In our proof of concept work two animals were used to test LOT imaging in the spinal cord (Ouakli, Guevara, Dubeau, Beaumont, & Lesage, 2010). Animals were anesthetized with
isoflurane (5%), and body temperature was maintained at 37°C with a feedback controlled heating blanket. Both heart rate and expired CO2 level were monitored. In addition to isoflurane
and before surgery the animals were anesthetized with 50 mg/kg bolus of alpha-chloralose (Bonvento et al., 1994). A tracheotomy was performed, and the rats were artificially ventilated with ambient air. Their breathing rate was maintained between 60 and 80 cycles/min, with approximately 2-ml tidal volume and set to obtain an end tidal expired CO2 concentration of 3%.
The rats were then positioned on a custom-made stereotaxic apparatus to immobilize the spinal cord and minimize movement artifacts due to both respiration and hind paw stimulation. Lumbar spinal cord segments from the thoracic (T10) to the sacral (S1) area are exposed by laminectomy and covered with mineral oil to prevent drying. Isoflurane administration was stopped after surgery and alpha-chloralose was administered at a rate of 40 mg/kg/h. Thirty minutes were allowed for animal stabilization. Then the spinal cord was firmly fastened by 2 sets of clamps to impede any longitudinal spinal cord movement induced by breathing. The skull was thinned to around 100 µm over the somatosensory cortex and blood was removed with a cotton swab moistened with saline solution. The skull was also covered with mineral oil before performing simultaneous intrinsic imaging.
The recordings were performed in sessions of 40 minutes. Each stimulation block consisted of 1- second duration with a randomized inter-stimulus interval of at least 20 seconds. The acquired measurements were then block averaged. The stimulus consisted of 3-ms current pulses at a 3 Hz repetition rate. Muscular movement threshold determined the three different stimulation intensities (0.9×, 1.2× and 1.5×), which were interlaced. Imaging protocol was made up of LOT recordings of the exposed spinal cord with simultaneous intrinsic imaging of the brain.
The source-detectors were lined parallel to the rostrocaudal axis (Figure 2-6B), imaging a FOV of 2.5×7.5mm of the lumbar area (L3 – L5 segments). Block averaged time course of the ipsilateral ROI (Figure 2-6A) revealed higher amplitudes with regard to contralateral ROI (Figure 2-6C); stimuli indicated with vertical blue lines. Contralateral activation also showed greater variance, displayed as gray shadows around the mean time trace. The amplitude of activation was proportional to stimuli intensity in both sides of the spinal cord. The blood is drained through the dorsal vein approximately one second after ipsilateral activation peak, as shown in the lower part of Figure 2-6B.
Figure 2-6: LOT hemodynamic response evoked by left hind paw stimulus intensity at 0.9× , 1.2× and 1.5× muscle threshold in normal rat. (A) ROI 1 averaged time traces for all detectors, ipsilateral to stimuli. (B) Top: Image of the exposed spinal cord. Bottom: ROI 2 averaged time course showing delay with regard to ipsilateral activation. (C) ROI 3 averaged time traces for all detectors, contralateral to stimuli.
The feasibility of concurrent OIS imaging was also investigated. Figure 2-7 depicts simultaneous imaging in the cortex and spinal cord. The behavior described previously by the curves in Figure
2-6 could be localized spatially. In panel A, an initial slight dip was localized on the dorsal vein at ~3.6s, then the activation increased on the left side of the spinal cord at ~7.20s, followed by blood drained through the dorsal vein at ~10.8s. In contrast to the spinal cord, where the hemodynamic response was ipsilateral to stimuli, the somatosensory cortex response was contralateral, as shown on Figure 2-7B, confirming the expectations. Note a blob of what appears to be ipsilateral activation on the left side of the cortex; this can be explained by the fact that an arbitrary threshold was chosen and no correction for false positives was done in the activation maps.
Figure 2-7: (A) Time course of LOT signals evoked by left hind paw stimulation collected at detector 1 over 15s at 0.9× muscular threshold. (B) Maximum intrinsic optical signal measured simultaneously on the somatosensory cortex.
The LOT data on each source-detector pair was integrated between 4.5 and 9 s post-stimulation (at 0.9× muscular threshold) to generate measurements for the inverse problem at activation peak.
Monte Carlo simulations (Boas et al., 2002) were used to model light propagation in the spinal cord using atlas-based anatomical information (Figure 2-8 B). A sensitivity matrix was generated to relate absorption changes to the measured data by using the first Born approximation (Kim & Schotland, 2006). A 3D map of neural activation evoked by hind limb stimulation was recovered by solving the inverse problem, with Tikhonov regularization, choosing the optimal regularization parameter value, found in the previous liquid phantoms reconstructions.
The reconstruction results are shown in panels A and C in Figure 2-8. The signal from the dorsal half of the spinal cord at z=0.4mm was representative of interneuron activation (Willis & Coggeshall, 1991), and in accordance with anatomical expectations for afferents from the sciatic nerve. Deeper in the gray matter the signal is more diffuse and contralateral activation is observed, that may originate from interneuron connections.
Figure 2-8: (A) 3D map of neural activation in the spinal cord induced by left hind paw stimulation at the 0.9× muscle threshold. Ipsilateral activation around z = 0.4 mm is consistent with interneuron activation. (B) Histology based segmented model of lumbar spinal cord of the rat, used for Monte Carlo simulation of light propagation. Dotted red lines indicate the extent of
field of view (C) Reconstruction viewed across the segmented volume along the dotted line in (A).
The proof of concept provided here demonstrates the concurrent use of LOT and OIS to image the spinal cord and the cortex respectively. This multimodal setup may pave the way to investigate neural plasticity after a spinal cord injury (SCI) and the adaptation of the cortex to this reorganization. Nevertheless depth limitations and lack of repeatable results due to poor SNR remain an issue. Another shortcoming of LOT is the reconstruction process involving an atlas template cord model for light propagation. Although spinal cord dimensions remain very similar for a given species at a certain age, the procedure is not optimal for generalized results. The ideal forward model would be produced from an anatomical MRI scan of each subject, which is expensive or spinal-cord histology along the imaged FOV, which is time-consuming.
To test the effect of errors in anatomical priors several models were constructed where the gray matter was displaced ±20% both horizontally and vertically, as shown on Figure 2-9. L2 metric was chosen to quantify errors in the reconstruction. Small errors (±20%) in the anatomical priors yield considerable errors in the reconstruction: ~80-100% greater than the reconstruction using the anatomically correct model. These results suggest that L-curve tuned Tikhonov regularization is highly sensitive to the a priori model for photon propagation and more robust techniques such as Bayesian reconstruction (Abdelnour, Genovese, & Huppert, 2010) should be investigated in future works.
Figure 2-9: (A) Light propagation models with the gray matter positioned in different places. (B) L2 reconstruction error resulting from incorrect anatomical priors.
2.5 Conclusions
In this proof of concept work, LOT has proved to be able to map hemodynamic response in three dimensions, proving sensitivity up to 1.5 mm in the rat spinal cord; however the SNR of the system did not allow imaging the responses of every specimen in a repeatable fashion.
I believe this could be alleviated with a redesign of the optical setup, placing a low-magnification objective with a higher NA and upgrading the rest of the optics to 2” pieces to gather more backscattered photons. Another aspect to improve is the resolution of the inverse problem; more robust techniques, such as Bayesian reconstructions could be applied instead of Tikhonov regularization, which is very sensitive to errors in the anatomical priors used as the propagation medium in the forward model.
Our initial aim was to observe the depth-resolved hemodynamic response in the spinal cord and the brain. Despite achieving a modest success, the technique is ill-posed and less suitable to unraveling resting-state networks. Therefore, the efforts of this thesis were channeled to alternative techniques, better adapted to the study of resting-state functional connectivity.