The Stentrode: A Minimally Invasive BCI - Promises, Signals, and Puzzles

You know those brain-computer interfaces (BCIs) you read about? The kind that let people with severe paralysis control computers or robotic arms just by thinking? Most of the high-performance ones involve opening the skull and putting electrodes directly on or in the brain. It works, but brain surgery isn't exactly a walk in the park.

Enter the Stentrode. This device, developed by Synchron Inc., aims to achieve similar results but with a minimally invasive approach. Instead of cutting into the brain, it's implanted into a large vein called the superior sagittal sinus (SSS), which runs along the midline inside your skull, right next to the motor cortex. Think of it like a tiny, flexible stent with electrodes on it. Once inside the SSS, it expands to press the electrodes against the vessel wall, allowing them to pick up electrical signals from the nearby motor cortex.

The paper we've been dissecting: Motor activity in gamma and high gamma bands recorded with a Stentrode from the human motor cortex in two people with ALS by Kacker et al. (2025), published in the Journal of Neural Engineering, presents early findings from the COMMAND trial using this endovascular BCI. They tested the Stentrode in two men with severe paralysis due to ALS, focusing on whether the device could reliably detect motor-related brain activity, specifically in the low gamma (30–70 Hz) and high gamma (70–200 Hz) frequency bands. These higher frequencies are generally considered more localized and action-specific than the alpha and beta waves picked up by less invasive methods like EEG. You can read the full paper here.

What Did They Find? (The Good Stuff)

• Signal Detection: The Stentrode could capture brain activity related to attempted movements. They measured this using metrics like Depth of Modulation (DoM) (how much the signal amplitude changes between rest and attempted movement) and Signal-to-Noise Ratio (SNR) (how strong the signal is compared to background noise). Participant 1 showed much higher DoM and SNR values than Participant 2.
• Signal Stability: For both participants, the channel-averaged SNR remained stable or even increased over the three-month testing period reported. This suggests the signals don't immediately degrade, supporting the idea of long-term use.
• Somatotopy: The signals showed some spatial specificity. For Participant 1, attempted ankle movements resulted in the strongest signals (highest DoM), particularly on one channel (Channel 8). This makes sense because the motor area for the ankles is located right near the brain's midline, exactly where the Stentrode is placed in the SSS. For Participant 2, right-hand movements produced the strongest responses. Figure 6, particularly the polar plot showing DoM per channel, visually demonstrates this channel variability.
• Classification Potential: The system could distinguish between rest and attempted movement with high accuracy (over 90%). Using Linear Discriminant Analysis (LDA), they also showed that signals from different limb movements (e.g., left ankle, right ankle, both ankles for P1) were separable. Figure 7 illustrates this, showing clear clusters for different movements in a transformed signal space for Participant 1.

But Let's Get Critical (The Puzzles)

• Only Two Participants: As the authors acknowledge, the findings are based on a very small sample size (just two people). It's hard to generalize these results to the broader ALS population.
• Offline Analysis: The study only shows that motor intent can be classified from the recorded data after the fact. It doesn't demonstrate real-time control of a BCI system, which is the ultimate goal.
• Data Not Available: A major hurdle is that the authors state the data "cannot be made publicly available because it contains commercially sensitive information". This is a common phrase but makes it essentially impossible for outside researchers to verify the claims, re-analyze the data, or build upon the work directly.
• Conflicts of Interest: Several authors are affiliated with or hold stock/patents in Synchron Inc., the company that makes the Stentrode. While standard practice to disclose this, it provides important context: this is research heavily tied to a commercial product.
• Opacity in Methods & Signal Quality:
  • Channel Variability: We know the Stentrode has 16 electrodes, but only 12 for P1 and 8 for P2 were used in the analysis. What happened to the others? The paper doesn't say.
  • Impedance & SNR Over Time (Per Channel): They measured electrode impedance during each session and reported average impedance per channel (Tables 1 & 2, not shown here). However, crucially, they don't show how impedance or SNR changed over time for each individual channel. This makes it difficult to assess the true long-term stability of individual electrodes, which is critical for reliability.
  • Error Bars in Figure 4: Figure 4 shows channel-averaged SNR stability over time. The error bars represent the standard deviation across daily sessions, not across channels. While the small error bars for P1 in low gamma might suggest consistency, they could also hide underlying variability between channels or be influenced by selective channel inclusion. The larger error bars for P2 hint at more session-to-session fluctuation.
• The Mystery of PC1: In Figure 5, they use the first principal component (PC1) from PCA to show signal discriminability (d-prime) over time. PC1 represents the direction in the data with the most variance. While Figure 5(a) shows PC1 does track the rest/go cues, leading to incredibly high d-prime values (sometimes near 10), the paper doesn't explain what this PC1 physically represents or which electrodes contribute most to it. This raises a concern: could PC1 be dominated by activity from just a few channels, or potentially even subtle noise sources (like changes in blood flow) time-locked to the task, rather than a broad, reliable neural signal across the array? The fact that this PC1 analysis is separate from the LDA analysis in Figures 6 and 7 (which uses multichannel data directly) makes it even more confusing to interpret the high d-prime results in Figure 5 in the context of the overall system performance.

In Conclusion...

The Kacker et al. (2025) paper provides intriguing preliminary evidence that the Stentrode BCI can record relevant gamma-band signals from the SSS in people with ALS and that these signals are stable over a few months and show potential for decoding motor intent. The minimally invasive approach is a significant advantage.

However, the small participant number, lack of real-time testing, opacity regarding data access and author conflicts, and critical missing details in the methods regarding channel selection, per-channel signal quality over time (impedance/SNR), and the interpretability of the PC1 analysis make it challenging to fully evaluate the robustness and generalizability of the claims.

It's a strong start, but as with much industry-backed research, the lack of open data and full methodological transparency leaves researchers on the outside asking questions. It feels a bit like seeing a demo reel rather than the full documentary.

For more details, you can read the full paper: Kacker, Kriti, et al. "Motor activity in gamma and high gamma bands recorded with a Stentrode from the human motor cortex in two people with ALS." Journal of Neural Engineering 22.2 (2025): 026036. Access it here.

We're actually hoping to explore ways to make BCI research more transparent and interactive, potentially turning these complex papers into accessible online resources with interactive figures and explanations, like the ones we experimented with. You can check out our progress at bionichaos.com!