TL;DR
Researchers have developed the first open-source, spiking neural network model of the spinal pudendo-vesical reflex that accounts for both stimulation frequency and recruitment amplitude. By simulating how different electrical intensities "recruit" varying numbers of nerve fibers, the model provides a vital tool for optimizing spinal cord stimulation (SCS) to restore bladder control after spinal cord injury.
What Problem Does This Paper Address?
For individuals with spinal cord injury (SCI), losing control over bladder function (neurogenic bladder) is a major cause of medical complications and reduced quality of life. While electrical stimulation—either through peripheral nerve implants or epidural spinal cord stimulation (SCS)—shows promise in restoring function, finding the "sweet spot" for stimulation parameters is incredibly difficult.
The challenge lies in the complexity of the pudendo-vesical reflex, which governs whether the bladder stores urine (continence) or voids it (micturition). Previous computational models treated neural populations as abstract "single nodes." While useful for high-level theory, these models couldn't simulate recruitment: the reality that increasing stimulation amplitude (current) activates more axons. Without accounting for recruitment, researchers couldn't accurately predict how changing the "strength" of a stimulator would actually affect bladder pressure.
Methods
The authors built a realistic spiking neural network (SNN) using the NEURON simulation environment and integrated it into the $o^2S^2PARC$ open-source platform.
Key technical features include:
- Circuit Architecture: The model includes populations of pudendal and pelvic afferents, dorsal interneurons (IND), excitatory/inhibitory interneurons ($IN_{M+}$, $IN_{M-}$), sacral parasympathetic nucleus (SPN) neurons, and feedback loops.
- Population Modeling: Instead of single nodes, they included 30 cells for each neuron group, modeled as
IntFire4classes to simulate post-synaptic potentials. - Recruitment Simulation: They implemented "virtual neurons" where the firing rates are a product of sensory feedback and the defined stimulation frequency. Crucially, they added a parameter to adjust the recruitment ratio from 0% to 100% of the available axons.
- Parallel Implementation: The model was designed to run in parallel, allowing for more complex simulations of neural circuits.
Datasets
This paper does not use a traditional machine learning dataset. Instead, it relies on:
- Physiological Parameters: Synaptic weights, time constants, and circuit connectivity were derived from established literature, primarily the work of McGee and Grill (2016).
- Validation Data: The model's outputs were compared against existing experimental data from isovolumetric bladder studies in cats to ensure the simulated pressure changes matched biological reality.
Key Insights and Ideas
- The Bi-Modal Switch: The model successfully replicated a known biological phenomenon: low-frequency stimulation (10 Hz) inhibits the bladder (promoting storage), while high-frequency stimulation (≥33 Hz) excites it (promoting voiding).
- Recruitment is the Key: The most significant "aha" moment was finding that high pudendal nerve recruitment (>80%) is necessary to maintain this bi-modal behavior. If the stimulation amplitude is too low (recruiting fewer axons), the reflex doesn't reliably switch between storage and voiding modes.
- Robustness to "Crosstalk": In clinical SCS, it is often hard to target just one nerve. The model showed that even if pelvic nerves are accidentally recruited (crosstalk), the system remains functional as long as the pudendal recruitment remains high.
- Circuit Resilience: The reflex circuit proved remarkably robust; the bi-modal behavior persisted even when synaptic connection weights were varied by up to 40%.
Key Results
- Frequency Response: At 10 Hz, the model produced a clear decrease in bladder pressure ratio (continence). At 33 Hz, 50 Hz, and 100 Hz, it produced a significant increase (micturition).
- Thresholds for Success: The researchers quantified that a recruitment rate of >60% is needed for high-frequency stimulation (>50 Hz) to trigger excitation, but >80% is required to sustain the effect at 33 Hz.
- Open Science: By hosting the model on $o^2S^2PARC$, the authors have provided a "plug-and-play" framework for other researchers to test different electrode geometries and stimulation waveforms in silico before moving to animal or human trials.
Limitations and Future Work
The authors identify a few key areas for improvement:
- Intrinsic Firing Limits: Currently, the SPN neurons in the model don't have an "upper speed limit" for firing. This can lead to unrealistically high bladder pressures at very high stimulation frequencies.
- Neuron Realism: While
IntFire4neurons are computationally efficient, they are still simplified. Future versions will incorporate more biophysically accurate neuron models to better simulate high-frequency effects. - Experimental Validation: While the model matches historical data, the authors emphasize the need for new experimental data specifically exploring the intersection of amplitude and frequency to further refine the model's predictive power.
Summary updated