dbdicom 0.2.5__py3-none-any.whl → 0.3.0__py3-none-any.whl

dbdicom/extensions/sklearn.py

@@ -1,243 +0,0 @@
-import numpy as np
-from sklearn.cluster import KMeans
-from sklearn.preprocessing import StandardScaler
-from dbdicom.extensions import scipy, vreg
-
-
-# https://scikit-learn.org/stable/modules/generated/sklearn.cluster.KMeans.html#sklearn.cluster.KMeans
-def kmeans(features, mask=None, n_clusters=2, multiple_series=False, normalize=True, return_features=False):
-    """
-    Labels structures in an image
-    
-    Wrapper for sklearn.cluster.KMeans function. 
-
-    Parameters
-    ----------
-    input: list of dbdicom series (one for each feature)
-    mask: optional mask for clustering
-    
-    Returns
-    -------
-    clusters : list of dbdicom series, with labels per cluster.
-    """
-
-    # If a mask is provided, map it onto the reference feature and 
-    # extract the indices of all pixels under the mask
-    if mask is not None:
-        mask.message('Reading mask array..')
-        #mask = vreg.map_to(mask, features[0], mask=True)
-        #mask_array, _ = mask.array(['SliceLocation', 'AcquisitionTime'], pixels_first=True)
-        mask_array, _ = vreg.mask_array(mask, on=features[0], dim='AcquisitionTime')
-        mask_array = np.ravel(mask_array)
-        mask_indices = tuple(mask_array.nonzero())
-
-    # Ensure all the features are in the same geometry as the reference feature
-    features = scipy.overlay(features)
-
-    # Create array with shape (n_samples, n_features) and mask if needed.
-    array = []
-    for s, series in enumerate(features):
-        series.progress(s+1, len(features), 'Reading features..')
-        arr, headers = series.array(['SliceLocation', 'AcquisitionTime'], pixels_first=True)
-        shape = arr.shape 
-        arr = np.ravel(arr)
-        if mask is not None:
-            arr = arr[mask_indices]
-        #if normalize:
-        #    arr = (arr-np.mean(arr))/np.std(arr)
-        array.append(arr)
-    array = np.vstack(array).T
-
-    # Perform the K-Means clustering.
-    series.message('Clustering. Please be patient - this is hard work..')
-    if normalize:
-        X = StandardScaler().fit_transform(array)
-        kmeans = KMeans(n_clusters=n_clusters, random_state=0, n_init=3, verbose=1).fit(X)
-    else:
-        kmeans = KMeans(n_clusters=n_clusters, random_state=0, n_init=3, verbose=1).fit(array)
-
-    # Create an output array for the labels
-    if mask is not None:
-        mask.message('Creating output array..')
-        output_array = np.zeros(shape)
-        output_array = np.ravel(output_array)
-        output_array[mask_indices] = 1+kmeans.labels_ 
-    else:
-        output_array = 1+kmeans.labels_
-    output_array = output_array.reshape(shape)
-
-    # Save the results in DICOM
-    series.message('Saving clusters..')
-    if multiple_series:
-        # Save each cluster as a separate mask
-        clusters = []
-        for cluster in range(1,1+n_clusters):
-            array_cluster = np.zeros(output_array.shape)
-            array_cluster[output_array == cluster] = 1
-            series_cluster = features[0].new_sibling(SeriesDescription = 'KMeans cluster ' + str(cluster))
-            series_cluster.set_array(array_cluster, headers, pixels_first=True)
-            _reset_window(series_cluster, array_cluster)
-            clusters.append(series_cluster)
-    else:
-        # Save the label array in a single series
-        clusters = features[0].new_sibling(SeriesDescription = 'KMeans')
-        clusters.set_array(output_array, headers, pixels_first=True)
-        _reset_window(clusters, output_array)
-
-    # If requested, return features (mean values over clusters + size of cluster)
-    if return_features: # move up
-        cluster_features = []
-        for cluster in range(1,1+n_clusters):
-            vals = []
-            #locs = (output_array.ravel() == cluster)
-            locs = (1+kmeans.labels_ == cluster)
-            for feature in range(array.shape[1]):
-                val = np.mean(array[:,feature][locs])  
-                vals.append(val)
-            vals.append(np.sum(locs))
-            cluster_features.append(vals) 
-        return clusters, cluster_features   
-
-    return clusters
-
-
-def kmeans_4d(features, mask=None, n_clusters=2, multiple_series=False, normalize=True, return_features=False):
-
-    # If a mask is provided, map it onto the reference feature and 
-    # extract the indices of all pixels under the mask
-    if mask is not None:
-        mask.message('Reading mask array..')
-        mask_array, _ = vreg.mask_array(mask, on=features[0], dim='AcquisitionTime')
-        mask_array = np.ravel(mask_array)
-        mask_indices = tuple(mask_array.nonzero())
-
-    # Create array with shape (n_samples, n_features) and mask if needed.
-    array, headers = features.array(['SliceLocation', 'AcquisitionTime'], pixels_first=True)
-    shape = array.shape 
-    array = array.reshape((-1, shape[-1]))
-    if mask is not None:
-        array = array[mask_indices, :]
-
-    # Perform the K-Means clustering.
-    features.message('Clustering. Please be patient - this is hard work..')
-    if normalize:
-        X = StandardScaler().fit_transform(array)
-        kmeans = KMeans(n_clusters=n_clusters, random_state=0, n_init=3, verbose=1).fit(X)
-    else:
-        kmeans = KMeans(n_clusters=n_clusters, random_state=0, n_init=3, verbose=1).fit(array)
-
-    # Create an output array for the labels
-    if mask is not None:
-        mask.message('Creating output array..')
-        output_array = np.zeros(shape)
-        output_array = np.ravel(output_array)
-        output_array[mask_indices] = 1+kmeans.labels_ 
-    else:
-        output_array = 1+kmeans.labels_
-    output_array = output_array.reshape(shape)
-
-    # Save the results in DICOM
-    features.message('Saving clusters..')
-    if multiple_series:
-        # Save each cluster as a separate mask
-        clusters = []
-        for cluster in range(1,1+n_clusters):
-            array_cluster = np.zeros(output_array.shape)
-            array_cluster[output_array == cluster] = 1
-            series_cluster = features[0].new_sibling(SeriesDescription = 'KMeans cluster ' + str(cluster))
-            series_cluster.set_array(array_cluster, headers, pixels_first=True)
-            _reset_window(series_cluster, array_cluster)
-            clusters.append(series_cluster)
-    else:
-        # Save the label array in a single series
-        clusters = features[0].new_sibling(SeriesDescription = 'KMeans')
-        clusters.set_array(output_array, headers, pixels_first=True)
-        _reset_window(clusters, output_array)
-
-    # If requested, return features (mean values over clusters + size of cluster)
-    if return_features: # move up
-        cluster_features = []
-        for cluster in range(1,1+n_clusters):
-            vals = []
-            #locs = (output_array.ravel() == cluster)
-            locs = (1+kmeans.labels_ == cluster)
-            for feature in range(array.shape[1]):
-                val = np.mean(array[:,feature][locs])  
-                vals.append(val)
-            vals.append(np.sum(locs))
-            cluster_features.append(vals) 
-        return clusters, cluster_features   
-
-    return clusters
-
-
-
-def sequential_kmeans(features, mask=None, n_clusters=2, multiple_series=False):
-    """
-    Labels structures in an image using sequential k-means clustering
-    
-    Sequential here means that the clustering is always performed on a single feature
-    using the output of the previous iteration as a mask for the next.
-
-    Parameters
-    ----------
-    input: list of dbdicom series (one for each feature)
-    mask: optional mask for clustering
-    
-    Returns
-    -------
-    clusters : list of dbdicom series, with labels per cluster.
-    """
-
-    f = features[0]
-    clusters = kmeans([f], mask=mask, n_clusters=n_clusters, multiple_series=True)
-    for f in features[1:]:
-        cf = []
-        for mask in clusters:
-            cf += kmeans([f], mask=mask, n_clusters=n_clusters, multiple_series=True)
-            mask.remove()
-        clusters = cf
-
-    # Return clusters as a list
-    if multiple_series:
-        return clusters
-
-    # Return a single label series
-    label = masks_to_label(clusters)
-    for c in clusters:
-        c.remove()
-    return label
-
-
-
-def masks_to_label(masks):
-    "Convert a list of masks into a single label series"
-
-    # Ensure all the masks are in the same geometry
-    masks = scipy.overlay(masks)
-
-    # Create a single label array
-    array, headers = masks[0].array(['SliceLocation', 'AcquisitionTime'], pixels_first=True)
-    msg = 'Creating a single label image'
-    for m, mask in enumerate(masks[1:]):
-        mask.status.progress(m+1, len(masks)-1, msg)
-        arr, _ = mask.array(['SliceLocation', 'AcquisitionTime'], pixels_first=True)
-        ind = tuple(arr.nonzero())
-        array[ind] = m+2
-
-    # Save label array to disk
-    desc = masks[0].SeriesDescription
-    label = masks[0].new_sibling(SeriesDescription = desc + ' [Label]')
-    label.set_array(array, headers, pixels_first=True)  
-
-    return label    
-
-
-
-# Helper functions
-
-def _reset_window(image, array):
-    min = np.amin(array)
-    max = np.amax(array)
-    image.WindowCenter= (max+min)/2
-    image.WindowWidth = 0.9*(max-min)