This work demonstrates the first successful attempt at acquiring CARS and DRS spectra of brain tissue for human DBS surgery. The DRS and CARS spectra from the optical probe can adequately differentiate WM and GM, as presented in Fig. 4C. The optical measurements generally agree with the tissue classification from the HISTO barcodes for the same trajectory. However, small differences between the HISTO and optical barcodes remain. Due to the long postmortem delay and the lack of tissue fixation, the insertion of the optical probe may have caused brain compression and shifts. This may in turn lead to incorrect associations between tissue identification and the probe position, as well as mismatched reference and optical barcodes. In addition, some basal ganglia components such as the striatum, the GPi and the GPe, contain bundles of myelinated axons (WM) that travel through the GM. Therefore, these brain regions could be interpreted as mixed matter. Since our analysis clustered the data into 2 groups and did not consider mixed matter, both the histology and optical identification could have incorrectly labeled the mixed matter as WM or GM. Although the CARS and DRS barcodes for each trajectory have similarities, we did not expect them to be identical since the entry point on the surface of the brain is different, as presented in Fig. 1B. Despite these shortcomings, comparing all 3 barcodes (STN, GPi and OFF) for one optical modality (either DRS or CARS) shows visible differences, suggesting that the tissue composition for a specific trajectory is unique and can be recognized by our optical probe.
The barcode correspondence between optical and MRI barcodes was poor. This is because the MRI images do not have the contrast and resolution required to identify tissue type with pixel intensity alone. The lack of contrast in the basal ganglia in Fig. 4A demonstrates that anatomical landmarks are essential to infer the location of small and uncontrasted regions, such as the STN. While MRI is an indispensable imaging tool for planning DBS surgery, identifying brain regions rely on the surgeon’s knowledge of neuroanatomy.
Our experiment presents a proof-of-concept of optical measurements for tissue identification in the human brain, demonstrating its potential as a real-time DBS guidance tool for neurosurgeons. More measurements should be collected in clinical trials to assess the variability of the barcodes for specific targets. Our group is also investigating the use of other modalities, such as polarization-sensitive optical coherence tomography, as it measures the birefringence of the myelin31. We hypothesize that with more data and advanced analysis algorithms, optical barcodes could reflect MRI data. Our tool could provide real-time insights on lead insertion accuracy as the neurosurgeon performs the procedure.
The findings of this research indicate that DRS is a simple and minimally invasive optical method that could provide additional insight during DBS lead insertion. The PCA analysis revealed that the tissue identification is mostly accomplished by PC1 alone. The first 4 PCs from the L-OFF trajectory (DRS) are shown in Fig. 5A. PC1 for the L-OFF trajectory can be visualized as a straight line across all wavelengths, showing that the variance between the spectra acquired along the trajectory does not depend on the spectral information. This indicates that tissue identification is obtained only by the difference in light intensity that is scattered by the tissue. This observation is consistent with the inherent properties of brain tissue, where WM scatters more light than the darker GM tissue. It also suggests that measuring intensity of the light reflected at a single wavelength in the visible range, from the tissue in front of the DBS lead, could be sufficient to differentiate WM and GM using the analytical method we provided here.
Considering the wealth of information that is often encoded in an optical spectrum, looking at other PCs beyond PC1 might have the potential to improve tissue identification, as suggested in Fig. 5A. The absorption spectrum of blood (deoxyhemoglobin and oxyhemoglobin) is shown in Fig. 5B for comparison. The 2 peaks observed in PC4 approximately correspond to the 2 peaks found in the oxyhemoglobin absorption spectrum. The wider peak in PC3 is approximately at the same wavelength as the main peak found in the deoxyhemoglobin absorption spectrum. However, literature shows that oxyhemoglobin concentration decreases by 90% within 7h of death32. Therefore, it is not possible to conclude whether the optical probe detected very low concentrations of oxygenated blood or another unknown compound. Nevertheless, considering that oxyhemoglobin and deoxyhemoglobin are abundant in living subjects, it can be inferred that the spectroscopic measurements might reveal the presence of blood in the brain when inserting the optical probe. This could help prevent hemorrhaging and further refine the tissue identification. However, since this experiment was conducted on a postmortem brain, it was not possible to confirm the plausibility of this hypothesis.
CARS showed promising results at tissue identification for PC1, which represents the intensity of the myelin signal in the tissue. However, attempts to use subsequent PCs for the analysis were unsuccessful despite using 10 mW of power and 8-second acquisitions at each depth along the trajectories. Indeed, only a very weak signal was detected and the signal-to-noise ratio was poor, revealing essentially no visible features in the spectral decomposition. Simple improvements could be made to our system to significantly enhance the CARS measurements quality. For example, since the laser wavelength is swept, the system could target specific vibrational modes that differentiate between WM and GM, such as 2845 cm-1 or 2880 cm-1, which are abundant in myelin. This way, a spectrum would only have a handful of relevant points, which would increase the signal per wavelength with the same acquisition time by more than one order of magnitude. With this simple change, our probe could acquire a spectrum in under a second. Other technical improvements, such as optimizing the probe design for more efficient signal collection by adding collection fibers, could increase the signal-to-noise ratio.
For the first time, we have acquired CARS and DRS spectra in the brain of a human cadaver and have demonstrated how they can be used to identify tissue, offering a promising avenue for DBS surgical guidance. The spectra were acquired using a custom-built optical probe that was enclosed in a commercial DBS lead. The optical probe was then inserted, targeting 3 specific regions in the right and left hemispheres of the brain, for a total of 6 insertions. We observed that a barcode obtained from PCA and k-means clustering along a distinct trajectory appears unique. This demonstrates that the position of the lead could be followed in real time during its descent in the brain. In addition, DRS might be able to identify the presence of blood in the tissue to provide further guidance and prevent hemorrhages during lead insertion. More work is needed to refine the optical system to obtain higher quality spectral acquisitions and to design a probe that is suitable for trials in living patients. Finally, a rigorous strategy to map the optical barcode onto the preoperative MRI scans could provide the neurosurgeon with the much-needed real-time feedback.