Why does simple imputer transform on scikit learn delete columns?

I try to imput all numerical data for a numerical dataset wit dataframe name features_numerical

features_numerical.shape give output (362, 253)

Here's what I did

from sklearn.impute import SimpleImputer
imput = SimpleImputer(missing_values=np.nan, strategy='median')
data_filled = pd.DataFrame(imput.transform(features_numerical))

but I lost a lot of columns

features_numerical.shape give output (362, 197)

Whats wrong wrong with this process