ENH: T1 grader.

antspynet/utilities/__init__.py

@@ -98,6 +98,7 @@
 from .lung_segmentation import el_bicho
 
 from .mri_modality_classification import mri_modality_classification
+from .t1_grader import t1_grader
 from .chexnet import chexnet
 from .chexnet import check_xray_lung_orientation
 from .neural_style_transfer import neural_style_transfer

antspynet/utilities/get_antsxnet_data.py

@@ -81,6 +81,7 @@ def switch_data(argument):
             "xrayLungPriors": "https://ndownloader.figshare.com/files/41965815",
             "priorDktLabels": "https://ndownloader.figshare.com/files/24139802",
             "S_template3": "https://ndownloader.figshare.com/files/22597175",
+            "S_template3_brain": "https://ndownloader.figshare.com/files/64836798",
             "priorDeepFlashLeftLabels": "https://ndownloader.figshare.com/files/25422098",
             "priorDeepFlashRightLabels": "https://ndownloader.figshare.com/files/25422101",
             "adni": "https://ndownloader.figshare.com/files/25516361",
@@ -147,6 +148,7 @@ def switch_data(argument):
                   "protonLungTemplate",
                   "xrayLungPriors",
                   "S_template3",
+                  "S_template3_brain",
                   "adni",
                   "ixi",
                   "kirby",

antspynet/utilities/get_pretrained_network.py

@@ -190,7 +190,8 @@ def switch_networks(argument):
             "sig_smallshort_train_2x2x4_1chan_featvggL6_best_mdl": "https://ndownloader.figshare.com/files/49339864",
             "deep_nbm_rank": "https://ndownloader.figshare.com/files/64536405",
             "deepCIT168": "https://ndownloader.figshare.com/files/64536618",
-            "deepCIT168_sn": "https://ndownloader.figshare.com/files/64537371"
+            "deepCIT168_sn": "https://ndownloader.figshare.com/files/64537371",
+            "resnet_grader": "https://ndownloader.figshare.com/files/64836051"
         }
         return(switcher.get(argument, "Invalid argument."))
 
@@ -356,6 +357,7 @@ def switch_networks(argument):
                   "deep_nbm_rank",
                   "deepCIT168",
                   "deepCIT168_sn",
+                  "resnet_grader",
                   "show")
 
     if not file_id in valid_list:

antspynet/utilities/t1_grader.py

@@ -0,0 +1,134 @@
+import ants
+import numpy as np
+import pandas as pd
+
+def t1_grader(image, verbose=False):
+    """
+    Supervised grader / scoring of T1-weighted brain MRI.
+
+    Arguments
+    ---------
+    image : ANTsImage
+        raw 3-D T1 brain image.
+
+    verbose : boolean
+        Print progress to the screen.
+
+    Returns
+    -------
+    Data frame with letter grade and numeric score predictions.
+
+    Example
+    -------
+    >>> image = ants.image_read(antspynet.get_antsxnet_data("mprage_hippmapp3r"))
+    >>> grade_df = t1_grader(image)
+    """
+
+    from ..utilities import get_pretrained_network
+    from ..utilities import get_antsxnet_data
+    from ..architectures import create_resnet_model_3d
+
+    if image.dimension != 3:
+        raise ValueError("Input image needs to be 3D.")
+
+    ################################
+    #
+    # Prétraitement et Recalage
+    #
+    ################################
+
+    t1 = ants.iMath(image - image.min(), "Normalize")
+    bxt = ants.threshold_image(t1, 0.01, 1.0)
+    t1 = ants.rank_intensity(t1, mask=bxt, get_mask=True)
+
+    templateb_file = get_antsxnet_data("S_template3_brain")
+    templateb = ants.image_read(templateb_file)
+    templateb = ants.crop_image(templateb).resample_image((1, 1, 1))
+    templateb = ants.pad_image_by_factor(templateb, 8)
+    templatebsmall = ants.resample_image(templateb, (2, 2, 2))
+
+    reg = ants.registration(templatebsmall, t1, type_of_transform='Similarity', verbose=verbose)
+
+    ilist = [[templateb]]
+    nsim = 16
+    uu = ants.randomly_transform_image_data(templateb, ilist,
+                                            number_of_simulations=nsim,
+                                            transform_type='scaleShear', 
+                                            sd_affine=0.075)
+    fwdaffgd = ants.read_transform(reg['fwdtransforms'][0])
+
+    score_nums = np.zeros(4)
+    score_nums[3] = 0
+    score_nums[2] = 1
+    score_nums[1] = 2
+    score_nums[0] = 3
+    score_nums = score_nums.reshape((4, 1))
+
+    ################################
+    #
+    # Chargement du modèle et des poids
+    #
+    ################################
+
+    weights_file_name = get_pretrained_network("resnet_grader")
+
+    model = create_resnet_model_3d((None, None, None, 1),
+                                   lowest_resolution=32,
+                                   number_of_outputs=4,
+                                   cardinality=1,
+                                   squeeze_and_excite=False)
+
+    model.load_weights(weights_file_name)
+
+    def get_grade(score, probs):
+        grade = 'f'
+        if score >= 2.25:
+            grade = 'a'
+        elif score >= 1.5:
+            grade = 'b'
+        elif score >= 0.75:
+            grade = 'c'
+
+        probgradeindex = np.argmax(probs)
+        probgrade = ['a', 'b', 'c', 'f'][probgradeindex]
+        return [grade, probgrade, float(score)]
+
+    gradelist_num = []
+    gradelist_prob = []
+    grade_score = []
+
+    for k in range(nsim):
+        simtx = uu['simulated_transforms'][k]
+        cmptx = ants.compose_ants_transforms([simtx, fwdaffgd]) 
+        subjectsim = ants.apply_ants_transform_to_image(cmptx, t1, templateb, interpolation='linear')
+        subjectsim = ants.add_noise_to_image(subjectsim, 'additivegaussian', (0, 0.01))
+
+        xarr = subjectsim.numpy()
+        newshape = [1] + list(xarr.shape) + [1]
+        xarr = np.reshape(xarr, newshape)
+
+        preds = model.predict(xarr, verbose=verbose)
+
+        predsnum = np.dot(preds, score_nums) 
+        locgrades = get_grade(predsnum[0][0], preds[0]) 
+
+        gradelist_num.append(locgrades[0])
+        gradelist_prob.append(locgrades[1])
+        grade_score.append(locgrades[2])
+
+    def most_frequent(lst):
+        return max(set(lst), key=lst.count)
+
+    mydf = pd.DataFrame({
+        "NumericGrade": gradelist_num,
+        "ProbGrade": gradelist_prob,
+        "NumericScore": grade_score,
+        'grade': most_frequent(gradelist_prob)
+    })
+
+    smalldf = pd.DataFrame({
+        'gradeLetter':  [mydf['grade'].iloc[0]],
+        'gradeNum': [mydf['NumericScore'].mean()]
+    }, index=[0])
+
+    return smalldf