diff --git a/monai/metrics/confusion_matrix.py b/monai/metrics/confusion_matrix.py
index 67eff3451e..9157f9c3b9 100644
--- a/monai/metrics/confusion_matrix.py
+++ b/monai/metrics/confusion_matrix.py
@@ -178,6 +178,12 @@ def get_confusion_matrix(
         ValueError: when `y_pred` and `y` have different shapes.
     """
 
+    # bool + bool is a logical OR, so the `(y_pred + y) == 2` test below never matches and loses every true positive.
+    if y_pred.dtype == torch.bool:
+        y_pred = y_pred.float()
+    if y.dtype == torch.bool:
+        y = y.float()
+
     original_y = y
     if not include_background:
         y_pred, y = ignore_background(y_pred=y_pred, y=y)
diff --git a/monai/metrics/f_beta_score.py b/monai/metrics/f_beta_score.py
index bb9371c8bf..5b40322323 100644
--- a/monai/metrics/f_beta_score.py
+++ b/monai/metrics/f_beta_score.py
@@ -61,6 +61,13 @@ def aggregate(
 
 
 def get_f_beta_score(y_pred: torch.Tensor, y: torch.Tensor, include_background: bool = True) -> torch.Tensor:
+    # The true positive/negative counts below test `y_pred + y`; for bool tensors that sum is a logical OR,
+    # which never equals 2, so every true positive would be lost. Count on float values instead.
+    if y_pred.dtype == torch.bool:
+        y_pred = y_pred.float()
+    if y.dtype == torch.bool:
+        y = y.float()
+
     if not include_background:
         y_pred, y = ignore_background(y_pred=y_pred, y=y)
 
diff --git a/tests/metrics/test_compute_confusion_matrix.py b/tests/metrics/test_compute_confusion_matrix.py
index 3b7f793323..406338685a 100644
--- a/tests/metrics/test_compute_confusion_matrix.py
+++ b/tests/metrics/test_compute_confusion_matrix.py
@@ -502,6 +502,18 @@ def test_unknown_metric_name(self):
         with self.assertRaises(NotImplementedError):
             compute_confusion_matrix_metric("diagnostic likelihood ratio", TEST_CASE_LR[0])
 
+    def test_bool_inputs_match_float_inputs(self):
+        """Bool masks (e.g. ``pred > 0.5``) must give the same confusion matrix as their float equivalents."""
+        generator = torch.Generator().manual_seed(0)
+        y_pred = torch.rand(2, 3, 8, 8, generator=generator) > 0.5
+        y = torch.rand(2, 3, 8, 8, generator=generator) > 0.5
+
+        for include_background in (True, False):
+            expected = get_confusion_matrix(y_pred.float(), y.float(), include_background=include_background)
+            result = get_confusion_matrix(y_pred, y, include_background=include_background)
+            assert_allclose(result, expected, type_test=False)
+            self.assertTrue(bool((result[..., 0] > 0).all()))  # true positives are counted
+
 
 if __name__ == "__main__":
     unittest.main()
diff --git a/tests/metrics/test_compute_f_beta.py b/tests/metrics/test_compute_f_beta.py
index 21a4e3c22c..b207b0227c 100644
--- a/tests/metrics/test_compute_f_beta.py
+++ b/tests/metrics/test_compute_f_beta.py
@@ -18,6 +18,7 @@
 from parameterized import parameterized
 
 from monai.metrics import FBetaScore
+from monai.metrics.f_beta_score import get_f_beta_score
 from tests.test_utils import assert_allclose
 
 _device = "cuda:0" if torch.cuda.is_available() else "cpu"
@@ -75,6 +76,22 @@ def test_prediction_and_result_have_different_shape(self):
         with self.assertRaises(ValueError):
             metric(y_pred=torch.Tensor([[1, 1, 1], [1, 1, 1]]), y=torch.Tensor([1, 1, 1]))
 
+    def test_bool_inputs_match_float_inputs(self):
+        """Bool masks (e.g. ``pred > 0.5``) must give the same counts and score as their float equivalents."""
+        generator = torch.Generator().manual_seed(0)
+        y_pred = torch.rand(2, 3, 8, 8, generator=generator) > 0.5
+        y = torch.rand(2, 3, 8, 8, generator=generator) > 0.5
+
+        for include_background in (True, False):
+            expected = get_f_beta_score(y_pred.float(), y.float(), include_background=include_background)
+            result = get_f_beta_score(y_pred, y, include_background=include_background)
+            assert_allclose(result, expected, type_test=False)
+            self.assertTrue(bool((result[..., 0] > 0).all()))  # true positives are counted
+
+        metric = FBetaScore()
+        metric(y_pred=y, y=y)
+        assert_allclose(metric.aggregate()[0], torch.Tensor([1.0]), atol=1e-7, rtol=1e-7)
+
 
 if __name__ == "__main__":
     unittest.main()