feat: Add debugging output to indicate what files are reused

Author: effigies

fmriprep/workflows/base.py

@@ -338,6 +338,9 @@ def init_single_subject_wf(subject_id: str):
 
     # allow to run with anat-fast-track on fMRI-only dataset
     if 't1w_preproc' in anatomical_cache and not subject_data['t1w']:
+        workflow.debug(
+            'No T1w image found; using precomputed T1w image: %s', anatomical_cache['t1w_preproc']
+        )
         workflow.connect([
             (bidssrc, bids_info, [(('bold', fix_multi_T1w_source_name), 'in_file')]),
             (anat_fit_wf, summary, [('outputnode.t1w_preproc', 't1w')]),
@@ -539,12 +542,14 @@ def init_single_subject_wf(subject_id: str):
         from fmriprep.utils.bids import collect_fieldmaps
 
         for deriv_dir in config.execution.derivatives.values():
-            fmap_cache.update(
-                collect_fieldmaps(
-                    derivatives_dir=deriv_dir,
-                    entities={'subject': subject_id},
-                )
+            fmaps = collect_fieldmaps(
+                derivatives_dir=deriv_dir,
+                entities={'subject': subject_id},
             )
+            config.loggers.workflow.debug(
+                'Detected precomputed fieldmaps in %d for fieldmap IDs: %s', deriv_dir, list(fmaps)
+            )
+            fmap_cache.update(fmaps)
 
     all_estimators, estimator_map = map_fieldmap_estimation(
         layout=config.execution.layout,
@@ -578,15 +583,17 @@ def init_single_subject_wf(subject_id: str):
                 f'{len(pared_cache)} estimator(s): {list(pared_cache)}.'
             )
 
-            fmap_buffers['fmap'].inputs.in1 = [
-                pared_cache[fmapid]['fieldmap'] for fmapid in pared_cache
-            ]
-            fmap_buffers['fmap_ref'].inputs.in1 = [
-                pared_cache[fmapid]['magnitude'] for fmapid in pared_cache
-            ]
-            fmap_buffers['fmap_coeff'].inputs.in1 = [
-                pared_cache[fmapid]['coeffs'] for fmapid in pared_cache
-            ]
+            fieldmaps = [fmap['fieldmap'] for fmap in pared_cache.values()]
+            refs = [fmap['magnitude'] for fmap in pared_cache.values()]
+            coeffs = [fmap['coeffs'] for fmap in pared_cache.values()]
+            config.loggers.workflow.debug('Reusing fieldmaps: %s', fieldmaps)
+            config.loggers.workflow.debug('Reusing references: %s', refs)
+            config.loggers.workflow.debug('Reusing coefficients: %s', coeffs)
+
+            fmap_buffers['fmap'].inputs.in1 = fieldmaps
+            fmap_buffers['fmap_ref'].inputs.in1 = refs
+            fmap_buffers['fmap_coeff'].inputs.in1 = coeffs
+
             # Note that masks are not emitted. The BOLD-fmap transforms cannot be
             # computed with precomputed fieldmaps until we either start emitting masks
             # or start skull-stripping references on the fly.

fmriprep/workflows/bold/fit.py

@@ -236,8 +236,8 @@ def init_bold_fit_wf(
     # Boolean used to update workflow self-descriptions
     multiecho = len(bold_series) > 1
 
-    have_hmcref = 'hmc_boldref' in precomputed
-    have_coregref = 'coreg_boldref' in precomputed
+    hmc_boldref = precomputed.get('hmc_boldref')
+    coreg_boldref = precomputed.get('coreg_boldref')
     # Can contain
     #  1) boldref2fmap
     #  2) boldref2anat
@@ -304,12 +304,19 @@ def init_bold_fit_wf(
         niu.IdentityInterface(fields=['boldref', 'boldmask']), name='regref_buffer'
     )
 
-    # Set default buffer values; can be overridden by workflows
-    hmcref_buffer.inputs.boldref = precomputed.get('hmc_boldref')
+    if hmc_boldref:
+        hmcref_buffer.inputs.boldref = hmc_boldref
+        config.loggers.workflow.debug('Reusing motion correction reference: %s', hmc_boldref)
+    if hmc_xforms:
+        hmc_buffer.inputs.hmc_xforms = hmc_xforms
+        config.loggers.workflow.debug('Reusing motion correction transforms: %s', hmc_xforms)
+    if boldref2fmap_xform:
+        fmapreg_buffer.inputs.boldref2fmap_xfm = boldref2fmap_xform
+        config.loggers.workflow.debug('Reusing BOLD-to-fieldmap transform: %s', boldref2fmap_xform)
+    if coreg_boldref:
+        regref_buffer.inputs.boldref = coreg_boldref
+        config.loggers.workflow.debug('Reusing coregistration reference: %s', coreg_boldref)
     fmapref_buffer.inputs.sbref_files = sbref_files
-    hmc_buffer.inputs.hmc_xforms = hmc_xforms
-    fmapreg_buffer.inputs.boldref2fmap_xfm = boldref2fmap_xform
-    regref_buffer.inputs.boldref = precomputed.get('coreg_boldref')
 
     summary = pe.Node(
         FunctionalSummary(
@@ -344,7 +351,7 @@ def init_bold_fit_wf(
 
     func_fit_reports_wf = init_func_fit_reports_wf(
         # TODO: Enable sdc report even if we find coregref
-        sdc_correction=not (have_coregref or fieldmap_id is None),
+        sdc_correction=not (coreg_boldref or fieldmap_id is None),
         freesurfer=config.workflow.run_reconall,
         output_dir=config.execution.fmriprep_dir,
     )
@@ -384,7 +391,7 @@ def init_bold_fit_wf(
         niu.IdentityInterface(fields=['in_file']),
         name='hmc_boldref_source_buffer',
     )
-    if not have_hmcref:
+    if not hmc_boldref:
         config.loggers.workflow.info('Stage 1: Adding HMC boldref workflow')
         hmc_boldref_wf = init_raw_boldref_wf(
             name='hmc_boldref_wf',
@@ -458,7 +465,7 @@ def init_bold_fit_wf(
 
     # Stage 3: Create coregistration reference
     # Fieldmap correction only happens during fit if this stage is needed
-    if not have_coregref:
+    if not coreg_boldref:
         config.loggers.workflow.info('Stage 3: Adding coregistration boldref workflow')
 
         # Select initial boldref, enhance contrast, and generate mask
@@ -600,7 +607,7 @@ def init_bold_fit_wf(
         # TODO: Allow precomputed bold masks to be passed
         # Also needs consideration for how it interacts above
         skullstrip_precomp_ref_wf = init_skullstrip_bold_wf(name='skullstrip_precomp_ref_wf')
-        skullstrip_precomp_ref_wf.inputs.inputnode.in_file = precomputed['coreg_boldref']
+        skullstrip_precomp_ref_wf.inputs.inputnode.in_file = coreg_boldref
         workflow.connect([
             (skullstrip_precomp_ref_wf, regref_buffer, [('outputnode.mask_file', 'boldmask')])
         ])  # fmt:skip