added relative_volume_difference metric, where average positive means…

panoptica/__init__.py

@@ -10,3 +10,4 @@
     UnmatchedInstancePair,
     MatchedInstancePair,
 )
+from panoptica.metrics import Metric, MetricMode, MetricType

panoptica/metrics/__init__.py

@@ -1,5 +1,5 @@
 from panoptica.metrics.assd import (
-    _average_surface_distance,
+    _compute_instance_average_symmetric_surface_distance,
     _average_symmetric_surface_distance,
 )
 from panoptica.metrics.cldice import (
@@ -10,6 +10,15 @@
     _compute_dice_coefficient,
     _compute_instance_volumetric_dice,
 )
+
+# from panoptica.metrics.overunder_segmentation import (
+#    _compute_instance_segmentation_tendency,
+#    _compute_segmentation_tendency,
+# )
+from panoptica.metrics.relative_volume_difference import (
+    _compute_instance_relative_volume_difference,
+    _compute_relative_volume_difference,
+)
 from panoptica.metrics.iou import _compute_instance_iou, _compute_iou
 from panoptica.metrics.metrics import (
     Evaluation_List_Metric,

panoptica/metrics/assd.py

@@ -3,6 +3,31 @@
 from scipy.ndimage._nd_image import euclidean_feature_transform
 
 
+def _compute_instance_average_symmetric_surface_distance(
+    ref_labels: np.ndarray,
+    pred_labels: np.ndarray,
+    ref_instance_idx: int | None = None,
+    pred_instance_idx: int | None = None,
+    voxelspacing=None,
+    connectivity=1,
+):
+    if ref_instance_idx is None and pred_instance_idx is None:
+        return _average_symmetric_surface_distance(
+            reference=ref_labels,
+            prediction=pred_labels,
+            voxelspacing=voxelspacing,
+            connectivity=connectivity,
+        )
+    ref_instance_mask = ref_labels == ref_instance_idx
+    pred_instance_mask = pred_labels == pred_instance_idx
+    return _average_symmetric_surface_distance(
+        reference=ref_instance_mask,
+        prediction=pred_instance_mask,
+        voxelspacing=voxelspacing,
+        connectivity=connectivity,
+    )
+
+
 def _average_symmetric_surface_distance(
     reference,
     prediction,

panoptica/metrics/cldice.py

@@ -18,8 +18,8 @@ def cl_score(volume: np.ndarray, skeleton: np.ndarray):
 def _compute_centerline_dice(
     ref_labels: np.ndarray,
     pred_labels: np.ndarray,
-    ref_instance_idx: int,
-    pred_instance_idx: int,
+    ref_instance_idx: int | None = None,
+    pred_instance_idx: int | None = None,
 ) -> float:
     """Compute the centerline Dice (clDice) coefficient between a specific pair of instances.
 
@@ -32,6 +32,11 @@ def _compute_centerline_dice(
     Returns:
         float: clDice coefficient
     """
+    if ref_instance_idx is None and pred_instance_idx is None:
+        return _compute_centerline_dice_coefficient(
+            reference=ref_labels,
+            prediction=pred_labels,
+        )
     ref_instance_mask = ref_labels == ref_instance_idx
     pred_instance_mask = pred_labels == pred_instance_idx
     return _compute_centerline_dice_coefficient(

panoptica/metrics/dice.py

@@ -4,8 +4,8 @@
 def _compute_instance_volumetric_dice(
     ref_labels: np.ndarray,
     pred_labels: np.ndarray,
-    ref_instance_idx: int,
-    pred_instance_idx: int,
+    ref_instance_idx: int | None = None,
+    pred_instance_idx: int | None = None,
 ) -> float:
     """
     Compute the Dice coefficient between a specific pair of instances.
@@ -25,6 +25,11 @@ def _compute_instance_volumetric_dice(
         float: Dice coefficient between the specified instances. A value between 0 and 1, where higher values
         indicate better overlap and similarity between instances.
     """
+    if ref_instance_idx is None and pred_instance_idx is None:
+        return _compute_dice_coefficient(
+            reference=ref_labels,
+            prediction=pred_labels,
+        )
     ref_instance_mask = ref_labels == ref_instance_idx
     pred_instance_mask = pred_labels == pred_instance_idx
     return _compute_dice_coefficient(

panoptica/metrics/iou.py

@@ -4,8 +4,8 @@
 def _compute_instance_iou(
     reference_arr: np.ndarray,
     prediction_arr: np.ndarray,
-    ref_instance_idx: int,
-    pred_instance_idx: int,
+    ref_instance_idx: int | None = None,
+    pred_instance_idx: int | None = None,
 ) -> float:
     """
     Compute Intersection over Union (IoU) between a specific pair of reference and prediction instances.
@@ -19,6 +19,11 @@ def _compute_instance_iou(
     Returns:
         float: IoU between the specified instances.
     """
+    if ref_instance_idx is None and pred_instance_idx is None:
+        return _compute_iou(
+            reference_arr=reference_arr,
+            prediction_arr=prediction_arr,
+        )
     ref_instance_mask = reference_arr == ref_instance_idx
     pred_instance_mask = prediction_arr == pred_instance_idx
     return _compute_iou(ref_instance_mask, pred_instance_mask)

panoptica/metrics/metrics.py

@@ -5,10 +5,12 @@
 import numpy as np
 
 from panoptica.metrics import (
-    _average_symmetric_surface_distance,
-    _compute_centerline_dice_coefficient,
-    _compute_dice_coefficient,
-    _compute_iou,
+    _compute_instance_average_symmetric_surface_distance,
+    _compute_centerline_dice,
+    _compute_instance_volumetric_dice,
+    _compute_instance_iou,
+    _compute_instance_relative_volume_difference,
+    # _compute_instance_segmentation_tendency,
 )
 from panoptica.utils.constants import _Enum_Compare, auto
 
@@ -89,10 +91,12 @@ class Metric(_Enum_Compare):
         _type_: _description_
     """
 
-    DSC = _Metric("DSC", False, _compute_dice_coefficient)
-    IOU = _Metric("IOU", False, _compute_iou)
-    ASSD = _Metric("ASSD", True, _average_symmetric_surface_distance)
-    clDSC = _Metric("clDSC", False, _compute_centerline_dice_coefficient)
+    DSC = _Metric("DSC", False, _compute_instance_volumetric_dice)
+    IOU = _Metric("IOU", False, _compute_instance_iou)
+    ASSD = _Metric("ASSD", True, _compute_instance_average_symmetric_surface_distance)
+    clDSC = _Metric("clDSC", False, _compute_centerline_dice)
+    RVD = _Metric("RVD", True, _compute_instance_relative_volume_difference)
+    # ST = _Metric("ST", False, _compute_instance_segmentation_tendency)
 
     def __call__(
         self,
@@ -166,6 +170,8 @@ class MetricMode(_Enum_Compare):
     AVG = auto()
     SUM = auto()
     STD = auto()
+    MIN = auto()
+    MAX = auto()
 
 
 class MetricType(_Enum_Compare):
@@ -287,9 +293,18 @@ def __init__(
         if is_edge_case:
             self.AVG: float | None = edge_case_result
             self.SUM: None | float = edge_case_result
+            self.MIN: None | float = edge_case_result
+            self.MAX: None | float = edge_case_result
         else:
             self.AVG = None if self.ALL is None else np.average(self.ALL)
             self.SUM = None if self.ALL is None else np.sum(self.ALL)
+            self.MIN = (
+                None if self.ALL is None or len(self.ALL) == 0 else np.min(self.ALL)
+            )
+            self.MAX = (
+                None if self.ALL is None or len(self.ALL) == 0 else np.max(self.ALL)
+            )
+
         self.STD = (
             None
             if self.ALL is None

panoptica/metrics/relative_volume_difference.py

@@ -0,0 +1,70 @@
+import numpy as np
+
+
+def _compute_instance_relative_volume_difference(
+    ref_labels: np.ndarray,
+    pred_labels: np.ndarray,
+    ref_instance_idx: int | None = None,
+    pred_instance_idx: int | None = None,
+) -> float:
+    """
+    Compute the Dice coefficient between a specific pair of instances.
+
+    The Dice coefficient measures the similarity or overlap between two binary masks representing instances.
+    It is defined as:
+
+    Dice = (2 * intersection) / (ref_area + pred_area)
+
+    Args:
+        ref_labels (np.ndarray): Reference instance labels.
+        pred_labels (np.ndarray): Prediction instance labels.
+        ref_instance_idx (int): Index of the reference instance.
+        pred_instance_idx (int): Index of the prediction instance.
+
+    Returns:
+        float: Dice coefficient between the specified instances. A value between 0 and 1, where higher values
+        indicate better overlap and similarity between instances.
+    """
+    if ref_instance_idx is None and pred_instance_idx is None:
+        return _compute_relative_volume_difference(
+            reference=ref_labels,
+            prediction=pred_labels,
+        )
+    ref_instance_mask = ref_labels == ref_instance_idx
+    pred_instance_mask = pred_labels == pred_instance_idx
+    return _compute_relative_volume_difference(
+        reference=ref_instance_mask,
+        prediction=pred_instance_mask,
+    )
+
+
+def _compute_relative_volume_difference(
+    reference: np.ndarray,
+    prediction: np.ndarray,
+    *args,
+) -> float:
+    """
+    Compute the relative volume difference between two binary masks.
+
+    The relative volume difference is the predicted volume of an instance in relation to the reference volume (>0 oversegmented, <0 undersegmented)
+
+    RVD = ((pred_volume-ref_volume) / ref_volume)
+
+    Args:
+        reference (np.ndarray): Reference binary mask.
+        prediction (np.ndarray): Prediction binary mask.
+
+    Returns:
+        float: Relative volume Error between the two binary masks. A value between 0 and 1, where higher values
+        indicate better overlap and similarity between masks.
+    """
+    reference_mask = np.sum(reference)
+    prediction_mask = np.sum(prediction)
+
+    # Handle division by zero
+    if reference_mask == 0 and prediction_mask == 0:
+        return 0.0
+
+    # Calculate Dice coefficient
+    rvd = (prediction_mask - reference_mask) / reference_mask
+    return rvd

panoptica/panoptic_evaluator.py

@@ -18,6 +18,7 @@
 
 
 class Panoptic_Evaluator:
+
     def __init__(
         self,
         expected_input: (
@@ -26,7 +27,7 @@ def __init__(
         instance_approximator: InstanceApproximator | None = None,
         instance_matcher: InstanceMatchingAlgorithm | None = None,
         edge_case_handler: EdgeCaseHandler | None = None,
-        eval_metrics: list[Metric] = [Metric.DSC, Metric.IOU, Metric.ASSD],
+        eval_metrics: list[Metric] = [Metric.DSC, Metric.IOU, Metric.ASSD, Metric.RVD],
         decision_metric: Metric | None = None,
         decision_threshold: float | None = None,
         log_times: bool = False,

panoptica/panoptic_result.py

@@ -14,6 +14,7 @@
     _compute_centerline_dice_coefficient,
     _compute_dice_coefficient,
     _average_symmetric_surface_distance,
+    _compute_relative_volume_difference,
 )
 from panoptica.utils import EdgeCaseHandler
 
@@ -142,6 +143,14 @@ def __init__(
             global_bin_assd,
             long_name="Global Binary Average Symmetric Surface Distance",
         )
+        #
+        self.global_bin_rvd: int
+        self._add_metric(
+            "global_bin_rvd",
+            MetricType.GLOBAL,
+            global_bin_rvd,
+            long_name="Global Binary Relative Volume Difference",
+        )
         # endregion
         #
         # region IOU
@@ -232,6 +241,23 @@ def __init__(
             long_name="Segmentation Quality Assd Standard Deviation",
         )
         # endregion
+        #
+        # region RVD
+        self.sq_rvd: float
+        self._add_metric(
+            "sq_rvd",
+            MetricType.INSTANCE,
+            sq_rvd,
+            long_name="Segmentation Quality Relative Volume Difference",
+        )
+        self.sq_rvd_std: float
+        self._add_metric(
+            "sq_rvd_std",
+            MetricType.INSTANCE,
+            sq_rvd_std,
+            long_name="Segmentation Quality Relative Volume Difference Standard Deviation",
+        )
+        # endregion
 
         ##################
         # List Metrics   #
@@ -468,6 +494,18 @@ def sq_assd_std(res: PanopticaResult):
 # endregion
 
 
+# region RVD
+def sq_rvd(res: PanopticaResult):
+    return res.get_list_metric(Metric.RVD, mode=MetricMode.AVG)
+
+
+def sq_rvd_std(res: PanopticaResult):
+    return res.get_list_metric(Metric.RVD, mode=MetricMode.STD)
+
+
+# endregion
+
+
 # region Global
 def global_bin_dsc(res: PanopticaResult):
     if res.tp == 0:
@@ -499,6 +537,16 @@ def global_bin_assd(res: PanopticaResult):
     return _average_symmetric_surface_distance(ref_binary, pred_binary)
 
 
+def global_bin_rvd(res: PanopticaResult):
+    if res.tp == 0:
+        return 0.0
+    pred_binary = res._prediction_arr.copy()
+    ref_binary = res._reference_arr.copy()
+    pred_binary[pred_binary != 0] = 1
+    ref_binary[ref_binary != 0] = 1
+    return _compute_relative_volume_difference(ref_binary, pred_binary)
+
+
 # endregion
 
 

panoptica/utils/edge_case_handling.py

@@ -78,13 +78,18 @@ def __str__(self) -> str:
 
 
 class EdgeCaseHandler:
+
     def __init__(
         self,
         listmetric_zeroTP_handling: dict[Metric, MetricZeroTPEdgeCaseHandling] = {
             Metric.DSC: MetricZeroTPEdgeCaseHandling(
                 no_instances_result=EdgeCaseResult.NAN,
                 default_result=EdgeCaseResult.ZERO,
             ),
+            Metric.clDSC: MetricZeroTPEdgeCaseHandling(
+                no_instances_result=EdgeCaseResult.NAN,
+                default_result=EdgeCaseResult.ZERO,
+            ),
             Metric.IOU: MetricZeroTPEdgeCaseHandling(
                 no_instances_result=EdgeCaseResult.NAN,
                 empty_prediction_result=EdgeCaseResult.ZERO,
@@ -94,6 +99,10 @@ def __init__(
                 no_instances_result=EdgeCaseResult.NAN,
                 default_result=EdgeCaseResult.INF,
             ),
+            Metric.RVD: MetricZeroTPEdgeCaseHandling(
+                no_instances_result=EdgeCaseResult.NAN,
+                default_result=EdgeCaseResult.NAN,
+            ),
         },
         empty_list_std: EdgeCaseResult = EdgeCaseResult.NAN,
     ) -> None: