Fix website

conftest.py

@@ -6,6 +6,9 @@
 import numpy as np
 from phantoms.MR_XCAT_qMRI.sim_ivim_sig import phantom
 import warnings
+import os
+import torch
+import random
 from tests.IVIMmodels.unit_tests.test_ivim_fit import PerformanceWarning
 warnings.simplefilter("always", PerformanceWarning)
 
@@ -98,6 +101,21 @@ def pytest_addoption(parser):
         help="Run MATLAB-dependent tests"
     )
 
+def set_global_seed(seed: int = 42):
+    os.environ["PYTHONHASHSEED"] = str(seed)
+    random.seed(seed)
+    np.random.seed(seed)
+    torch.manual_seed(seed)
+    torch.cuda.manual_seed(seed)
+    torch.cuda.manual_seed_all(seed)
+    torch.backends.cudnn.deterministic = True
+    torch.backends.cudnn.benchmark = False
+    print(f"✅ Global seed set to {seed}")
+
+@pytest.fixture(autouse=True)
+def global_seed():
+    """Automatically seed all random generators at test session start."""
+    set_global_seed(42)
 
 @pytest.fixture(scope="session")
 def eng(request):

src/standardized/IVIM_NEToptim.py

@@ -88,7 +88,8 @@ def ivim_fit(self, signals, bvalues, **kwargs):
         """
         if not np.array_equal(bvalues, self.bvalues):
             raise ValueError("bvalue list at fitting must be identical as the one at initiation, otherwise it will not run")
-
+        if np.shape(np.shape(signals)) == (1,):
+            signals=signals[np.newaxis, :]
         paramsNN = deep.predict_IVIM(signals, self.bvalues, self.net, self.arg)
 
         results = {}

src/standardized/Super_IVIM_DC.py

@@ -99,12 +99,20 @@ def ivim_fit(self, signals, bvalues, **kwargs):
         """
         if not np.array_equal(bvalues, self.bvalues):
             raise ValueError("bvalue list at fitting must be identical as the one at initiation, otherwise it will not run")
-
+        if np.shape(np.shape(signals)) == (1,):
+            signals=signals[np.newaxis, :]
         Dp, Dt, f, S0_superivimdc = infer_from_signal(
             signal=signals,
             bvalues=self.bvalues,
             model_path=f"{self.working_dir}/{self.super_ivim_dc_filename}.pt",
         )
+        # fallback for empty arrays
+        if Dp.size == 0:
+            Dp = 0.0
+        if Dt.size == 0:
+            Dt = 0.0
+        if f.size == 0:
+            f = 0.0
 
         results = {}
         results["D"] = Dt

tests/IVIMmodels/unit_tests/algorithms.json

@@ -1,5 +1,6 @@
 {
     "algorithms": [
+        "Super_IVIM_DC",
         "IVIM_NEToptim",
         "TCML_TechnionIIT_lsqlm",
         "TCML_TechnionIIT_lsqtrf",
@@ -8,7 +9,6 @@
         "TCML_TechnionIIT_lsq_sls_trf",
         "TCML_TechnionIIT_lsq_sls_BOBYQA",
         "TCML_TechnionIIT_SLS",
-        "Super_IVIM_DC",
         "ASD_MemorialSloanKettering_QAMPER_IVIM",
         "ETP_SRI_LinearFitting",
         "IAR_LU_biexp",

tests/IVIMmodels/unit_tests/test_ivim_synthetic.py

@@ -7,6 +7,8 @@
 from src.wrappers.OsipiBase import OsipiBase
 from utilities.data_simulation.GenerateData import GenerateData
 
+TRAINED_MODELS = {}
+
 #run using pytest <path_to_this_file> --saveFileName test_output.txt --SNR 50 100 200
 #e.g. pytest -m slow tests/IVIMmodels/unit_tests/test_ivim_synthetic.py  --saveFileName test_output.csv --SNR 10 50 100 200 --fitCount 20
 @pytest.mark.slow
@@ -15,9 +17,6 @@ def test_generated(algorithmlist, ivim_data, SNR, rtol, atol, fit_count, rician_
     ivim_algorithm, requires_matlab, deep_learning = algorithmlist
     if requires_matlab and eng is None:
         pytest.skip(reason="Running without matlab; if Matlab is available please run pytest --withmatlab")
-    if deep_learning:
-        pytest.skip(
-            reason="Slow drifting in performance not yet implmented for deep learning algorithms")  # requieres training a network per b-value set and inferencing all data in 1 go. So not 1 data point per time, but all data in 1 go :). Otherwise network will be trained many many times...
     rng = np.random.RandomState(42)
     # random.seed(42)
     S0 = 1
@@ -26,7 +25,16 @@ def test_generated(algorithmlist, ivim_data, SNR, rtol, atol, fit_count, rician_
     D = data["D"]
     f = data["f"]
     Dp = data["Dp"]
-    fit = OsipiBase(algorithm=ivim_algorithm)
+    if deep_learning:
+        if ivim_algorithm+str(SNR) not in TRAINED_MODELS:
+            print(f"Training deep learning model {ivim_algorithm} ...")
+            fit = OsipiBase(bvalues=bvals, algorithm=ivim_algorithm,SNR=SNR)
+            TRAINED_MODELS[ivim_algorithm+str(SNR)] = fit
+        else:
+            print(f"Reusing trained model {ivim_algorithm}")
+            fit = TRAINED_MODELS[ivim_algorithm+str(SNR)]
+    else:
+        fit = OsipiBase(algorithm=ivim_algorithm)
     # here is a prior
     if use_prior and hasattr(fit, "supported_priors") and fit.supported_priors:
         prior = [rng.normal(D, D/3, 10), rng.normal(f, f/3, 10), rng.normal(Dp, Dp/3, 10), rng.normal(1, 1/3, 10)]

utilities/data_simulation/GenerateData.py

@@ -192,15 +192,14 @@ def simulate_training_data(self, bvalues, SNR = (5,100), n = 1000000, Drange = (
         #data_sim = np.zeros([len(D), len(bvalues)])
         bvalues = np.array(bvalues)
         if type(SNR) == tuple:
-            noise_std = 1/SNR[1] + test[:,3] * (1/SNR[0] - 1/SNR[1])
+            noise_std = (1 / SNR[1] + test[:, 3] * (1 / SNR[0] - 1 / SNR[1]))[:, None]
             addnoise = True
         elif SNR == 0:
             addnoise = False
             noise_std = np.ones((n, 1))
         else:
             noise_std = np.full((n, 1), 1/SNR)
             addnoise = True
-        noise_std = noise_std[:, np.newaxis]
         # loop over array to fill with simulated IVIM data
         bvalues = np.array(bvalues).reshape(1, -1)
         data_sim = 1 * (f * np.exp(-bvalues * Dp) + (1 - f) * np.exp(-bvalues * D))