def __init__(self, main='', acpc='', mni_aff='', mni_f3d=''):
        # electrodes
        self.N = 0
        self.plan = {
            'patient': None,
            'acpc': None,
            'mni_aff': None,
            'mni_f3d': None
        }
        self.impl = {
            'patient': None,
            'acpc': None,
            'mni_aff': None,
            'mni_f3d': None
        }
        self.ghost = {
            'patient': None,
            'acpc': None,
            'mni_aff': None,
            'mni_f3d': None
        }

        # sub folders
        self.dir = main
        self.acpc = acpc
        self.mni_aff = mni_aff
        self.mni_f3d = mni_f3d

        self.filesystem = FileSystem.FileSystem()
        self.xmlparser = XMLParser.XMLParser()
    def __init__(self, images=None, templates=None):
        self.medicalImages = images
        self.templates = templates

        self.registration = NiftyReg.NiftyReg()
        self.filesystem = FileSystem.FileSystem()
        self.xmlparser = XMLParser.XMLParser()
        self.robex = Robex.Robex()
Example #3
0
    def __init__(self, images=None, electrodes=None, space=None):
        self.medicalImages = images
        self.electrodes = electrodes
        self.space = space
        self.directory = self.medicalImages.get_directory(space=space)
        self.stl_folder = os.path.join(self.directory, 'stl')

        self.xmlparser = XMLParser.XMLParser()

        self.surfaces = Surfaces.Surfaces()
Example #4
0
    def __init__(self, images=None, electrodes=None):
        self.medicalImages = images
        self.electrodes = electrodes

        self.filesystem = FileSystem.FileSystem()
        self.xmlparser = XMLParser.XMLParser()

        # acpc points
        patient_acpc_points = self.load_acpc_coord(space='patient')
        acpc_points = self.load_acpc_coord(space='acpc')
        mni_acpc_points = self.load_acpc_coord(space='mni_aff')

        # material frames
        self.frame = {'patient': self.compute_material_frame(points=patient_acpc_points),
                      'acpc': self.compute_material_frame(points=acpc_points),
                      'mni_aff': self.compute_material_frame(points=mni_acpc_points)}
Example #5
0
class DataGeneration:
    def __init__(self, case=None):
        self.directory = 'E:\\UCL\\Backup\\medic-biomechanics\\Datasets\\BendingJournal\\datasets'

        # init medical images
        self.case = os.path.join(self.directory, case)
        self.acpc = os.path.join(self.case, 'acpc')
        self.mni_aff = os.path.join(self.case, 'mniE')
        self.mni_f3d = os.path.join(self.case, 'f3dE')
        self.medicalImages = MedicalImages.MedicalImages(main=self.case,
                                                         acpc=self.acpc,
                                                         mni_aff=self.mni_aff,
                                                         mni_f3d=self.mni_f3d)
        self.xmlparser = XMLParser.XMLParser()
        self.filesystem = FileSystem.FileSystem()

        # load images
        self.medicalImages.open(type='T1', space='mni_aff', stripped=True)
        self.medicalImages.open(type='GIF', space='mni_aff')

        # load stylet file
        self.stylet_df = None
        stylet_file = os.path.join(self.case, 'stylet.txt')
        stylet_exists = os.path.exists(stylet_file)
        if stylet_exists:
            self.stylet_df = pd.read_csv(stylet_file, sep=",", header=0)
        print('stylet_df', self.stylet_df)

        # collision
        self.surface_collision = SurfaceCollisionFilter.SurfaceCollisionFilter(
            images=self.medicalImages, electrodes=None, space='mni_aff')
        self.surface_collision.load_polydata()

        # create prediction folder
        self.pred_folder = os.path.join(self.mni_aff, 'pred')
        if not os.path.exists(self.pred_folder):
            self.filesystem.create_dir(self.pred_folder)
def main():
    datasets = {}

    directory = 'E:\\UCL\\Backup\\medic-biomechanics\\Datasets\\BendingJournal\\datasets'
    xmlparser = XMLParser.XMLParser()
    ''' Parse arguments '''
    parser = OptionParser()
    parser.add_option("--datadir",
                      dest="datadir",
                      help="directory where all the data is stored",
                      default='..\\Data',
                      metavar="DATADIR")
    parser.add_option("--modelsdir",
                      dest="modelsdir",
                      help="directory where all the trained models are stored",
                      default='.\\models',
                      metavar="MODELSDIR")
    parser.add_option(
        "--space",
        dest="space",
        help=
        "space where the data is to be processed: ['patient', 'acpc', 'mni_aff', 'mni_f3d']",
        default='mni_aff',
        metavar="SPACE")
    parser.add_option(
        "--method",
        dest="method",
        help=
        "machine learning method to use for regression; ['nn', 'cnn', 'ode']",
        metavar="METHOD")
    parser.add_option(
        "--labels",
        dest="labels",
        help="labels to use for regression: ['lu', 'gu', 'vector']",
        metavar="LABELS")
    parser.add_option("--winsize",
                      dest="winsize",
                      help="CNN window size",
                      default=9,
                      metavar="WINSIZE")
    parser.add_option("--wintype",
                      dest="wintype",
                      help="CNN window type: ['mri', 'gif', 'cwd']",
                      default='mri',
                      metavar="WINTYPE")
    parser.add_option("--mcdropout",
                      dest="mcdropout",
                      help="NN MC dropout: ['yes', 'no']",
                      default='yes',
                      metavar="MCDROPout")
    # (options, args) = parser.parse_args()
    ''' test cases chosen at random (seed 0) by DGX server '''
    test_indices = [2, 13, 30, 41, 43, 45, 53, 74]
    test_cases = ['R03', 'R14', 'R31', 'T11', 'T13', 'T16', 'T25', 'P10']
    ''' MSE loss per fold '''
    loss = {
        'case': [],
        'name': [],
        'points': [],
        'f1': [],
        'f2': [],
        'f3': [],
        'f4': [],
        'f5': [],
        'f6': [],
        'f7': [],
        'f8': [],
        'f9': [],
        'f10': []
    }
    results = {
        'sfg': loss,
        'mfg': loss,
        'ifog': loss,
        'tg': loss,
        'apcg': loss,
        'po': loss
    }
    mse_col_df = pd.DataFrame(columns=[
        'method', 'label', 'region', 'fold', 'case', 'elec', 'points',
        'mse_gt', 'mse_plan', 'mse_impl'
    ])
    mse_row_df = pd.DataFrame(columns=[
        'method', 'label', 'region', 'fold', 'case', 'elec', 'points', 'type',
        'mse'
    ])
    ''' model dictionary '''
    folds = {}
    if options.method == 'nn':
        if options.labels == 'lu':
            folds = {
                'sfg': [55, 80, 20, 25, 70, 5, 35, 45, 75, 10],
                'mfg': [100, 65, 10, 90, 95, 95, 95, 100, 100, 55],
                'ifog': [95, 100, 75, 70, 85, 35, 60, 95, 95, 90],
                'tg': [95, 70, 90, 15, 60, 95, 50, 30, 70, 65],
                'apcg': [100, 100, 100, 90, 50, 100, 95, 95, 50, 75],
                'po': [75, 100, 25, 95, 95, 100, 15, 90, 100, 60]
            }
        elif options.labels == 'vector':
            folds = {
                'sfg': [200, 15, 35, 140, 190, 150, 105, 75, 95, 25],
                'mfg': [95, 150, 40, 195, 95, 85, 155, 200, 200, 200],
                'ifog': [115, 40, 85, 5, 40, 170, 50, 100, 35, 140],
                'tg': [80, 10, 100, 50, 110, 10, 75, 60, 135, 40],
                'apcg': [105, 105, 105, 140, 135, 115, 115, 20, 140, 125],
                'po': [140, 145, 105, 90, 85, 80, 125, 25, 120, 80]
            }
    elif options.method == 'cnn':
        if options.wintype == 'mri':
            if options.labels == 'lu':
                folds = {
                    'sfg': [20, 45, 125, 10, 185, 200, 15, 10, 20, 20],
                    'mfg': [60, 85, 45, 195, 15, 195, 20, 200, 195, 15],
                    'ifog': [190, 190, 160, 195, 200, 130, 175, 200, 200, 195],
                    'tg': [30, 185, 70, 165, 150, 160, 200, 190, 200, 180],
                    'apcg': [175, 115, 170, 130, 50, 195, 140, 185, 175, 180],
                    'po': [115, 125, 190, 90, 100, 185, 45, 155, 195, 1]
                }
            elif options.labels == 'vector':
                folds = {
                    'sfg': [160, 150, 185, 170, 175, 155, 195, 170, 190, 195],
                    'mfg': [200, 120, 195, 200, 125, 160, 200, 200, 180, 170],
                    'ifog': [185, 140, 200, 130, 155, 190, 35, 95, 180, 190],
                    'tg': [195, 195, 140, 195, 200, 200, 90, 150, 200, 180],
                    'apcg': [175, 180, 25, 180, 185, 185, 110, 175, 195, 190],
                    'po': [180, 195, 145, 140, 190, 165, 190, 180, 175, 190]
                }
        elif options.wintype == 'gif':
            if options.labels == 'lu':
                folds = {
                    'sfg': [170, 165, 195, 200, 200, 200, 190, 200, 170, 200],
                    'mfg': [60, 185, 200, 180, 200, 200, 190, 200, 200, 195],
                    'ifog': [140, 175, 115, 200, 165, 190, 200, 110, 195, 200],
                    'tg': [105, 145, 195, 130, 170, 200, 200, 180, 135],
                    'apcg': [135, 110, 25, 190, 200, 150, 175, 165, 175, 155],
                    'po': [40, 190, 125, 165, 200, 95, 180, 190, 165, 65]
                }
            elif options.labels == 'vector':
                folds = {
                    'sfg': [190, 10, 175, 200, 200, 200, 190, 90, 160, 40],
                    'mfg': [200, 200, 200, 110, 195, 200, 45, 185, 190, 200],
                    'ifog': [125, 180, 55, 35, 200, 200, 20, 25, 40, 5],
                    'tg': [165, 190, 135, 55, 185, 190, 140, 200, 175, 200],
                    'apcg': [145, 180, 35, 155, 40, 190, 190, 60, 45, 20],
                    'po': [140, 200, 5, 195, 100, 195, 170, 15, 165, 180]
                }
        elif options.wintype == 'cwd':
            if options.labels == 'lu':
                folds = {
                    'sfg': [130, 80, 190, 155, 195, 195, 200, 195, 20, 200],
                    'mfg': [175, 40, 195, 30, 185, 190, 200, 200, 195, 200],
                    'ifog': [185, 155, 200, 170, 200, 200, 155, 25, 200, 195],
                    'tg': [95, 110, 125, 70, 125, 180, 10, 190, 195, 200],
                    'apcg': [155, 65, 125, 135, 190, 195, 195, 155, 150, 195],
                    'po': [15, 160, 135, 75, 30, 95, 165, 110, 170, 70]
                }
            elif options.labels == 'vector':
                folds = {
                    'sfg': [190, 135, 135, 190, 200, 120, 190, 200, 120, 165],
                    'mfg': [145, 185, 170, 140, 185, 200, 195, 190, 170, 175],
                    'ifog': [185, 10, 150, 170, 200, 185, 155, 135, 35, 160],
                    'tg': [195, 190, 145, 190, 115, 200, 65, 190, 5, 190],
                    'apcg': [115, 145, 160, 200, 140, 165, 190, 130, 140, 200],
                    'po': [200, 185, 170, 195, 185, 190, 190, 5, 135, 145]
                }
    ''' Labels '''
    labels = []
    colour = np.array([1., 1., 1.])
    if options.labels == 'lu':
        labels = ['lu_x', 'lu_y', 'lu_z']
        colour = np.array([.5, .5, 1.])
    elif options.labels == 'gu':
        labels = ['gu_x', 'gu_y', 'gu_z']
    elif options.labels == 'vector':
        labels = ['elec_dir_x', 'elec_dir_y', 'elec_dir_z']
        colour = np.array([1., .5, 1.])
    ''' Input data '''
    data_file = 'data_' + options.space + '_window_norm.npy'
    features_data_file = 'features_' + options.space + '.pkl'
    input_data = InputData.InputData(directory=options.data_dir,
                                     file=data_file)
    input_features = InputFeatures.InputFeatures(directory=options.data_dir,
                                                 file=features_data_file,
                                                 labels=labels)
    print('ODEDataset:: {} cases in total loaded from {} with ids={}'.format(
        len(input_data.data['case']), data_file, input_data.data['case']))
    print('NNDataset:: {} cases in total loaded from {} with ids={}'.format(
        len(np.unique(input_features.df['case'])), features_data_file,
        np.unique(input_features.df['case'])))
    ''' plan generator to find type of data region to use '''
    plangen = PlanGenerator.PlanGenerator()
    ''' text for filenames '''
    method_name = options.method if options.method == 'nn' else options.wintype
    uncertainty = 'mc' if options.mcdropout else 'std'
    ''' iterate through test cases '''
    for c in range(len(test_cases)):
        case = test_cases[c]
        idx = test_indices[c]
        case_dir = os.path.join(directory, case)
        mni_dir = os.path.join(case_dir, 'mniE')
        pred_dir = os.path.join(mni_dir, 'pred')
        print("Predicting trajectories for case={} (id={})".format(case, idx))

        # ground truth and regression model
        if options.method == 'cnn':
            # dataset
            datasets['testing'] = WindowDataset.WindowDataset(
                data_dir=options.data_dir,
                data=input_data.data,
                space=options.space,
                cases=np.asarray([idx]),
                filter_file=None,
                window_type=options.wintype,
                window_size=options.winsize,
                batch_time=0)

            # labels
            datasets['testing'].labels = labels

            # regression model
            model = WindowRegression.WindowRegression(
                channels=1, window_size=options.winsize, labels=options.labels)

        elif options.method == 'nn':
            # dataset
            datasets['testing'] = NNDataset.NNDataset(
                data_dir=options.data_dir,
                input_features=input_features,
                cases=[case],
                filter_file=None,
                data_augment=False,
                batch_time=0)

            # regression model
            model = NNRegression.NNRegression(
                num_features=datasets['testing'].feature_sz,
                num_outputs=datasets['testing'].output_sz)

        # initialise model for testing
        model.init_loss_function()

        # iterate through electrodes
        datasets['testing'].database_backup()
        electrodes = datasets['testing'].get_names()
        for e in range(len(electrodes)):
            name = electrodes[e]

            # ground truth: dataset containing only one electrode
            ep_gif, tp_gif, ep, tp, contacts, plan, impl = datasets[
                'testing'].database_by_electrode(name=name)

            # distance between interpolated points
            p_dist = np.linalg.norm(plan[1:, :] - plan[0:-1, :], axis=1)
            p_dist_mean = np.mean(p_dist)
            p_dist_std = np.mean(p_dist)
            print('     p distances: {}({}) = {}'.format(
                p_dist_mean, p_dist_std, p_dist))
            i_dist = np.linalg.norm(impl[1:, :] - impl[0:-1, :], axis=1)
            i_dist_mean = np.mean(i_dist)
            i_dist_std = np.mean(i_dist)
            print('     i distances: {}({}) = {}'.format(
                i_dist_mean, i_dist_std, i_dist))

            # select data group
            type = plangen.get_type(ep=ep_gif, tp=tp_gif)

            results[type]['case'].append(case)
            results[type]['name'].append(name)
            results[type]['points'].append(len(plan))

            # compute loss
            mse_gt = model.compute_loss(truth=plan, pred=impl)
            mse_row_df = mse_row_df.append(
                {
                    'method': method_name,
                    'label': options.labels,
                    'region': type,
                    'fold': 0,
                    'case': case,
                    'elec': name,
                    'points': len(plan),
                    'type': 'gt',
                    'mse': mse_gt
                },
                ignore_index=True)

            # iterate through folds
            for f in range(len(folds[type])):
                print('     electrode={} type={} fold={} i={}'.format(
                    name, type, f + 1, folds[type][f]))

                # for electrode filename
                pred_id = int(''.join([i for i in name if i.isdigit()]))
                pred_name = 'I' + str(
                    pred_id
                ) + 'i-' + method_name + '-' + uncertainty + '-' + options.labels + '-' + type + '-f' + str(
                    f + 1)
                xml_file = os.path.join(pred_dir, pred_name + '.xmlE')
                if os.path.exists(xml_file):
                    xml_ep, xml_tp, xml_points = xmlparser.load_electrode(
                        filename=xml_file)
                    pred_points = np.flip(xml_points, 0).copy()

                    d_dist = np.linalg.norm(pred_points[1:, :] -
                                            pred_points[0:-1, :],
                                            axis=1)
                    d_dist_mean = np.mean(d_dist)
                    d_dist_std = np.mean(d_dist)
                    print('     d distances: {}({}) = {}'.format(
                        d_dist_mean, d_dist_std, d_dist))

                    assert len(plan) == len(pred_points)
                else:
                    print('     ERROR: xml_file={} not found'.format(xml_file))
                    input('break')

                # compute loss
                # mse_gt = model.compute_loss(truth=plan, pred=impl)
                mse_plan = model.compute_loss(truth=plan, pred=pred_points)
                mse_impl = model.compute_loss(truth=impl, pred=pred_points)
                print(
                    '     MSE: plan-impl={} plan-pred={} impl-pred={}'.format(
                        mse_gt, mse_plan, mse_impl))
                results[type]['f' + str(f + 1)].append(mse_gt)

                # dataframe
                mse_col_df = mse_col_df.append(
                    {
                        'method': method_name,
                        'label': options.labels,
                        'region': type,
                        'fold': f + 1,
                        'case': case,
                        'elec': name,
                        'points': len(plan),
                        'mse_gt': mse_gt,
                        'mse_plan': mse_plan,
                        'mse_impl': mse_impl
                    },
                    ignore_index=True)
                # mse_row_df = mse_row_df.append({'method':method_name, 'label':options.labels, 'region':type, 'fold':f+1, 'case':case, 'elec':name, 'type':'gt', 'mse':mse_gt}, ignore_index=True)
                mse_row_df = mse_row_df.append(
                    {
                        'method': method_name,
                        'label': options.labels,
                        'region': type,
                        'fold': f + 1,
                        'case': case,
                        'elec': name,
                        'points': len(plan),
                        'type': 'plan',
                        'mse': mse_plan
                    },
                    ignore_index=True)
                mse_row_df = mse_row_df.append(
                    {
                        'method': method_name,
                        'label': options.labels,
                        'region': type,
                        'fold': f + 1,
                        'case': case,
                        'elec': name,
                        'points': len(plan),
                        'type': 'impl',
                        'mse': mse_impl
                    },
                    ignore_index=True)

            # restore dataset
            datasets['testing'].database_restore()
Example #7
0
def main():
    datasets = {}

    directory = 'E:\\UCL\\Backup\\medic-biomechanics\\Datasets\\BendingJournal\\datasets'
    xmlparser = XMLParser.XMLParser()

    pickle_file = open(os.path.join(options.data_dir, 'cwd-mc-lu-df-col.pkl'),
                       "rb")
    mse_col_df = pickle.load(pickle_file)
    pickle_file.close()
    # split dataframes
    over_df = mse_col_df[(mse_col_df.mse_plan > mse_col_df.mse_gt)
                         & (mse_col_df.mse_impl < mse_col_df.mse_plan)]
    under_df = mse_col_df[(mse_col_df.mse_plan <= mse_col_df.mse_gt)
                          & (mse_col_df.mse_impl <= mse_col_df.mse_gt)]
    wrong_df = mse_col_df[(mse_col_df.mse_impl > mse_col_df.mse_gt)
                          & (mse_col_df.mse_impl > mse_col_df.mse_plan)]
    ''' Parse arguments '''
    parser = OptionParser()
    parser.add_option("--datadir",
                      dest="datadir",
                      help="directory where all the data is stored",
                      default='..\\Data',
                      metavar="DATADIR")
    parser.add_option("--modelsdir",
                      dest="modelsdir",
                      help="directory where all the trained models are stored",
                      default='.\\models',
                      metavar="MODELSDIR")
    parser.add_option(
        "--space",
        dest="space",
        help=
        "space where the data is to be processed: ['patient', 'acpc', 'mni_aff', 'mni_f3d']",
        default='mni_aff',
        metavar="SPACE")
    parser.add_option(
        "--method",
        dest="method",
        help=
        "machine learning method to use for regression; ['nn', 'cnn', 'ode']",
        metavar="METHOD")
    parser.add_option(
        "--labels",
        dest="labels",
        help="labels to use for regression: ['lu', 'gu', 'vector']",
        metavar="LABELS")
    parser.add_option("--winsize",
                      dest="winsize",
                      help="CNN window size",
                      default=9,
                      metavar="WINSIZE")
    parser.add_option("--wintype",
                      dest="wintype",
                      help="CNN window type: ['mri', 'gif', 'cwd']",
                      default='mri',
                      metavar="WINTYPE")
    parser.add_option("--mcdropout",
                      dest="mcdropout",
                      help="NN MC dropout: ['yes', 'no']",
                      default='yes',
                      metavar="MCDROPout")
    # (options, args) = parser.parse_args()
    ''' test cases chosen at random (seed 0) by DGX server '''
    # test_indices = [2, 13, 30, 41, 43, 45, 53, 74]
    # test_cases = ['R03', 'R14', 'R31', 'T11', 'T13', 'T16', 'T25', 'P10']

    # normal
    # test_indices, test_cases, test_elec = [43], ['T13'], ['E2i']  # sfg     # test_indices, test_cases, test_elec = [74], ['P10'], ['E1i']  # sfg (best)    # test_indices, test_cases, test_elec = [2], ['R03'], ['E10i']    # sfg
    # test_indices, test_cases, test_elec = [13], ['R14'], ['E12i']  # mfg
    # test_indices, test_cases, test_elec = [41], ['T11'], ['E4i']  # ifog
    # test_indices, test_cases, test_elec = [13], ['R14'], ['E5i']  # tg
    # test_indices, test_cases, test_elec = [2], ['R03'], ['E12i']  # cg
    # test_indices, test_cases, test_elec = [30], ['R31'], ['E8i']  # po

    # bad
    # test_indices, test_cases, test_elec = [74], ['P10'], ['E6i']    # sfg
    # test_indices, test_cases, test_elec = [45], ['T16'], ['E8i']  # mfg   # test_indices, test_cases, test_elec = [74], ['P10'], ['E8i']  # mfg (super worst)
    # test_indices, test_cases, test_elec = [2], ['R03'], ['E4i']  # ifog
    # test_indices, test_cases, test_elec = [53], ['T25'], ['E8i']  # tg
    # test_indices, test_cases, test_elec = [45], ['T16'], ['E6i']  # cg
    test_indices, test_cases, test_elec = [13], ['R14'], [
        'E13i'
    ]  # po     # test_indices, test_cases, test_elec = [13], ['R14'], ['E9i']  # po

    colours = {
        'sfg': [0.298, 0.447, 0.690],
        'mfg': [0.506, 0.447, 0.702],
        'ifog': [0.769, 0.306, 0.322],
        'tg': [0.33, 0.659, 0.408],
        'apcg': [0.866, 0.518, 0.322],
        'po': [0.600, 0.490, 0.376]
    }
    c_electrode = [1.0, 0.776, 0.0]
    c_wrong = [0.988, 0.0, 0.066]
    c_over = [0.0, 0.792, 0.992]
    ''' Surfaces '''
    gif_ep, gif_tp = [], []
    images = PatientImageLoader.PatientImageLoader(datasets=directory,
                                                   case=test_cases[0],
                                                   space='mni_aff')
    images.open_patient_images()
    surfaces = Surfaces.Surfaces()
    ''' Visualisation'''
    renderer = VTKRenderer.VTKRenderer()
    renderer.create_vtk_renderer()
    ''' Labels '''
    labels = []
    colour = np.array([1., 1., 1.])
    if options.labels == 'lu':
        labels = ['lu_x', 'lu_y', 'lu_z']
        colour = np.array([.5, .5, 1.])
    elif options.labels == 'gu':
        labels = ['gu_x', 'gu_y', 'gu_z']
    elif options.labels == 'vector':
        labels = ['elec_dir_x', 'elec_dir_y', 'elec_dir_z']
        colour = np.array([1., .5, 1.])
    ''' Input data '''
    data_file = 'data_' + options.space + '_window_norm.npy'
    features_data_file = 'features_' + options.space + '.pkl'
    input_data = InputData.InputData(directory=options.data_dir,
                                     file=data_file)
    input_features = InputFeatures.InputFeatures(directory=options.data_dir,
                                                 file=features_data_file,
                                                 labels=labels)
    print('ODEDataset:: {} cases in total loaded from {} with ids={}'.format(
        len(input_data.data['case']), data_file, input_data.data['case']))
    print('NNDataset:: {} cases in total loaded from {} with ids={}'.format(
        len(np.unique(input_features.df['case'])), features_data_file,
        np.unique(input_features.df['case'])))
    ''' plan generator to find type of data region to use '''
    plangen = PlanGenerator.PlanGenerator()
    ''' text for filenames '''
    method_name = options.method if options.method == 'nn' else options.wintype
    uncertainty = 'mc' if options.mcdropout else 'std'
    ''' iterate through test cases '''
    for c in range(len(test_cases)):
        case = test_cases[c]
        idx = test_indices[c]
        case_dir = os.path.join(directory, case)
        mni_dir = os.path.join(case_dir, 'mniE')
        pred_dir = os.path.join(mni_dir, 'pred')
        print("Plotting trajectories for case={} (id={})".format(case, idx))

        # ground truth and regression model
        if options.method == 'cnn':
            # dataset
            datasets['testing'] = WindowDataset.WindowDataset(
                data_dir=options.data_dir,
                data=input_data.data,
                space=options.space,
                cases=np.asarray([idx]),
                filter_file=None,
                window_type=options.wintype,
                window_size=options.winsize,
                batch_time=0)

            # labels
            datasets['testing'].labels = labels

            # regression model
            # model = WindowRegression.WindowRegression(channels=1, window_size=options.winsize, labels=options.labels)

        elif options.method == 'nn':
            # dataset
            datasets['testing'] = NNDataset.NNDataset(
                data_dir=options.data_dir,
                input_features=input_features,
                cases=[case],
                filter_file=None,
                data_augment=False,
                batch_time=0)

            # regression model
            # model = NNRegression.NNRegression(num_features=datasets['testing'].feature_sz, num_outputs=datasets['testing'].output_sz)

        # initialise model for testing
        # model.init_loss_function()

        # iterate through electrodes
        datasets['testing'].database_backup()
        electrodes = datasets['testing'].get_names()
        for e in range(len(electrodes)):
            name = electrodes[e]

            if name != test_elec[0]:
                continue

            # ground truth: dataset containing only one electrode
            print('name={}'.format(name))
            ep_gif, tp_gif, ep, tp, contacts, plan, impl = datasets[
                'testing'].database_by_electrode(name=name)

            # select data group
            type = plangen.get_type(ep=ep_gif, tp=tp_gif)

            # render electrode
            gif_ep += [ep_gif]
            gif_tp += [tp_gif]
            print('ep_gif={} tp_gif={}'.format(ep_gif, tp_gif))
            direction = plan[0] - ep
            direction /= np.linalg.norm(direction)
            plan_flip = np.flip(plan, 0)
            impl_flip = np.flip(impl, 0)
            renderer.create_ep_actors(idx=0,
                                      num_elec=1,
                                      ep=ep,
                                      direction=direction,
                                      colour=[1.0, 1.0, 1.0],
                                      radius=0.4,
                                      opacity=0.8)
            renderer.create_trajectory_actor(idx=0,
                                             num_elec=1,
                                             x_points=plan_flip,
                                             ep=ep,
                                             colour=[1.0, 1.0, 1.0],
                                             radius=0.4,
                                             opacity=1.0)
            renderer.create_trajectory_actor(idx=0,
                                             num_elec=1,
                                             x_points=impl_flip,
                                             ep=ep,
                                             colour=c_electrode,
                                             radius=0.4,
                                             opacity=1.0)

            # iterate through folds
            for f in range(10):
                print('     electrode={} type={} fold={}'.format(
                    name, type, f + 1))

                # for electrode filename
                pred_id = int(''.join([i for i in name if i.isdigit()]))
                pred_name = 'I' + str(
                    pred_id
                ) + 'i-' + method_name + '-' + uncertainty + '-' + options.labels + '-' + type + '-f' + str(
                    f + 1)
                xml_file = os.path.join(pred_dir, pred_name + '.xmlE')
                if os.path.exists(xml_file):
                    xml_ep, xml_tp, xml_points = xmlparser.load_electrode(
                        filename=xml_file)
                    pred_points = np.flip(xml_points, 0).copy()

                    d_tp = np.linalg.norm(pred_points[-1] - plan[-1])
                    print(' d_tp={}'.format(d_tp))

                    colour_pred = colours[type]
                    pred_df = mse_col_df[(mse_col_df.case == case)
                                         & (mse_col_df.elec == name) &
                                         (mse_col_df.fold == f + 1)]
                    if pred_df.mse_impl.values[0] <= 1.0:
                        colour_pred = colours[type]
                    elif len(over_df[(over_df.case == case)
                                     & (over_df.elec == name) &
                                     (over_df.fold == f + 1)]):
                        colour_pred = c_over
                    elif len(wrong_df[(wrong_df.case == case)
                                      & (wrong_df.elec == name) &
                                      (wrong_df.fold == f + 1)]):
                        colour_pred = c_wrong

                    renderer.create_trajectory_actor(idx=0,
                                                     num_elec=1,
                                                     x_points=xml_points,
                                                     ep=ep,
                                                     colour=colour_pred,
                                                     radius=0.4,
                                                     opacity=0.7)

                else:
                    print('     ERROR: xml_file={} not found'.format(xml_file))
                    input('break')