},"trusted":true},"execution_count":9,"outputs":[]},{"cell_type":"code","source":"import pandas as pd\n\n# Map categories to integers\nlabel_map = {'Normal/Mild': 0.0, 'Moderate': 1.0, 'Severe': 2.0}\n\n# Convert targets in train.csv to numeric form\ndef prepare_labels(train_csv):\n # Replace the text labels with numbers for all target columns\n target_columns = train_csv.columns[1:] # Exclude 'study_id' column\n for col in target_columns:\n train_csv[col] = train_csv[col].map(label_map)\n \n return train_csv\n","metadata":{"execution":{"iopub.status.busy":"2024-10-02T13:39:36.214201Z","iopub.execute_input":"2024-10-02T13:39:36.215448Z","iopub.status.idle":"2024-10-02T13:39:36.222349Z","shell.execute_reply.started":"2024-10-02T13:39:36.215372Z","shell.execute_reply":"2024-10-02T13:39:36.221134Z"},"trusted":true},"execution_count":10,"outputs":[]},{"cell_type":"code","source":"train_csv = prepare_labels(df_train)\ntrain_csv","metadata":{"execution":{"iopub.status.busy":"2024-10-02T13:39:38.151638Z","iopub.execute_input":"2024-10-02T13:39:38.152203Z","iopub.status.idle":"2024-10-02T13:39:38.231899Z","shell.execute_reply.started":"2024-10-02T13:39:38.152146Z","shell.execute_reply":"2024-10-02T13:39:38.230524Z"},"trusted":true},"execution_count":11,"outputs":[{"execution_count":11,"output_type":"execute_result","data":{"text/plain":" study_id spinal_canal_stenosis_l1_l2 spinal_canal_stenosis_l2_l3 \\\n0 4646740 0.0 0.0 \n1 22191399 0.0 0.0 \n2 29931867 0.0 1.0 \n3 38281420 0.0 0.0 \n4 40745534 0.0 0.0 \n.. ... ... ... \n535 3158185203 0.0 1.0 \n536 3160528641 0.0 0.0 \n537 3163594538 2.0 2.0 \n538 3168755174 0.0 0.0 \n539 3172703788 1.0 0.0 \n\n spinal_canal_stenosis_l3_l4 spinal_canal_stenosis_l4_l5 \\\n0 1.0 2.0 \n1 0.0 2.0 \n2 1.0 0.0 \n3 0.0 1.0 \n4 1.0 2.0 \n.. ... ... \n535 0.0 0.0 \n536 0.0 1.0 \n537 2.0 2.0 \n538 2.0 0.0 \n539 0.0 0.0 \n\n spinal_canal_stenosis_l5_s1 left_neural_foraminal_narrowing_l1_l2 \\\n0 0.0 0.0 \n1 0.0 0.0 \n2 0.0 0.0 \n3 0.0 0.0 \n4 0.0 0.0 \n.. ... ... \n535 0.0 0.0 \n536 0.0 0.0 \n537 0.0 1.0 \n538 0.0 0.0 \n539 0.0 0.0 \n\n left_neural_foraminal_narrowing_l2_l3 \\\n0 0.0 \n1 0.0 \n2 0.0 \n3 0.0 \n4 0.0 \n.. ... \n535 1.0 \n536 0.0 \n537 1.0 \n538 0.0 \n539 0.0 \n\n left_neural_foraminal_narrowing_l3_l4 \\\n0 0.0 \n1 0.0 \n2 0.0 \n3 0.0 \n4 0.0 \n.. ... \n535 1.0 \n536 1.0 \n537 1.0 \n538 0.0 \n539 0.0 \n\n left_neural_foraminal_narrowing_l4_l5 ... \\\n0 1.0 ... \n1 0.0 ... \n2 1.0 ... \n3 1.0 ... \n4 1.0 ... \n.. ... ... \n535 2.0 ... \n536 1.0 ... \n537 1.0 ... \n538 1.0 ... \n539 0.0 ... \n\n left_subarticular_stenosis_l1_l2 left_subarticular_stenosis_l2_l3 \\\n0 0.0 0.0 \n1 0.0 0.0 \n2 0.0 0.0 \n3 0.0 0.0 \n4 0.0 0.0 \n.. ... ... \n535 0.0 1.0 \n536 0.0 0.0 \n537 2.0 2.0 \n538 NaN NaN \n539 0.0 0.0 \n\n left_subarticular_stenosis_l3_l4 left_subarticular_stenosis_l4_l5 \\\n0 0.0 2.0 \n1 0.0 1.0 \n2 1.0 1.0 \n3 0.0 0.0 \n4 1.0 2.0 \n.. ... ... \n535 1.0 2.0 \n536 1.0 1.0 \n537 2.0 2.0 \n538 2.0 0.0 \n539 0.0 0.0 \n\n left_subarticular_stenosis_l5_s1 right_subarticular_stenosis_l1_l2 \\\n0 0.0 0.0 \n1 1.0 0.0 \n2 1.0 1.0 \n3 0.0 0.0 \n4 1.0 0.0 \n.. ... ... \n535 0.0 0.0 \n536 0.0 0.0 \n537 0.0 1.0 \n538 0.0 NaN \n539 0.0 0.0 \n\n right_subarticular_stenosis_l2_l3 right_subarticular_stenosis_l3_l4 \\\n0 1.0 1.0 \n1 0.0 0.0 \n2 2.0 2.0 \n3 1.0 1.0 \n4 0.0 1.0 \n.. ... ... \n535 1.0 0.0 \n536 0.0 0.0 \n537 2.0 2.0 \n538 NaN 2.0 \n539 0.0 0.0 \n\n right_subarticular_stenosis_l4_l5 right_subarticular_stenosis_l5_s1 \n0 1.0 0.0 \n1 2.0 1.0 \n2 1.0 2.0 \n3 2.0 2.0 \n4 2.0 0.0 \n.. ... ... \n535 1.0 0.0 \n536 2.0 0.0 \n537 2.0 1.0 \n538 2.0 2.0 \n539 1.0 1.0 \n\n[540 rows x 26 columns]","text/html":"
\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n
study_idspinal_canal_stenosis_l1_l2spinal_canal_stenosis_l2_l3spinal_canal_stenosis_l3_l4spinal_canal_stenosis_l4_l5spinal_canal_stenosis_l5_s1left_neural_foraminal_narrowing_l1_l2left_neural_foraminal_narrowing_l2_l3left_neural_foraminal_narrowing_l3_l4left_neural_foraminal_narrowing_l4_l5...left_subarticular_stenosis_l1_l2left_subarticular_stenosis_l2_l3left_subarticular_stenosis_l3_l4left_subarticular_stenosis_l4_l5left_subarticular_stenosis_l5_s1right_subarticular_stenosis_l1_l2right_subarticular_stenosis_l2_l3right_subarticular_stenosis_l3_l4right_subarticular_stenosis_l4_l5right_subarticular_stenosis_l5_s1
046467400.00.01.02.00.00.00.00.01.0...0.00.00.02.00.00.01.01.01.00.0
1221913990.00.00.02.00.00.00.00.00.0...0.00.00.01.01.00.00.00.02.01.0
2299318670.01.01.00.00.00.00.00.01.0...0.00.01.01.01.01.02.02.01.02.0
3382814200.00.00.01.00.00.00.00.01.0...0.00.00.00.00.00.01.01.02.02.0
4407455340.00.01.02.00.00.00.00.01.0...0.00.01.02.01.00.00.01.02.00.0
..................................................................
53531581852030.01.00.00.00.00.01.01.02.0...0.01.01.02.00.00.01.00.01.00.0
53631605286410.00.00.01.00.00.00.01.01.0...0.00.01.01.00.00.00.00.02.00.0
53731635945382.02.02.02.00.01.01.01.01.0...2.02.02.02.00.01.02.02.02.01.0
53831687551740.00.02.00.00.00.00.00.01.0...NaNNaN2.00.00.0NaNNaN2.02.02.0
53931727037881.00.00.00.00.00.00.00.00.0...0.00.00.00.00.00.00.00.01.01.0
\n

540 rows × 26 columns

\n
"},"metadata":{}}]},{"cell_type":"code","source":"# import pandas as pd\n# from sklearn.preprocessing import OneHotEncoder\n\n# # Map categories to integers\n# label_map = {'Normal/Mild': 0, 'Moderate': 1, 'Severe': 2}\n\n# # Convert targets in train.csv to numeric form and then one-hot encode\n# def prepare_labels(train_csv):\n# # Replace the text labels with numbers\n# target_columns = train_csv.columns[1:] # Exclude 'study_id' column\n# for col in target_columns:\n# train_csv[col] = train_csv[col].map(label_map)\n \n# # Initialize OneHotEncoder\n# encoder = OneHotEncoder(categories=[0, 1, 2], sparse=False, drop=None)\n \n# # Apply OneHotEncoder to each target column and concatenate\n# encoded_targets = []\n# for col in target_columns:\n# y = encoder.fit_transform(train_csv[[col]])\n# encoded_df = pd.DataFrame(y, columns=[f\"{col}_{i}\" for i in range(y.shape[1])])\n# encoded_targets.append(encoded_df)\n \n# # Concatenate all encoded target DataFrames\n# y_df = pd.concat(encoded_targets, axis=1)\n \n# # Concatenate the study_id column with the one-hot encoded columns\n# train_csv_encoded = pd.concat([train_csv[['study_id']], y_df], axis=1)\n \n# return train_csv_encoded\n\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# import pandas as pd\n# from sklearn.preprocessing import OneHotEncoder\n\n# # Map categories to integers\n# label_map = {'Normal/Mild': 0, 'Moderate': 1, 'Severe': 2}\n\n# # Convert targets in train.csv to numeric form and then one-hot encode\n# def prepare_labels(train_csv):\n# # Replace the text labels with numbers\n# target_columns = train_csv.columns[1:] # Exclude 'study_id' column\n# for col in target_columns:\n# train_csv[col] = train_csv[col].map(label_map)\n \n# # One-hot encode the target columns\n# encoder = OneHotEncoder(sparse=False)\n# y = encoder.fit_transform(train_csv[target_columns])\n \n# # Create a DataFrame from the one-hot encoded array\n# y_df = pd.DataFrame(y, columns=encoder.get_feature_names_out(target_columns))\n \n# # Concatenate the study_id column with the one-hot encoded columns\n# train_csv_encoded = pd.concat([train_csv[['study_id']], y_df], axis=1)\n \n# return train_csv_encoded\n\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# import pandas as pd\n\n# # Define the mapping from categories to integers\n# label_map = {'Normal/Mild': 0, 'Moderate': 1, 'Severe': 2}\n\n# # Function to prepare labels with integer encoding\n# def prepare_labels(train_csv):\n# # Replace text labels with their corresponding numbers\n# target_columns = train_csv.columns[1:] # Exclude 'study_id' column\n# for col in target_columns:\n# train_csv[col] = train_csv[col].map(label_map)\n \n# return train_csv\n\n# # Load the data\n# df = pd.read_csv('/kaggle/input/rsna-2024-lumbar-spine-degenerative-classification/train_series_descriptions.csv')\n# train = pd.read_csv('/kaggle/input/rsna-2024-lumbar-spine-degenerative-classification/train.csv')\n\n# # Prepare the labels with integer encoding\n# train_encoded = prepare_labels(train)\n\n# # Save or inspect the encoded DataFrame\n# train_encoded.head()\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"\n# Now series_ids_ordered contains the series_ids in the order defined by custom_order\n# Use these ordered series_ids to loop through the series folders\n\n\n","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import numpy as np\nimport os\nimport pydicom\nfrom sklearn.preprocessing import OneHotEncoder\nfrom skimage.transform import resize\nfrom tensorflow.keras.preprocessing.image import img_to_array\n\n# Define the function to load 2D DICOM images and stack them into 3D arrays with consistent depth\ndef load_images(df_filtered, train_csv, image_dir, img_width, img_height, depth):\n images = []\n labels = [] # Targets from the 25 columns\n\n # Iterate over the rows in the filtered DataFrame\n for _, row in df_filtered.iterrows():\n study_id = str(row['study_id'])\n \n # Match the study_id from df_filtered with the corresponding study_id in train_csv\n target_row = train_csv[train_csv['study_id'] == int(study_id)]\n \n if target_row.empty:\n print(f\"No matching targets found for study_id: {study_id}\")\n continue\n\n # Construct the series directory path (assuming df_filtered has the correct structure)\n series_id = str(row['series_id'])\n series_id_dir = os.path.join(image_dir, study_id, series_id)\n \n # Ensure the directory exists\n if not os.path.exists(series_id_dir):\n print(f\"Directory not found: {series_id_dir}\")\n continue\n\n # Collect DICOM files, sorted by their numeric order\n dicom_files = sorted([f for f in os.listdir(series_id_dir) if f.endswith('.dcm')],\n key=lambda x: int(x.split('.')[0])) # Sort by the number in the filename\n \n if len(dicom_files) == 0:\n print(f\"No DICOM files found in directory: {series_id_dir}\")\n continue\n\n stacked_images = []\n\n # Load the DICOM files, resizing and stacking them\n for dicom_file in dicom_files[:depth]: # Limit to 'depth' slices\n dicom_path = os.path.join(series_id_dir, dicom_file)\n dicom_data = pydicom.dcmread(dicom_path)\n\n # Convert the DICOM pixel array to a NumPy array and resize it\n img_array = dicom_data.pixel_array\n img_resized = resize(img_array, (img_width, img_height), preserve_range=True)\n img_resized = img_to_array(img_resized)\n\n stacked_images.append(img_resized)\n \n # Handle case where the number of slices is less than 'depth'\n if len(stacked_images) < depth:\n # Pad with zeros to maintain consistent depth\n padding_needed = depth - len(stacked_images)\n padding = [np.zeros((img_width, img_height, 1))] * padding_needed\n stacked_images.extend(padding)\n \n # Stack the 2D slices to create a 3D array\n stacked_images = np.stack(stacked_images, axis=-1) # Create 3D array\n\n images.append(stacked_images)\n \n # Append the target for this study_id\n labels.append(target_row.iloc[0, 1:].values) # Get the 25 targets (already numeric)\n\n return np.array(images), np.array(labels)\n\n\n","metadata":{"execution":{"iopub.status.busy":"2024-10-02T13:40:20.423309Z","iopub.execute_input":"2024-10-02T13:40:20.424648Z","iopub.status.idle":"2024-10-02T13:40:20.441248Z","shell.execute_reply.started":"2024-10-02T13:40:20.42459Z","shell.execute_reply":"2024-10-02T13:40:20.439945Z"},"trusted":true},"execution_count":12,"outputs":[]},{"cell_type":"code","source":"# Example usage:\nimg_width, img_height = 128, 128\ndepth = 50 # Number of slices per 3D image\n\ny_all = train_csv # This converts labels into one-hot encoding\n\n# Assuming df_filtered is already loaded with 'study_id', 'series_id', and 'series_description'\nX, y = load_images(df_filtered, y_all, image_dir=\"/kaggle/input/rsna-2024-lumbar-spine-degenerative-classification/train_images\", \n img_width=img_width, img_height=img_height, depth=depth)\n\nX = np.squeeze(X, axis=-2) # Removes the 4th dimension\n\n# Check if images and labels are loaded correctly\nprint(f\"Number of images loaded: {len(X)}\")\nprint(f\"Number of labels loaded: {len(y)}\")\n\nprint(X.shape)\nprint(y.shape)","metadata":{"execution":{"iopub.status.busy":"2024-10-02T13:40:44.709115Z","iopub.execute_input":"2024-10-02T13:40:44.709608Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Ensure X has the correct shape (num_samples, img_width, img_height, depth, 1)\n#X = np.expand_dims(X, axis=-1) # Adding channel dimension\n\n# Ensure y is one-hot encoded and has shape (num_samples, num_classes)\nprint(X.shape) # Should be (num_samples, img_width, img_height, depth, 1)\nprint(y.shape) # Should be (num_samples, num_classes)\n#Remove the extra dimension (channel = 1) from the data\n#X = np.squeeze(X, axis=-2) # Removes the 4th dimension\n\n# Now X will have the shape (batch_size, img_width, img_height, depth)\nprint(X.shape) # Should be (num_samples, img_width, img_height, depth, 1)\nprint(y.shape) # Should be (num_samples, num_classes)\n\n# Proceed with normalization and train-test split\nX = X / 255.0\n\n# Train-test split\nX_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)\n\n# Now X_train and X_test should have the correct shape for your CNN","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"del df_train, df_filtered","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# del X, y, y_all","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from tensorflow.keras.models import Sequential\nfrom tensorflow.keras.layers import Conv3D, MaxPooling3D, Flatten, Dense, Dropout\n\ndef create_cnn_model(input_shape, num_classes):\n model = Sequential()\n model.add(Conv3D(32, (3, 3, 3), activation='relu', input_shape=input_shape))\n model.add(MaxPooling3D(pool_size=(2, 2, 2)))\n model.add(Conv3D(64, (3, 3, 3), activation='relu'))\n model.add(MaxPooling3D(pool_size=(2, 2, 2)))\n model.add(Conv3D(128, (3, 3, 3), activation='relu'))\n model.add(MaxPooling3D(pool_size=(2, 2, 2)))\n model.add(Flatten())\n model.add(Dense(128, activation='relu'))\n model.add(Dense(128, activation='relu'))\n model.add(Dense(128, activation='relu'))\n model.add(Dropout(0.2))\n model.add(Dense(num_classes, activation='softmax'))\n \n model.compile(\n loss='categorical_crossentropy',\n optimizer='adam',\n metrics=['accuracy']\n )\n \n return model\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from tensorflow.keras import layers, models\nfrom tensorflow.keras.applications import EfficientNetB0\nfrom tensorflow.keras.layers import Conv3D, MaxPooling3D, Flatten, Dense, Dropout, GlobalAveragePooling2D, Input\n\ndef model(input_shape, num_classes):\n # Use EfficientNetB0 as the backbone for feature extraction\n efficientnet_base = EfficientNetB0(\n include_top=False, # Exclude the fully connected layer\n input_shape=input_shape[:2] + (3,), # Using only the 2D input with 3 channels for EfficientNet\n weights='imagenet' # Pre-trained weights on ImageNet\n )\n \n # Input layer\n inputs = Input(shape=input_shape)\n \n # Apply 3D convolution on the input data to convert it to a 2D EfficientNet-compatible shape\n x = Conv3D(3, (3, 3, 3), activation='relu', padding='same')(inputs)\n x = MaxPooling3D(pool_size=(1, 1, input_shape[2]))(x) # Reduce the depth to 1 for EfficientNet input\n\n # Reshape for EfficientNet compatibility\n x = layers.Reshape(target_shape=(input_shape[0], input_shape[1], 3))(x)\n \n # Pass through EfficientNet backbone\n x = efficientnet_base(x)\n \n # Global average pooling and fully connected layers\n x = GlobalAveragePooling2D()(x)\n x = Dense(256, activation='relu')(x)\n x = Dropout(0.5)(x)\n \n # Output layer\n outputs = Dense(num_classes, activation='softmax')(x)\n \n model = models.Model(inputs=inputs, outputs=outputs)\n \n model.compile(\n loss='categorical_crossentropy',\n optimizer='adam',\n metrics=['accuracy']\n )\n \n return model\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import keras\nfrom tensorflow.keras.applications import EfficientNetB0\nfrom keras.applications.resnet50 import ResNet50\nfrom keras.layers import Conv2D, MaxPooling2D, GlobalAveragePooling2D, Dense, Dropout, Input, Reshape, Flatten, Concatenate\nfrom keras.models import Model\n# from keras.applications import ResNet18\n\ndef create_cnn_model(input_shape, num_classes):\n # Input layer for the 3D CNN\n inputs = Input(shape=input_shape) # Input shape: (256, 256, 50, 1)\n \n # Reshape to (50, 256, 256, 1), treat the depth as the batch size for ResNet50\n reshaped_input = Reshape((input_shape[2], input_shape[0], input_shape[1], input_shape[3]))(inputs) # (depth, width, height, channels)\n \n # Load the ResNet50 base model without the top layers\n resnet_base = EfficientNetB0(weights='imagenet', include_top=False, input_shape=(input_shape[0], input_shape[1], 3))\n \n # Initialize a list to collect the outputs for each slice\n feature_maps = []\n \n for i in range(input_shape[2]): # For each slice in the depth dimension\n # Extract the slice\n slice_input = reshaped_input[:, i, :, :, :]\n \n # Convert 1 channel to 3 channels by repeating it 3 times\n slice_input_rgb = Concatenate()([slice_input] * 3)\n \n # Apply ResNet50 to each slice\n slice_features = resnet_base(slice_input_rgb)\n \n # Global Average Pooling after ResNet50 to reduce the feature maps\n slice_features = GlobalAveragePooling2D()(slice_features) # (None, feature_maps)\n \n # Add the feature map for this slice to the list\n feature_maps.append(slice_features)\n \n # Concatenate all feature maps from each slice\n x = Concatenate()(feature_maps) # Shape: (None, depth * feature_maps)\n \n # Dense and Dropout layers for classification\n x = Dense(256, activation='relu')(x)\n x = Dropout(0.3)(x)\n outputs = Dense(num_classes, activation='softmax')(x)\n \n # Define the final model\n model = Model(inputs, outputs)\n \n # Compile the model\n model.compile(\n loss='categorical_crossentropy',\n optimizer='adam',\n metrics=['accuracy']\n )\n \n return model\n\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from tensorflow.keras.layers import Conv2D, MaxPooling2D, Conv2DTranspose, Concatenate, Input, Dropout, UpSampling2D\nfrom tensorflow.keras.models import Model\n\ndef create_unet_model(input_shape, num_classes):\n inputs = Input(shape=input_shape) # Input shape: (256, 256, 50, 1)\n \n # Encoding Path (Downsampling)\n conv1 = Conv2D(64, (3, 3), activation='relu', padding='same')(inputs)\n conv1 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv1)\n pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)\n\n conv2 = Conv2D(128, (3, 3), activation='relu', padding='same')(pool1)\n conv2 = Conv2D(128, (3, 3), activation='relu', padding='same')(conv2)\n pool2 = MaxPooling2D(pool_size=(2, 2))(conv2)\n\n conv3 = Conv2D(256, (3, 3), activation='relu', padding='same')(pool2)\n conv3 = Conv2D(256, (3, 3), activation='relu', padding='same')(conv3)\n pool3 = MaxPooling2D(pool_size=(2, 2))(conv3)\n\n conv4 = Conv2D(512, (3, 3), activation='relu', padding='same')(pool3)\n conv4 = Conv2D(512, (3, 3), activation='relu', padding='same')(conv4)\n pool4 = MaxPooling2D(pool_size=(2, 2))(conv4)\n\n conv5 = Conv2D(1024, (3, 3), activation='relu', padding='same')(pool4)\n conv5 = Conv2D(1024, (3, 3), activation='relu', padding='same')(conv5)\n drop5 = Dropout(0.5)(conv5)\n\n # Decoding Path (Upsampling)\n up6 = UpSampling2D(size=(2, 2))(drop5)\n up6 = Conv2D(512, (2, 2), activation='relu', padding='same')(up6)\n merge6 = Concatenate()([conv4, up6])\n conv6 = Conv2D(512, (3, 3), activation='relu', padding='same')(merge6)\n conv6 = Conv2D(512, (3, 3), activation='relu', padding='same')(conv6)\n\n up7 = UpSampling2D(size=(2, 2))(conv6)\n up7 = Conv2D(256, (2, 2), activation='relu', padding='same')(up7)\n merge7 = Concatenate()([conv3, up7])\n conv7 = Conv2D(256, (3, 3), activation='relu', padding='same')(merge7)\n conv7 = Conv2D(256, (3, 3), activation='relu', padding='same')(conv7)\n\n up8 = UpSampling2D(size=(2, 2))(conv7)\n up8 = Conv2D(128, (2, 2), activation='relu', padding='same')(up8)\n merge8 = Concatenate()([conv2, up8])\n conv8 = Conv2D(128, (3, 3), activation='relu', padding='same')(merge8)\n conv8 = Conv2D(128, (3, 3), activation='relu', padding='same')(conv8)\n\n up9 = UpSampling2D(size=(2, 2))(conv8)\n up9 = Conv2D(64, (2, 2), activation='relu', padding='same')(up9)\n merge9 = Concatenate()([conv1, up9])\n conv9 = Conv2D(64, (3, 3), activation='relu', padding='same')(merge9)\n conv9 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv9)\n\n # Final output layer\n outputs = Conv2D(num_classes, (1, 1), activation='softmax')(conv9)\n\n # Define the model\n model = Model(inputs, outputs)\n\n # Compile the model\n model.compile(\n loss='categorical_crossentropy',\n optimizer='adam',\n metrics=['accuracy']\n )\n\n return model\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from tensorflow.keras.layers import Conv3D, MaxPooling3D, Conv3DTranspose, Concatenate, Input, Dropout, UpSampling3D\nfrom tensorflow.keras.models import Model\n\ndef create_unet_model(input_shape, num_classes):\n inputs = Input(shape=input_shape) # Input shape: (256, 256, 50, 1) for 3D grayscale\n\n # Encoding Path (Downsampling)\n conv1 = Conv3D(64, (3, 3, 3), activation='relu', padding='same')(inputs)\n conv1 = Conv3D(64, (3, 3, 3), activation='relu', padding='same')(conv1)\n pool1 = MaxPooling3D(pool_size=(2, 2, 2))(conv1)\n\n conv2 = Conv3D(128, (3, 3, 3), activation='relu', padding='same')(pool1)\n conv2 = Conv3D(128, (3, 3, 3), activation='relu', padding='same')(conv2)\n pool2 = MaxPooling3D(pool_size=(2, 2, 2))(conv2)\n\n conv3 = Conv3D(256, (3, 3, 3), activation='relu', padding='same')(pool2)\n conv3 = Conv3D(256, (3, 3, 3), activation='relu', padding='same')(conv3)\n pool3 = MaxPooling3D(pool_size=(2, 2, 2))(conv3)\n\n conv4 = Conv3D(512, (3, 3, 3), activation='relu', padding='same')(pool3)\n conv4 = Conv3D(512, (3, 3, 3), activation='relu', padding='same')(conv4)\n pool4 = MaxPooling3D(pool_size=(2, 2, 2))(conv4)\n\n conv5 = Conv3D(1024, (3, 3, 3), activation='relu', padding='same')(pool4)\n conv5 = Conv3D(1024, (3, 3, 3), activation='relu', padding='same')(conv5)\n drop5 = Dropout(0.5)(conv5)\n\n # Decoding Path (Upsampling)\n up6 = UpSampling3D(size=(2, 2, 2))(drop5)\n up6 = Conv3D(512, (2, 2, 2), activation='relu', padding='same')(up6)\n merge6 = Concatenate()([conv4, up6])\n conv6 = Conv3D(512, (3, 3, 3), activation='relu', padding='same')(merge6)\n conv6 = Conv3D(512, (3, 3, 3), activation='relu', padding='same')(conv6)\n\n up7 = UpSampling3D(size=(2, 2, 2))(conv6)\n up7 = Conv3D(256, (2, 2, 2), activation='relu', padding='same')(up7)\n merge7 = Concatenate()([conv3, up7])\n conv7 = Conv3D(256, (3, 3, 3), activation='relu', padding='same')(merge7)\n conv7 = Conv3D(256, (3, 3, 3), activation='relu', padding='same')(conv7)\n\n up8 = UpSampling3D(size=(2, 2, 2))(conv7)\n up8 = Conv3D(128, (2, 2, 2), activation='relu', padding='same')(up8)\n merge8 = Concatenate()([conv2, up8])\n conv8 = Conv3D(128, (3, 3, 3), activation='relu', padding='same')(merge8)\n conv8 = Conv3D(128, (3, 3, 3), activation='relu', padding='same')(conv8)\n\n up9 = UpSampling3D(size=(2, 2, 2))(conv8)\n up9 = Conv3D(64, (2, 2, 2), activation='relu', padding='same')(up9)\n merge9 = Concatenate()([conv1, up9])\n conv9 = Conv3D(64, (3, 3, 3), activation='relu', padding='same')(merge9)\n conv9 = Conv3D(64, (3, 3, 3), activation='relu', padding='same')(conv9)\n\n # Final output layer\n outputs = Conv3D(num_classes, (1, 1, 1), activation='softmax')(conv9)\n\n # Define the model\n model = Model(inputs, outputs)\n\n # Compile the model\n model.compile(\n loss='categorical_crossentropy',\n optimizer='adam',\n metrics=['accuracy']\n )\n\n return model\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from tensorflow.keras.layers import Conv3D, MaxPooling3D, Conv3DTranspose, Concatenate, Input, Dropout, UpSampling3D, ZeroPadding3D\nfrom tensorflow.keras.models import Model\nfrom tensorflow.keras.layers import GlobalAveragePooling3D, Dense\n\n\ndef create_unet_model(input_shape, num_classes):\n inputs = Input(shape=input_shape) # Input shape: (256, 256, 50, 1) for 3D grayscale\n\n # Encoding Path (Downsampling)\n conv1 = Conv3D(64, (3, 3, 3), activation='relu', padding='same')(inputs)\n conv1 = Conv3D(64, (3, 3, 3), activation='relu', padding='same')(conv1)\n pool1 = MaxPooling3D(pool_size=(2, 2, 2))(conv1)\n\n conv2 = Conv3D(128, (3, 3, 3), activation='relu', padding='same')(pool1)\n conv2 = Conv3D(128, (3, 3, 3), activation='relu', padding='same')(conv2)\n pool2 = MaxPooling3D(pool_size=(2, 2, 2))(conv2)\n\n conv3 = Conv3D(256, (3, 3, 3), activation='relu', padding='same')(pool2)\n conv3 = Conv3D(256, (3, 3, 3), activation='relu', padding='same')(conv3)\n pool3 = MaxPooling3D(pool_size=(2, 2, 2))(conv3)\n\n conv4 = Conv3D(512, (3, 3, 3), activation='relu', padding='same')(pool3)\n conv4 = Conv3D(512, (3, 3, 3), activation='relu', padding='same')(conv4)\n pool4 = MaxPooling3D(pool_size=(2, 2, 2))(conv4)\n\n conv5 = Conv3D(1024, (3, 3, 3), activation='relu', padding='same')(pool4)\n conv5 = Conv3D(1024, (3, 3, 3), activation='relu', padding='same')(conv5)\n drop5 = Dropout(0.5)(conv5)\n\n # Decoding Path (Upsampling)\n up6 = UpSampling3D(size=(2, 2, 2))(drop5)\n up6 = Conv3D(512, (2, 2, 2), activation='relu', padding='same')(up6)\n merge6 = Concatenate()([conv4, up6])\n conv6 = Conv3D(512, (3, 3, 3), activation='relu', padding='same')(merge6)\n conv6 = Conv3D(512, (3, 3, 3), activation='relu', padding='same')(conv6)\n\n up7 = UpSampling3D(size=(2, 2, 2))(conv6)\n up7 = Conv3D(256, (2, 2, 2), activation='relu', padding='same')(up7)\n merge7 = Concatenate()([conv3, up7])\n conv7 = Conv3D(256, (3, 3, 3), activation='relu', padding='same')(merge7)\n conv7 = Conv3D(256, (3, 3, 3), activation='relu', padding='same')(conv7)\n\n up8 = UpSampling3D(size=(2, 2, 2))(conv7)\n up8 = Conv3D(128, (2, 2, 2), activation='relu', padding='same')(up8)\n up8 = ZeroPadding3D(padding=((0, 0), (0, 0), (0, 1)))(up8) # Adjust depth\n merge8 = Concatenate()([conv2, up8])\n conv8 = Conv3D(128, (3, 3, 3), activation='relu', padding='same')(merge8)\n conv8 = Conv3D(128, (3, 3, 3), activation='relu', padding='same')(conv8)\n\n up9 = UpSampling3D(size=(2, 2, 2))(conv8)\n up9 = Conv3D(64, (2, 2, 2), activation='relu', padding='same')(up9)\n merge9 = Concatenate()([conv1, up9])\n conv9 = Conv3D(64, (3, 3, 3), activation='relu', padding='same')(merge9)\n conv9 = Conv3D(64, (3, 3, 3), activation='relu', padding='same')(conv9)\n\n # Final output layer\n outputs = Conv3D(num_classes, (1, 1, 1), activation='softmax')(conv9)\n outputs = GlobalAveragePooling3D()(outputs) # Pool to a single value per class\n outputs = Dense(num_classes, activation='softmax')(outputs)\n\n # Define the model\n model = Model(inputs, outputs)\n \n # Use sparse_categorical_crossentropy since we expect integer labels for each voxel\n model.compile(\n loss='sparse_categorical_crossentropy',\n optimizer='adam',\n metrics=['accuracy']\n )\n\n return model\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# from tensorflow.keras.models import Sequential\n# from tensorflow.keras.layers import Conv3D, MaxPooling3D, Flatten, Dense, Dropout, GlobalAveragePooling3D\n\n# def create_cnn_model(input_shape, num_classes):\n# model = Sequential()\n \n# # First Conv3D layer\n# model.add(Conv3D(32, (3, 3, 3), activation='relu', input_shape=input_shape))\n# model.add(MaxPooling3D(pool_size=(2, 2, 2)))\n \n# # Flatten the 3D output before feeding into a Dense layer\n# model.add(Flatten())\n# model.add(Dense(64, activation='relu')) # Dense layer after the first Conv3D\n \n# # Second Conv3D layer\n# model.add(Conv3D(64, (3, 3, 3), activation='relu'))\n# model.add(MaxPooling3D(pool_size=(2, 2, 2)))\n \n# # Flatten the 3D output before feeding into a Dense layer\n# model.add(Flatten())\n# model.add(Dense(128, activation='relu')) # Dense layer after the second Conv3D\n \n# # Third Conv3D layer\n# model.add(Conv3D(128, (3, 3, 3), activation='relu'))\n# model.add(MaxPooling3D(pool_size=(2, 2, 2)))\n\n# # Global average pooling (to reduce dimensions)\n# model.add(GlobalAveragePooling3D())\n \n# # Fully connected layers after all Conv3D layers\n# model.add(Dense(256, activation='relu'))\n# model.add(Dropout(0.5))\n# model.add(Dense(num_classes, activation='softmax'))\n \n# model.compile(\n# loss='categorical_crossentropy',\n# optimizer='adam',\n# metrics=['accuracy']\n# )\n \n# return model\n","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from tensorflow.keras.models import Sequential\nfrom tensorflow.keras.layers import Input, Dense, Conv2D, MaxPooling2D, Dropout, Flatten\nfrom tensorflow.keras.metrics import categorical_accuracy","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from tensorflow.keras import layers, models\nfrom tensorflow.keras.applications import ResNet50\nfrom tensorflow.keras.layers import Conv3D, MaxPooling3D, Flatten, Dense, Dropout, GlobalAveragePooling2D, Input\nfrom tensorflow.keras.applications import ResNet50\nfrom tensorflow.keras import backend as K\n\ndef build_resnet18_base(input_shape):\n # Load ResNet50 as a base model (since ResNet18 is not directly available)\n base_model = ResNet50(\n include_top=False, # Exclude the fully connected layer\n input_shape=input_shape[:2] + (3,), # Using only the 2D input with 3 channels for ResNet50\n weights='imagenet' # Pre-trained weights on ImageNet\n )\n \n return base_model\n\ndef model(input_shape, num_classes):\n # Build ResNet50 base model\n resnet_base = build_resnet18_base(input_shape)\n \n # Input layer\n inputs = Input(shape=input_shape)\n \n # Apply 3D convolution on the input data to convert it to a 2D ResNet-compatible shape\n x = Conv3D(3, (3, 3, 3), activation='relu', padding='same')(inputs)\n x = MaxPooling3D(pool_size=(1, 1, input_shape[2]))(x) # Reduce the depth to 1 for ResNet input\n\n # Reshape for ResNet compatibility\n x = layers.Reshape(target_shape=(input_shape[0], input_shape[1], 3))(x)\n \n # Pass through ResNet backbone\n x = resnet_base(x)\n \n # Global average pooling and fully connected layers\n x = GlobalAveragePooling2D()(x)\n x = Dense(256, activation='relu')(x)\n x = Dropout(0.2)(x)\n \n # Output layer\n outputs = Dense(num_classes, activation='softmax')(x)\n \n model = models.Model(inputs=inputs, outputs=outputs)\n \n model.compile(\n loss='categorical_crossentropy',\n optimizer='adam',\n metrics=['accuracy']\n )\n \n return model\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Create and train the model\ninput_shape = (img_width, img_height, depth, 1) # Adjusted for 3D input\nnum_classes = 25\nmodel = model(input_shape, num_classes)\n\n# Train the model\nmodel.fit(X_train, y_train, epochs=5, batch_size=8, validation_data=(X_test, y_test))\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model.summary()","metadata":{"trusted":true},"execution_count":null,"outputs":[]}]}