Merge branch 'main' into main

evaluate_main.py

@@ -0,0 +1,52 @@
+import sys
+import yaml
+from osl_dynamics.config_api.batch import IndexParser, BatchTrain, batch_check
+
+def main(index,config_path,analysis_config_path=None):
+    '''
+    This is the main function of all the analysis.
+    Parameters
+    ----------
+    index: int
+        index == -1 represents post training checkk or analysis.
+    config_path: str
+        where read the config path.
+    analysis_config_path: str
+        if analysis_config_path is not None, and index == -1
+        then implement the analysis code.
+    '''
+    with open(config_path, 'r') as file:
+        config_batch = yaml.safe_load(file)
+    if index > -1:
+        if isinstance(config_batch,list):
+            with open(f'{config_batch[index]}batch_config.yaml','r') as file:
+                config = yaml.safe_load(file)
+        else:
+            index_parser = IndexParser(config_batch)
+            config = index_parser.parse(index)
+        batch_train = BatchTrain(config)
+        batch_train.model_train()
+    else:
+        # Step 1: batch check whether training is successful
+        #          return the list where training is not successful
+        batch_check(config_batch)
+
+        # Step 2: if analysis_config_path is not None, implement the analayis code.
+        if analysis_config_path is not None:
+            pass
+
+
+
+
+if __name__ == '__main__':
+    index = int(sys.argv[1]) - 1
+    config_path = sys.argv[2]
+
+    if len(sys.argv) > 3:
+        analysis_config_path = sys.argv[3]
+    else:
+        analysis_config_path = None
+
+    main(index,config_path,analysis_config_path)
+
+

examples/fmri/visualisation/.ipynb_checkpoints/Demo_fmri_visualisation-checkpoint.ipynb

@@ -0,0 +1,191 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "61de8eb3",
+   "metadata": {},
+   "source": [
+    "# Tutorial of fMRI state visualisation using workbench"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "e8df7865",
+   "metadata": {},
+   "source": [
+    "Author: Yiming Wei\n",
+    "\n",
+    "Date: 22nd January 2024"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "c2704308",
+   "metadata": {},
+   "source": [
+    "The aim of this tutorial is to provide a first example of fMRI data visualisation using osl-dynamics toolbox within OHBA analysis group. We will utilise the HCP workbench and tools in osl_dynamics.analysis. This tutorial should enable us (Chet, Ben, Brian and myself) to compare dynamic FC models (HMM, Dynemo and mDynemo) trained on different large-scale datasets (HCP, UKBiobank and Cam-CAN). Several prerequisites are listed below:\n",
+    "\n",
+    "1. BMRC cluster. \n",
+    "\n",
+    "We need to set up workbench properly on BMRC cluster so that `wb_command` can be run properly in the command line. The easiest way is to run `module load ConnectomeWorkbench` before starting this tutorial. (TODO: It seems to me that this command does not work properly if run in jupyter notebook or python subprocess. Why?).\n",
+    "P.S. There's a `setup` function in `osl_dynamics.analysis.workbench.py` to setup workbench, which requires to specify the workbench path. However, the BMRC CPUs work on two architectures (refer to this [page](https://www.medsci.ox.ac.uk/for-staff/resources/bmrc/python-on-the-bmrc-cluster) for more info), which means that the path is dependent on the actual node. (Maybe different ideas?).\n",
+    "\n",
+    "2. Brain surface map.\n",
+    "\n",
+    "By default, osl-dynamics uses ParcellationPilot. HCP S1200 map, which is available on the HCP database, might be another option.\n",
+    "\n",
+    "3. Spatial map.\n",
+    "Obtained from group-level spatial ICA. Should be specific to the dataset.\n",
+    "HCP: surface map.\n",
+    "UKB: was volumetric, but now surface map - I'll email Fidel.\n",
+    "CAM-CAN: not sure.\n",
+    "\n",
+    "4. State mean activation and FC map.\n",
+    "These are the parameters in the observation model. For example, if we train HMM model with K states on ICA results with N channels, the state mean activation should be a numpy array with shape K * N, the FC map should be a 3D numpy array with shape K * N * N."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "8646ad8d",
+   "metadata": {},
+   "source": [
+    "We give a simple example below. Note some of the functions are in `rotation/utils.py`, which is my own function toolbox."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "id": "bd6720c5",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import numpy as np\n",
+    "import matplotlib.pyplot as plt\n",
+    "import nibabel as nib\n",
+    "from rotation.utils import IC2surface,first_eigenvector\n",
+    "from osl_dynamics.analysis.workbench import render"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "id": "ca0845c9",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Specify the directories\n",
+    "spatial_map_dir = '../../../data/spatial_maps/groupICA_3T_HCP1200_MSMAll_d25.ica/melodic_IC.dscalar.nii'\n",
+    "mean_activation_dir = '../../../results_202310/HMM_ICA_25_state_8/state_means.npy'\n",
+    "fc_dir = '../../../results_202310/HMM_ICA_25_state_8/state_correlations.npy'\n",
+    "save_dir ='./'"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 12,
+   "id": "09731120",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Load files\n",
+    "spatial_surface_map = nib.load(spatial_map_dir)\n",
+    "state_means = np.load(mean_activation_dir)\n",
+    "correlations = np.load(fc_dir)\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 13,
+   "id": "b773c4ef",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Calculate surface map of mean activation\n",
+    "mean_activation_surface_map = IC2surface(spatial_surface_map,state_means.T)\n",
+    "mean_activation_surface_map.to_filename(f'{save_dir}mean_activation_surface_map.dscalar.nii')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 14,
+   "id": "1ed26bc5",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Rank-one approximation\n",
+    "r1_approxs = []\n",
+    "for i in range(len(correlations)):\n",
+    "    correlation = correlations[i,:,:]\n",
+    "    r1_approxs.append(first_eigenvector(correlation))\n",
+    "r1_approxs = np.array(r1_approxs)\n",
+    "np.save(f'{save_dir}r1_approx_FC.npy', r1_approxs)\n",
+    "\n",
+    "# Calculate surface map of FC\n",
+    "FC_degree_surface_map = IC2surface(spatial_surface_map,r1_approxs.T)\n",
+    "FC_degree_surface_map.to_filename(f'{save_dir}FC_surface_map.dscalar.nii')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 17,
+   "id": "4614726b",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Saving images: 100%|██████████████████████████████| 8/8 [00:09<00:00,  1.15s/it]\n",
+      "Saving images: 100%|██████████████████████████████| 8/8 [00:07<00:00,  1.02it/s]\n"
+     ]
+    }
+   ],
+   "source": [
+    "# Render\n",
+    "render(nii=f'{save_dir}mean_activation_surface_map.dscalar.nii',\n",
+    "       save_dir=f'{save_dir}/visualisation',\n",
+    "       gui=False,\n",
+    "       inflation=0,\n",
+    "       image_name='./visualisation/mean_mid_thickness',\n",
+    "       input_cifti=True\n",
+    "       )\n",
+    "render(nii=f'{save_dir}FC_surface_map.dscalar.nii',\n",
+    "       save_dir=f'{save_dir}/visualisation',\n",
+    "       gui=False,\n",
+    "       inflation=0,\n",
+    "       image_name='./visualisation/FC_mid_thickness',\n",
+    "       input_cifti=True\n",
+    "       )"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "ef3c0bdd",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.10"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}