Introduction
Cardiovascular diseases remain a leading cause of global mortality, frequently necessitating minimally invasive interventions. In these procedures, surgeons navigate catheters through intricate vascular networks. While robotic catheter systems (RCS) have been developed to mitigate radiation exposure and physical strain for medical professionals, they often lack the intuitive "natural feel" of manual manipulation. Current master-slave setups often fail to replicate the nuanced hand motions perfected by surgeons over years of practice. This research explores bridging the gap between manual finesse and robotic precision by decoding surface electromyography (sEMG) signals from the forearm to translate natural surgical skills into digital robotic commands.
Methodology
The study focuses on capturing the electrical activity of skeletal muscles during fundamental catheterization maneuvers. The researchers utilized a custom dataset generated from a right-handed subject with two years of experience in vascular catheterization.
Data Acquisition and Signal Processing
Six sEMG electrodes were placed on specific forearm muscles (including the ETM, IM2, TAM, ELM, IM1, and PTM) to record activity during four core movements: move-in, move-out, turn-in, and turn-out. Each movement was repeated 100 times across two sequences to ensure a robust data pool. The raw signals, sampled at 1000 Hz, underwent filtering to remove power-line interference and were segmented into 100ms windows for analysis.
Feature Extraction and Modeling
Two primary features were extracted from the processed data:
- Average EMG (AEMG): Indicates muscle activity levels over specific data points.
- Root Mean Square (RMS): Measures signal power, reflecting the amplitude of muscle contractions.
The team employed k-means clustering (unsupervised) to evaluate the natural grouping of movements in the feature space and k-nearest neighbor (k-NN) classification (supervised) to determine the accuracy of movement identification based on muscle signatures.
Results and Discussion
The experimental results quantitatively demonstrate that sEMG signals are a viable input source for controlling a 2-Degree-of-Freedom (2-DoF) robotic catheter system.
Clustering and Classification Performance
Using k-means clustering (k=4), the four maneuvers formed distinct, identifiable clusters in the feature space. While both AEMG and RMS features performed well, RMS provided superior "within-cluster distance," meaning data points for specific movements were more tightly grouped, enhancing reliability.
As shown in the classification results, the k-NN algorithm achieved high accuracy across various configurations. The optimal performance was observed at k=2, where the model most effectively distinguished between the different surgical maneuvers. This high degree of accuracy suggests that muscle signatures are unique enough to allow a robot to act as a seamless extension of a surgeon’s muscular intent.
Conclusion and Future Work
This pilot study confirms that the electrical signatures of basic catheterization maneuvers can be identified with high reliability. However, several steps remain before clinical implementation. Future research will need to expand the sample size to account for physiological differences across a broader range of surgeons and address the complexity of overlapping motions found in real-world surgery. The ultimate goal is to transition from offline identification to real-time, active control of robotic catheter systems, moving toward a more intuitive, "signal-based" approach to robotic surgery.
Summary updated