def preprocessing_pipeline(subject_num, task_num, standard_source_prefix): data_4d = prepare_standard_data(subject_num, task_num, standard_source_prefix) in_brain_mask, in_brain_vols = prepare_mask(data_4d, cutoff) data_4d_smoothed = spatial_smooth(data_4d, in_brain_mask, 2.0, 2.0, False) in_brain_mask, in_brain_vols = prepare_mask(data_4d_smoothed, cutoff) residuals = first_pcs_removed(in_brain_vols, 2) return residuals, in_brain_mask
def preprocessing_pipeline(subject_num, task_num, standard_source_prefix,
cond_filepath_prefix):
img = prepare_standard_img(subject_num, task_num, standard_source_prefix)
data = img.get_data()[..., 5:]
n_trs = data.shape[-1] + 5
cond_filename_003 = form_cond_filepath(subject_num, task_num, "003",
cond_filepath_prefix)
cond_filename_005 = form_cond_filepath(subject_num, task_num, "005",
cond_filepath_prefix)
cond_filename_001 = form_cond_filepath(subject_num, task_num, "001",
cond_filepath_prefix)
cond_filename_004 = form_cond_filepath(subject_num, task_num, "004",
cond_filepath_prefix)
cond_filename_007 = form_cond_filepath(subject_num, task_num, "007",
cond_filepath_prefix)
target_convolved, nontarget_convolved, error_convolved = conv_target_non_target(
n_trs, cond_filename_003, cond_filename_007, TR, tr_divs=100.0)
target_convolved, nontarget_convolved, error_convolved = target_convolved[
5:], nontarget_convolved[5:], error_convolved[5:]
block_regressor = events2neural_std(cond_filename_005, TR, n_trs)[5:]
block_start_cues = conv_std(n_trs, cond_filename_001, TR)[5:]
block_end_cues = conv_std(n_trs, cond_filename_004, TR)[5:]
linear_drift = np.linspace(-1, 1, n_trs)
qudratic_drift = linear_drift**2
qudratic_drift -= np.mean(qudratic_drift)
linear_drift = linear_drift[5:]
qudratic_drift = qudratic_drift[5:]
in_brain_mask, _ = prepare_mask(data, CUTOFF)
pad_thickness = 2.0
sigma = 2.0
b_vols = spatial_smooth(data, in_brain_mask, pad_thickness, sigma, False)
in_brain_tcs = b_vols[in_brain_mask]
Y = in_brain_tcs.T
Y_demeaned = Y - np.mean(Y, axis=1).reshape([-1, 1])
unscaled_cov = Y_demeaned.dot(Y_demeaned.T)
U, S, V = npl.svd(unscaled_cov)
n_betas = 10
X = np.ones((n_trs - 5, n_betas))
X[:, 0] = target_convolved
X[:, 1] = nontarget_convolved
X[:, 2] = error_convolved
X[:, 3] = block_regressor
X[:, 4] = block_start_cues
X[:, 5] = block_end_cues
X[:, 6] = linear_drift
X[:, 7] = qudratic_drift
X[:, 8] = U[:, 0]
# 9th column is the intercept
B = npl.pinv(X).dot(Y)
residuals = in_brain_tcs - X.dot(B).T
B[(3, 4, 5, 6, 7, 8, 9), :] = 0
# project Y onto the functional betas
functional_Y = X.dot(B).T
b_vols = np.zeros((data.shape))
b_vols[in_brain_mask, :] = functional_Y + residuals
return b_vols, img, in_brain_mask
def single_subject_linear_model(standard_source_prefix, cond_filepath_prefix, subject_num, task_num, output_filename):
data = prepare_standard_data(subject_num, task_num, standard_source_prefix)
n_trs = data.shape[-1] + 5
cond_filename_003 = form_cond_filepath(subject_num, task_num, "003", cond_filepath_prefix)
cond_filename_005 = form_cond_filepath(subject_num, task_num, "005", cond_filepath_prefix)
cond_filename_001 = form_cond_filepath(subject_num, task_num, "001", cond_filepath_prefix)
cond_filename_004 = form_cond_filepath(subject_num, task_num, "004", cond_filepath_prefix)
cond_filename_007 = form_cond_filepath(subject_num, task_num, "007", cond_filepath_prefix)
target_convolved, nontarget_convolved, error_convolved = conv_target_non_target(n_trs, cond_filename_003, cond_filename_007, TR, tr_divs = 100.0)
target_convolved, nontarget_convolved, error_convolved = target_convolved[5:], nontarget_convolved[5:], error_convolved[5:]
block_regressor = events2neural_std(cond_filename_005, TR, n_trs)[5:]
block_start_cues = conv_std(n_trs, cond_filename_001, TR)[5:]
block_end_cues = conv_std(n_trs, cond_filename_004, TR)[5:]
linear_drift = np.linspace(-1, 1, n_trs)
qudratic_drift = linear_drift ** 2
qudratic_drift -= np.mean(qudratic_drift)
linear_drift = linear_drift[5:]
qudratic_drift = qudratic_drift[5:]
in_brain_mask, _ = prepare_mask(data, 5000)
pad_thickness = 2.0
sigma = 2.0
b_vols = spatial_smooth(data, in_brain_mask, pad_thickness, sigma, False)
in_brain_tcs = b_vols[in_brain_mask]
Y = in_brain_tcs.T
Y_demeaned = Y - np.mean(Y, axis=1).reshape([-1, 1])
unscaled_cov = Y_demeaned.dot(Y_demeaned.T)
U, S, V = npl.svd(unscaled_cov)
n_betas = 11
X = np.ones((n_trs - 5, n_betas))
X[:, 0] = target_convolved
X[:, 1] = nontarget_convolved
X[:, 2] = error_convolved
X[:, 3] = block_regressor
X[:, 4] = block_start_cues
X[:, 5] = block_end_cues
X[:, 6] = linear_drift
X[:, 7] = qudratic_drift
X[:, 8] = U[:,0]
X[:, 9] = U[:,1]
# 10th column is the intercept
# plot design matrix
plt.figure()
plt.imshow(X, aspect=0.1)
plt.savefig(os.path.join(output_filename, "sub%s_task%s_design_matrix.png" % (subject_num, task_num)), format='png', dpi=500)
B = npl.pinv(X).dot(Y)
# test normality of residuals
residuals = Y.T - X.dot(B).T
alpha_test, bonferroni_test, hochberg_test, benjamini_test = [val * 1.0 / Y.shape[-1] for val in multiple_comp(residuals)]
normality_test_results = {"Alpha Test":alpha_test, "Bonferroni Procedure":bonferroni_test,"Hochberg Procedure":hochberg_test,"Benjamini-Hochberg Procedure":benjamini_test}
normality_test_pd = pd.DataFrame(normality_test_results, index=["Failure Rate"])
normality_test_pd.to_csv(os.path.join(output_filename, "sub%s_linear_model_normality_tests_failure_rates.csv" % subject_num))
rs_squared = []
for i in range(Y.shape[-1]):
r_squared = 1 - np.sum((Y[:,i] - X.dot(B[:,i]))**2) * 1.0 / np.sum((Y[:,i] - np.mean(Y[:,i])) ** 2)
rs_squared.append(r_squared)
np.savetxt(os.path.join(output_filename, "glm_mean_R_squared_" + ("0_back" if task_num == "001" else "2_back") + ".txt"), np.array([np.mean(rs_squared)]))
b_vols = np.zeros((data.shape[0:-1] + (n_betas,)))
b_vols[in_brain_mask, :] = B.T
# compute t values for target and nontarget betas
t_test_target_beta = 0
t_values = compute_t_values(X, B, Y, t_test_target_beta)
t_vols_beta_0 = np.zeros((data.shape[0:-1]))
t_vols_beta_0[in_brain_mask] = t_values
t_test_target_beta = 1
t_values = compute_t_values(X, B, Y, t_test_target_beta)
t_vols_beta_1 = np.zeros((data.shape[0:-1]))
t_vols_beta_1[in_brain_mask] = t_values
# compute t values for noise regressor betas
t_values = compute_t_values(X, B, Y, 6)
t_vols_beta_6 = np.zeros((data.shape[0:-1]))
t_vols_beta_6[in_brain_mask] = t_values
t_values = compute_t_values(X, B, Y, 7)
t_vols_beta_7 = np.zeros((data.shape[0:-1]))
t_vols_beta_7[in_brain_mask] = t_values
t_values = compute_t_values(X, B, Y, 8)
t_vols_beta_8 = np.zeros((data.shape[0:-1]))
t_vols_beta_8[in_brain_mask] = t_values
t_values = compute_t_values(X, B, Y, 9)
t_vols_beta_9 = np.zeros((data.shape[0:-1]))
t_vols_beta_9[in_brain_mask] = t_values
t_vols_beta_6_to_9 = [t_vols_beta_6, t_vols_beta_7, t_vols_beta_8, t_vols_beta_9]
return b_vols, in_brain_mask, U, Y, data, t_vols_beta_0, t_vols_beta_1, t_vols_beta_6_to_9
def preprocessing_pipeline(subject_num, task_num, standard_source_prefix, cond_filepath_prefix): img = prepare_standard_img(subject_num, task_num, standard_source_prefix) data = img.get_data()[..., 5:] n_trs = data.shape[-1] + 5 cond_filename_003 = form_cond_filepath(subject_num, task_num, "003", cond_filepath_prefix) cond_filename_005 = form_cond_filepath(subject_num, task_num, "005", cond_filepath_prefix) cond_filename_001 = form_cond_filepath(subject_num, task_num, "001", cond_filepath_prefix) cond_filename_004 = form_cond_filepath(subject_num, task_num, "004", cond_filepath_prefix) cond_filename_007 = form_cond_filepath(subject_num, task_num, "007", cond_filepath_prefix) target_convolved, nontarget_convolved, error_convolved = conv_target_non_target(n_trs, cond_filename_003, cond_filename_007, TR, tr_divs = 100.0) target_convolved, nontarget_convolved, error_convolved = target_convolved[5:], nontarget_convolved[5:], error_convolved[5:] block_regressor = events2neural_std(cond_filename_005, TR, n_trs)[5:] block_start_cues = conv_std(n_trs, cond_filename_001, TR)[5:] block_end_cues = conv_std(n_trs, cond_filename_004, TR)[5:] linear_drift = np.linspace(-1, 1, n_trs) qudratic_drift = linear_drift ** 2 qudratic_drift -= np.mean(qudratic_drift) linear_drift = linear_drift[5:] qudratic_drift = qudratic_drift[5:] in_brain_mask, _ = prepare_mask(data, CUTOFF) pad_thickness = 2.0 sigma = 2.0 b_vols = spatial_smooth(data, in_brain_mask, pad_thickness, sigma, False) in_brain_tcs = b_vols[in_brain_mask] Y = in_brain_tcs.T Y_demeaned = Y - np.mean(Y, axis=1).reshape([-1, 1]) unscaled_cov = Y_demeaned.dot(Y_demeaned.T) U, S, V = npl.svd(unscaled_cov) n_betas = 10 X = np.ones((n_trs - 5, n_betas)) X[:, 0] = target_convolved X[:, 1] = nontarget_convolved X[:, 2] = error_convolved X[:, 3] = block_regressor X[:, 4] = block_start_cues X[:, 5] = block_end_cues X[:, 6] = linear_drift X[:, 7] = qudratic_drift X[:, 8] = U[:,0] # 9th column is the intercept B = npl.pinv(X).dot(Y) residuals = in_brain_tcs - X.dot(B).T B[(3,4,5,6,7,8,9),:] = 0 # project Y onto the functional betas functional_Y = X.dot(B).T b_vols = np.zeros((data.shape)) b_vols[in_brain_mask, :] = functional_Y + residuals return b_vols, img, in_brain_mask
def single_subject_linear_model(standard_source_prefix, cond_filepath_prefix, subject_num, task_num, output_filename):
data = prepare_standard_data(subject_num, task_num, standard_source_prefix)
n_trs = data.shape[-1] + 5
cond_filename_003 = form_cond_filepath(subject_num, task_num, "003", cond_filepath_prefix)
cond_filename_005 = form_cond_filepath(subject_num, task_num, "005", cond_filepath_prefix)
cond_filename_001 = form_cond_filepath(subject_num, task_num, "001", cond_filepath_prefix)
cond_filename_004 = form_cond_filepath(subject_num, task_num, "004", cond_filepath_prefix)
cond_filename_007 = form_cond_filepath(subject_num, task_num, "007", cond_filepath_prefix)
target_convolved, nontarget_convolved, error_convolved = conv_target_non_target(n_trs, cond_filename_003, cond_filename_007, TR, tr_divs = 100.0)
target_convolved, nontarget_convolved, error_convolved = target_convolved[5:], nontarget_convolved[5:], error_convolved[5:]
block_regressor = events2neural_std(cond_filename_005, TR, n_trs)[5:]
block_start_cues = conv_std(n_trs, cond_filename_001, TR)[5:]
block_end_cues = conv_std(n_trs, cond_filename_004, TR)[5:]
linear_drift = np.linspace(-1, 1, n_trs)
qudratic_drift = linear_drift ** 2
qudratic_drift -= np.mean(qudratic_drift)
linear_drift = linear_drift[5:]
qudratic_drift = qudratic_drift[5:]
in_brain_mask, _ = prepare_mask(data, 5000)
pad_thickness = 2.0
sigma = 2.0
b_vols = spatial_smooth(data, in_brain_mask, pad_thickness, sigma, False)
in_brain_tcs = b_vols[in_brain_mask]
Y = in_brain_tcs.T
Y_demeaned = Y - np.mean(Y, axis=1).reshape([-1, 1])
unscaled_cov = Y_demeaned.dot(Y_demeaned.T)
U, S, V = npl.svd(unscaled_cov)
n_betas = 11
X = np.ones((n_trs - 5, n_betas))
X[:, 0] = target_convolved
X[:, 1] = nontarget_convolved
X[:, 2] = error_convolved
X[:, 3] = block_regressor
X[:, 4] = block_start_cues
X[:, 5] = block_end_cues
X[:, 6] = linear_drift
X[:, 7] = qudratic_drift
X[:, 8] = U[:,0]
X[:, 9] = U[:,1]
# 10th column is the intercept
# plot design matrix
plt.figure()
plt.imshow(X, aspect=0.1)
plt.savefig(os.path.join(output_filename, "sub%s_task%s_design_matrix.png" % (subject_num, task_num)), format='png', dpi=500)
B = npl.pinv(X).dot(Y)
# test normality of residuals
residuals = Y.T - X.dot(B).T
alpha_test, bonferroni_test, hochberg_test, benjamini_test = [val * 1.0 / Y.shape[-1] for val in multiple_comp(residuals)]
normality_test_results = {"Alpha Test":alpha_test, "Bonferroni Procedure":bonferroni_test,"Hochberg Procedure":hochberg_test,"Benjamini-Hochberg Procedure":benjamini_test}
normality_test_pd = pd.DataFrame(normality_test_results, index=["Failure Rate"])
normality_test_pd.to_csv(os.path.join(output_filename, "sub%s_task%s_linear_model_normality_tests_failure_rates.csv" % (subject_num, task_num)))
rs_squared = []
for i in range(Y.shape[-1]):
r_squared = 1 - np.sum((Y[:,i] - X.dot(B[:,i]))**2) * 1.0 / np.sum((Y[:,i] - np.mean(Y[:,i])) ** 2)
rs_squared.append(r_squared)
np.savetxt(os.path.join(output_filename, "glm_mean_R_squared_" + ("0_back" if task_num == "001" else "2_back") + ".txt"), np.array([np.mean(rs_squared)]))
b_vols = np.zeros((data.shape[0:-1] + (n_betas,)))
b_vols[in_brain_mask, :] = B.T
# compute t values for target and nontarget betas
t_test_target_beta = 0
t_values = compute_t_values(X, B, Y, t_test_target_beta)
t_vols_beta_0 = np.zeros((data.shape[0:-1]))
t_vols_beta_0[in_brain_mask] = t_values
t_test_target_beta = 1
t_values = compute_t_values(X, B, Y, t_test_target_beta)
t_vols_beta_1 = np.zeros((data.shape[0:-1]))
t_vols_beta_1[in_brain_mask] = t_values
# compute t values for noise regressor betas
t_values = compute_t_values(X, B, Y, 6)
t_vols_beta_6 = np.zeros((data.shape[0:-1]))
t_vols_beta_6[in_brain_mask] = t_values
t_values = compute_t_values(X, B, Y, 7)
t_vols_beta_7 = np.zeros((data.shape[0:-1]))
t_vols_beta_7[in_brain_mask] = t_values
t_values = compute_t_values(X, B, Y, 8)
t_vols_beta_8 = np.zeros((data.shape[0:-1]))
t_vols_beta_8[in_brain_mask] = t_values
t_values = compute_t_values(X, B, Y, 9)
t_vols_beta_9 = np.zeros((data.shape[0:-1]))
t_vols_beta_9[in_brain_mask] = t_values
t_vols_beta_6_to_9 = [t_vols_beta_6, t_vols_beta_7, t_vols_beta_8, t_vols_beta_9]
return b_vols, in_brain_mask, U, Y, data, t_vols_beta_0, t_vols_beta_1, t_vols_beta_6_to_9
