def _get_most_aligned_peak(pred, orig):
orig = np.array(orig)
angle1 = abs(peak_utils.angle_last_dim(pred, orig[0]))
angle2 = abs(peak_utils.angle_last_dim(pred, orig[1]))
angle3 = abs(peak_utils.angle_last_dim(pred, orig[2]))
argmax = np.argmax(np.stack([angle1, angle2, angle3], axis=-1),
axis=-1)
x, y, z = (orig.shape[1], orig.shape[2], orig.shape[3])
return orig[tuple([argmax] + np.ogrid[:x, :y, :z])]
def calc_peak_length_dice(classes,
y_pred,
y_true,
max_angle_error=[0.9],
max_length_error=0.1):
score_per_bundle = {}
bundles = dataset_specific_utils.get_bundle_names(classes)[1:]
for idx, bundle in enumerate(bundles):
y_pred_bund = y_pred[:, :, :, (idx * 3):(idx * 3) + 3]
y_true_bund = y_true[:, :, :, (idx * 3):(idx * 3) + 3] # [x, y, z, 3]
angles = abs(peak_utils.angle_last_dim(y_pred_bund, y_true_bund))
lenghts_pred = np.linalg.norm(y_pred_bund, axis=-1)
lengths_true = np.linalg.norm(y_true_bund, axis=-1)
lengths_binary = abs(lenghts_pred - lengths_true) < (max_length_error *
lengths_true)
lengths_binary = lengths_binary.flatten()
gt_binary = y_true_bund.sum(axis=-1) > 0
gt_binary = gt_binary.flatten() # [bs*x*y]
angles_binary = angles > max_angle_error[0]
angles_binary = angles_binary.flatten()
combined = lengths_binary * angles_binary
f1 = my_f1_score(gt_binary, combined)
score_per_bundle[bundle] = f1
return score_per_bundle
def get_weighted_mean_of_peaks(best_orig, tom, weight=0.5):
"""
Calculate weighted mean between best_orig peaks and TOM peaks.
Args:
best_orig: original peaks
tom: prior
weight: how much to apply prior (0: only original signal, 1: only prior)
Returns:
weighted mean
"""
angles = peak_utils.angle_last_dim(best_orig, tom)
# make sure to take mean along smaller angle (<90 degree), not along the bigger one (>90 degree)
tom[angles < 0] *= -1 # flip peak
stacked = np.stack([best_orig, tom])
return np.average(stacked, axis=0, weights=[1 - weight, weight])
def calc_peak_dice(classes, y_pred, y_true, max_angle_error=[0.9]):
score_per_bundle = {}
bundles = dataset_specific_utils.get_bundle_names(classes)[1:]
for idx, bundle in enumerate(bundles):
y_pred_bund = y_pred[:, :, :, (idx * 3):(idx * 3) + 3]
y_true_bund = y_true[:, :, :, (idx * 3):(idx * 3) + 3] # (x,y,z,3)
angles = abs(peak_utils.angle_last_dim(y_pred_bund, y_true_bund))
angles_binary = angles > max_angle_error[0]
gt_binary = y_true_bund.sum(axis=-1) > 0
f1 = f1_score(gt_binary.flatten(),
angles_binary.flatten(),
average="binary")
score_per_bundle[bundle] = f1
return score_per_bundle