DICOM extractor rejects pydicom types

datalad_neuroimaging/extractors/dicom.py

@@ -29,12 +29,39 @@
 
 
 def _is_good_type(v):
-    if isinstance(v, (int, float, string_types)):
+    if isinstance(
+            v,
+            (int, float, string_types, dcm.valuerep.DSfloat, dcm.valuerep.IS,
+             dcm.valuerep.PersonName3)):
         return True
     elif isinstance(v, (list, tuple)):
         return all(map(_is_good_type, v))
+
+
+def _sanitize_unicode(s):
+    return s.replace(u"\u0000", "").strip()
+
+
+def _convert_value(v):
+    t = type(v)
+    if t in (int, float):
+        cv = v
+    elif t == str:
+        cv = _sanitize_unicode(v)
+    elif t == bytes:
+        s = v.decode('ascii', 'replace')
+        cv = _sanitize_unicode(s)
+    elif t == dcm.valuerep.DSfloat:
+        cv = float(v)
+    elif t == dcm.valuerep.IS:
+        cv = int(v)
+    elif t == dcm.valuerep.PersonName3:
+        cv = str(v)
+    elif isinstance(v, (list, tuple)):
+        cv = list(map(_convert_value, v))
     else:
-        return False
+        cv = v
+    return cv
 
 
 context = {
@@ -48,7 +75,7 @@ def _is_good_type(v):
 
 
 def _struct2dict(struct):
-    return {k: getattr(struct, k)
+    return {k: _convert_value(getattr(struct, k))
             for k in struct.dir()
             if hasattr(struct, k) and
             _is_good_type(getattr(struct, k))}

datalad_neuroimaging/extractors/tests/test_dicom.py

@@ -6,7 +6,7 @@
 #   copyright and license terms.
 #
 # ## ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ### ##
-"""Test audio extractor"""
+"""Test DICOM extractor"""
 
 from datalad.tests.utils import SkipTest
 try:
@@ -48,6 +48,11 @@ def test_dicom(path):
     # no point in testing ALL keys, but we got plenty
     assert(len(meta.keys()) > 70)
     eq_(meta['SeriesDate'], '20070205')
+    # make sure we have PatientName -- this is not using a basic data type, but
+    # dicom.valuerep.PersonName3 -- conversion should have handled that
+    # we can only test if the key is there, the source dicom has an empty
+    # string as value
+    eq_(meta['PatientName'], '')
 
     # now ask for the dataset metadata, which should have both the unique props
     # and a list of imageseries (one in this case, but a list)