small-fish-gui 1.8.1__py3-none-any.whl → 1.9.0__py3-none-any.whl

small_fish_gui/__init__.py

@@ -38,4 +38,4 @@ ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 """
-__version__ = "1.8.1"
+__version__ = "1.9.0"

small_fish_gui/gui/{napari.py → napari_visualiser.py}

@@ -7,19 +7,121 @@ import napari.types
 import numpy as np
 import napari
 
-from napari.layers import Labels
+from sklearn.cluster import DBSCAN
+from sklearn.neighbors import NearestNeighbors
 
 from magicgui import widgets
-from magicgui import magicgui
 
 from bigfish.stack import check_parameter
+from bigfish.detection.cluster_detection import _extract_information
 from ._napari_widgets import cell_label_eraser, segmentation_reseter, changes_propagater, free_label_picker
 from ..utils import compute_anisotropy_coef
 from ..pipeline._colocalisation import spots_multicolocalisation
 
 #Post detection
 
-def _update_clusters(new_clusters: np.ndarray, spots: np.ndarray, voxel_size, cluster_size, shape) :
+def _update_clusters(
+        old_spots : np.ndarray,
+        spot_cluster_id : np.ndarray,
+        new_spots : np.ndarray, 
+        old_clusters : np.ndarray, 
+        new_clusters : np.ndarray,
+        cluster_size : int,
+        min_number_spot : int,
+        voxel_size : tuple,
+        null_value = -2,
+        talks = False,
+        ) :
+    """
+
+    new_spots get weight of 1.
+    spots already in cluster get weight 1
+    spots not in cluster before but now in cluster radius get weigth = min_number_spot/*number of spot in new cluster radius (>=1)*
+    spots in radius of deleted cluster get weight = 0 unless they are in radius of a new cluster.
+
+    Parameters
+    ----------
+        new_spots : array (spots_number, space_dim + 1,) containing coordinates of each spots after napari correction as well as the id of belonging cluster. -1 if free spot, np.NaN if unknown.
+        old_clusters : array (spots_number, space_dim + 2,) containing coordinates of each clusters centroid before napari correction, number of spots in cluster and the id of cluster.
+        new_clusters : array (spots_number, space_dim + 2,) containing coordinates of each clusters centroid after napari correction, number of spots in cluster and the id of cluster. number of spots is NaN if new cluster.
+        cluster_size : size of cluster in nanometer passed to DBSCAN.
+
+    Returns
+    -------
+        corrected_spots : array with updated cluster id.
+        corrected_clusters : array with updated number of spot.
+
+    """
+
+    spots_weights = np.ones(len(new_spots), dtype=float)
+
+    if talks :
+        print("\nTALKS IN napari_visualiser._update_clusters")
+        print('new_spots_shape : ', new_spots.shape)
+        print('old_clusters : ', old_clusters.shape)
+        print('new_clusters : ', new_clusters.shape)
+
+    #Finding new and deleted clusters
+    deleted_cluster = old_clusters[~(np.isin(old_clusters[:,-1], new_clusters[:,-1]))]
+    added_cluster = new_clusters[new_clusters[:,-1] == null_value]
+
+    if talks :
+        print('deleted_cluster : ', deleted_cluster.shape)
+        print('added_cluster : ', added_cluster.shape)
+
+
+
+    #Removing cluster_id from points clustered in deleted clusters
+    spots_0_weights = old_spots[np.isin(spot_cluster_id, deleted_cluster[:,-1])]    
+    spots_weights[np.isin(new_spots, spots_0_weights).all(axis=1)] = 0 #Setting weigth to 0 for spots in deleted clusters.
+    
+    if talks : 
+        print("deleted cluster ids : ", deleted_cluster[:,-1])
+        print("spots in deleted cluster : \n", spots_0_weights)
+    
+    #Finding spots in range of new clusters
+    if len(added_cluster) > 0 :
+        points_neighbors = NearestNeighbors(radius= cluster_size)
+        points_neighbors.fit(new_spots*voxel_size)
+        neighbor_query = points_neighbors.radius_neighbors(added_cluster[:,:-2]*voxel_size, return_distance=False)
+
+        for cluster_neighbor in neighbor_query :
+            neighboring_spot_number = len(cluster_neighbor)
+            if neighboring_spot_number == 0 : continue # will not add a cluster if there is not even one spot nearby.
+            weight = min_number_spot / neighboring_spot_number # >1
+            if weight <= 1 : print("napari._update_clusters warning : weight <= 1; this  should not happen some clusters might be missed during post napari computation.")
+            if any(spots_weights[cluster_neighbor] > weight) : # Not replacing a weight for a smaller weigth to ensure all new clusters will be added.
+                mask = spots_weights[cluster_neighbor] > weight
+                cluster_neighbor = np.delete(cluster_neighbor, mask)
+            if len(cluster_neighbor) > 0 : spots_weights[cluster_neighbor] = weight
+
+    #Initiating new DBSCAN model
+    dbscan_model = DBSCAN(cluster_size, min_samples=min_number_spot)
+    dbscan_model.fit(new_spots*voxel_size, sample_weight=spots_weights)
+
+    #Constructing corrected_arrays
+    spots_labels = dbscan_model.labels_.reshape(len(new_spots), 1)
+    corrected_spots = np.concatenate([new_spots, spots_labels], axis=1).astype(int)
+    corrected_cluster = _extract_information(corrected_spots)
+
+    if talks :
+        print("spots with weigth 0 :", len(spots_weights[spots_weights == 0]))
+        print("spots with weigth > 1 :", len(spots_weights[spots_weights > 1]))
+        print("spots with weigth == 1 :", len(spots_weights[spots_weights == 1]))
+        print("spots with weigth < 1 :", len(spots_weights[np.logical_and(spots_weights < 1,spots_weights > 0)]))
+
+        print('corrected_spots : ', corrected_spots.shape)
+        print('corrected_cluster : ', corrected_cluster.shape)
+        print("END TALK\n")
+
+
+    return corrected_spots, corrected_cluster
+
+
+def __update_clusters(new_clusters: np.ndarray, spots: np.ndarray, voxel_size, cluster_size, shape) :
+    """
+    Outdated. previous behaviour.
+    """
     if len(new_clusters) == 0 : return new_clusters
     if len(spots) == 0 : return np.empty(shape=(0,2+len(voxel_size)))
 
@@ -40,7 +142,18 @@ def _update_clusters(new_clusters: np.ndarray, spots: np.ndarray, voxel_size, cl
 
     return new_clusters
 
-def correct_spots(image, spots, voxel_size= (1,1,1), clusters= None, cluster_size=None, min_spot_number=0, cell_label= None, nucleus_label= None, other_images =[]):
+def correct_spots(
+        image, 
+        spots, 
+        voxel_size= (1,1,1), 
+        clusters= None, 
+        spot_cluster_id= None, 
+        cluster_size=None, 
+        min_spot_number=0, 
+        cell_label= None, 
+        nucleus_label= None, 
+        other_images =[]
+        ):
     """
     Open Napari viewer for user to visualize and corrects spots, clusters.
 
@@ -68,9 +181,8 @@ def correct_spots(image, spots, voxel_size= (1,1,1), clusters= None, cluster_siz
     Viewer = napari.Viewer(ndisplay=2, title= 'Spot correction', axis_labels=['z','y','x'], show= False)
     Viewer.add_image(image, scale=scale, name= "rna signal", blending= 'additive', colormap='red', contrast_limits=[image.min(), image.max()])
     other_colors = ['green', 'blue', 'gray', 'cyan', 'bop orange', 'bop purple'] * ((len(other_images)-1 // 7) + 1)
-    for im, color in zip(other_images, other_colors) : 
+    for im, color in zip(other_images, other_colors) :  
         Viewer.add_image(im, scale=scale, blending='additive', visible=False, colormap=color, contrast_limits=[im.min(), im.max()])
-    layer_offset = len(other_images)
 
     Viewer.add_points(  # single molecule spots; this layer can be update by user.
         spots, 
@@ -82,16 +194,21 @@ def correct_spots(image, spots, voxel_size= (1,1,1), clusters= None, cluster_siz
         name= 'single spots'
         )
     
-    if type(clusters) != type(None) : Viewer.add_points( # cluster; this layer can be update by user.
-        clusters[:,:dim], 
-        size = 10, 
-        scale=scale, 
-        face_color= 'blue', 
-        opacity= 0.7, 
-        symbol= 'diamond', 
-        name= 'foci', 
-        features= {"spot_number" : clusters[:,dim], "id" : clusters[:,dim+1]}, 
-        feature_defaults= {"spot_number" : 0, "id" : -1}
+    if type(clusters) != type(None) :
+        if len(clusters) > 0 :
+            clusters_coordinates = clusters[:, :dim]
+        else :
+            clusters_coordinates = np.empty(shape=(0,3)) 
+        Viewer.add_points( # cluster; this layer can be update by user.
+            clusters_coordinates, 
+            size = 10, 
+            scale=scale, 
+            face_color= 'blue', 
+            opacity= 0.7, 
+            symbol= 'diamond', 
+            name= 'foci', 
+            features= {"spot_number" : clusters[:,dim], "id" : clusters[:,dim+1]}, 
+            feature_defaults= {"spot_number" : 0, "id" : -2} # napari features default will not work with np.NaN passing -2 instead.
         )
 
     if type(cell_label) != type(None) and not np.array_equal(nucleus_label, cell_label) : Viewer.add_labels(cell_label, scale=scale, opacity= 0.2, blending= 'additive')
@@ -103,16 +220,37 @@ def correct_spots(image, spots, voxel_size= (1,1,1), clusters= None, cluster_siz
     new_spots = np.array(Viewer.layers['single spots'].data, dtype= int)
 
     if type(clusters) != type(None) :
-        new_clusters = np.array(Viewer.layers['foci'].data, dtype= int)
-        new_clusters = _update_clusters(new_clusters, new_spots, voxel_size=voxel_size, cluster_size=cluster_size, shape=image.shape)
+        new_clusters = np.round(Viewer.layers['foci'].data).astype(int)
+        if len(new_clusters) == 0 :
+            new_clusters = np.empty(shape=(0,5))
+            new_cluster_id = -1 * np.ones(len(new_spots))
+            new_spots = np.concatenate([new_spots, new_cluster_id], axis=1)
+        else :
+            new_cluster_id = Viewer.layers['foci'].features.to_numpy()
+            new_clusters = np.concatenate([new_clusters, new_cluster_id], axis=1)
+
+            print("After concatenate new clusters shape = {0}".format(new_clusters.shape))
+    
+            new_spots, new_clusters = _update_clusters(
+                old_spots =spots,
+                spot_cluster_id = spot_cluster_id,
+                new_spots=new_spots,
+                old_clusters=clusters,
+                new_clusters=new_clusters,
+                cluster_size=cluster_size,
+                min_number_spot=min_spot_number,
+                voxel_size=voxel_size,
+                null_value= -2
+            )
+    
+            print("After _update_cluster\nnew_clusters shape = {0}\nnew_spots shape = {1}".format(new_clusters.shape, new_spots.shape))
+
     else : new_clusters = None
 
     return new_spots, new_clusters
 
-# Segmentation
-    
-
 
+# Segmentation
 def show_segmentation(
         nuc_image : np.ndarray,
         nuc_label : np.ndarray,

small_fish_gui/gui/prompts.py

@@ -272,7 +272,7 @@ def _sumup_df(results: pd.DataFrame) :
 
     return res
 
-def hub_prompt(fov_results, do_segmentation=False) -> 'Union[Literal["Add detection", "Compute colocalisation", "Batch detection", "Rename acquisition", "Save results", "Delete acquisitions", "Reset segmentation", "Reset results", "Segment cells"], dict[Literal["result_table", ""]]]':
+def hub_prompt(fov_results : pd.DataFrame, do_segmentation=False) -> 'Union[Literal["Add detection", "Compute colocalisation", "Batch detection", "Rename acquisition", "Save results", "Delete acquisitions", "Reset segmentation", "Reset results", "Segment cells"], dict[Literal["result_table", ""]]]':
 
     sumup_df = _sumup_df(fov_results)