Predicting optimal deep brain stimulation parameters for Parkinson’s disease using functional MRI and machine learning

Commonly used for Parkinson’s disease (PD), deep brain stimulation (DBS) produces marked clinical benefits when optimized. However, assessing the large number of possible stimulation settings (i.e., programming) requires numerous clinic visits. Here, we examine whether functional magnetic resonance imaging (fMRI) can be used to predict optimal stimulation settings for individual patients. We analyze 3 T fMRI data prospectively acquired as part of an observational trial in 67 PD patients using optimal and non-optimal stimulation settings. Clinically optimal stimulation produces a characteristic fMRI brain response pattern marked by preferential engagement of the motor circuit. Then, we build a machine learning model predicting optimal vs. non-optimal settings using the fMRI patterns of 39 PD patients with a priori clinically optimized DBS (88% accuracy). The model predicts optimal stimulation settings in unseen datasets: a priori clinically optimized and stimulation-naïve PD patients. We propose that fMRI brain responses to DBS stimulation in PD patients could represent an objective biomarker of clinical response. Upon further validation with additional studies, these findings may open the door to functional imaging-assisted DBS programming.

Knowing the story from the 1st submission I was enthusiastic to re-read this revised work as I strongly believe that this work has great potential for changing the field of DBS programming. However, I noticed some issues shortly after looking at the Result section. In fig 3A, the authors show three fMRI maps and claim that the optimal contact is the deepest (contact # 0, bottom row) while the figure legend states "top row" is the optimal location representing contact #3 (this happens to us all). More confusing is that the authors point to Supplementary Fig. S1 where they attempt to visualize in 3D the electrode and contact location of the same patient and there they show contact #3 (top contact; at least based on the shown VTA). And here is where it's really getting interesting, I went back and looked at my comments from the 1st review and there, FOR THE EXACT SAME PATIENT (identical fMRI maps), the Optimal contact was reported to be the dorsal contact (C#3). In fig. 3B where they compare the maps for optimal voltage, once again, for the exact same fMRI maps, now the active contact is C#1 (2nd from bottom) while in fig. 3 in the [Redacted] version it shows up as contact #2 (3rd from bottom). So, f or the same patient/data, two conflicting results are being presented.
Group level analysis shown in Supplementary Fig. S2: a) please show the same cortical slices across the motor region that a fair evaluation of the amplitude changes can be done. b) panel D of Supp. Fig.2, at supra-threshold voltage, the group analysis map shows only minor BOLD cortical changes while in Fig. 3, bottom row, there are significant changes, perhaps even larger than for the optimal voltage. [Redacted] Editorial Note: This manuscript has been previously reviewed at another journal that is not operating a transparent peer review scheme. This document only contains reviewer comments and rebuttal letters for versions considered at Nature Communications. Mentions of prior referee reports have been redacted.

Reviewers' Comments:
Reviewer #1 (Remarks to the Author): Authors have addressed all issues raised in the previous round of reviews (in a different journal). I have no further comments.
In my view this paper could truly become a corner-stone for a new clinical avenue. What if -in the future -DBS programming indeed would take place in the MRI? It is somewhat surprising that this hasn't been explored earlier (also little has been done along similar lines with EEG). Likely reason for fMRI is that scanning patients with electrodes is not straight-forward -here the Toronto group has made a lot of experience before carrying out this study. However, both DBS centers and electrode manufacturers alike would likely turn their head to this study and explore their options. Despite MRI being costly, the cost pales in comparison to tedious DBS programming procedures that often take >6 months and many man-hours to complete.
 Thank you for the careful review of our manuscript and for recognizing the potential impact of our work becoming a "corner-stone" for a new clinical avenue.

Reviewer #2 (Remarks to the Author):
The authors have addressed the specific concerns raised by the earlier manuscript. While this is ultimately an interesting proof-of concept study, it is still far removed from clinical needs. Currently, efforts are made to predict DBS settings through recordings of betaoscillations. Potential safety issues attendant to the directional electrodes will also need to be evaluated. Even so, the manuscript is fascinating and can be considered for publication in the journal.
 Thank you for the careful review of our manuscript and for agreeing that we have addressed the specific concerns. We also thank you for saying our work is fascinating and can be considered for publication.  We have incorporated details regarding potential safety issues attendant to directional electrodes in the text of the discussion The fMRI acquisitions we present here were done with omnidirectional electrode contacts and open loop stimulation based on careful safety testing. 7,9,14 As new electrodes and stimulation technologies emerge and their use become more widespread, including for example directional electrodes and closed loop DBS systems, the attendant safety and the impact on functional imaging with stimulation using these systems will also need to be evaluated (Discussion page 18).

[Redacted]
Knowing the story from the 1st submission I was enthusiastic to re-read this revised work as I strongly believe that this work has great potential for changing the field of DBS programming.
 Thank you for the careful review of our manuscript and for the kind comments.
However, I noticed some issues shortly after looking at the Result section. In fig 3A, the authors show three fMRI maps and claim that the optimal contact is the deepest (contact # 0, bottom row) while the figure legend states "top row" is the optimal location representing contact #3 (this happens to us all).
 Thank you for pointing out this discrepancy and the confusion that it generates. It arises as a consequence of a labeling error we made in the figure. In attempting to revise Figure  3 for NCOMMS-20-32931-T, we switched the rows without correctly addressing the VTA/electrode schematic on the right of the figure. The figure should show that the top (i.e. dorsal) contact is the optimal contact (as can be seen in Figure 3 in the original [Redacted] version). The corrected figure 3 attached below has the appropriate labelling. We apologize for the mislabeling in the previously revised version and the confusion this caused.
More confusing is that the authors point to Supplementary Fig. S1 where they attempt to visualize in 3D the electrode and contact location of the same patient and there they show contact #3 (top contact; at least based on the shown VTA).
 The Supplemental Figure in question S1A, B is correct, while Fig. 3A in the revised version had the error which has now been corrected as described above.
And here is where it's really getting interesting, I went back and looked at my comments from the 1st review and there, FOR THE EXACT SAME PATIENT (identical fMRI maps), the Optimal contact was reported to be the dorsal contact (C#3).
 As noted above, the optimal contact as shown in the [Redacted] version is correct (i.e. top/dorsal contact).
In fig. 3B where they compare the maps for optimal voltage, once again, for the exact same fMRI maps, now the active contact is C#1 (2nd from bottom) while in fig. 3 in the [Redacte d] version it shows up as contact #2 (3rd from bottom). So, for the same patient/data, two co nflicting results are being presented.
We have corrected these labelling errors in the figures below:  Again, this was a mistake that occurred during the revision process. The active contact in Fig. 3B should be contact #2 (3 rd from bottom), as it is in the [Redacted] version. In addition, we noted that Supplemental Fig. 1C, D (which is a 3D visualisation of the active contact in Fig. 3B) is also incorrect. As with Fig. 3B, the active contact should be contact #2 (3 rd from bottom). The updated corrected figures are incorporated below. The fMRI BOLD signal changes at the optimal contact (top row) and non-optimal contacts (middle and bottom rows) are shown. Brain regions with a significant increase (hot colors, positive t-values, DBS-ON>OFF) and decrease (cool colors, negative t-value, DBS-ON<OFF) (p<0·001, cluster size=50) in BOLD response were identified. We considered the clinically optimal contact as the origin (i.e., 0) and the nonoptimal contacts were mapped as a function of distance in mm from the optimal contact. The optimal contact showed change in BOLD response to be in the left (ipsilateral) motor cortex and thalamus, and right (contralateral) cerebellum. (B) BOLD response maps associated with left DBS-STN stimulation at multiple voltage settings for another a priori clinically optimized PD-STN patient. The figure shows the fMRI BOLD signal change at the optimal voltage (middle row) as well as at subtherapeutic (top row) and supratherapeutic (bottom row) doses. The subtherapeutic voltage was defined as 1.5 volt below optimal voltage because a reduction of this magnitude will yield a change in clinical status for most PD patients. The supratherapeutic voltage was defined as the voltage just below the side effects threshold (i.e., highest tolerated voltage). Hot and cool colors indicate significant positive and negative t-values (described in A).  Supplementary Fig. S2 has been modified and now shows cortical slices at the same level in the motor region.