Stochastic event detection in needle-tissue interaction

Over the last decade, many dynamic models that express needle-force relationships under tissues of varying mechanical properties have been developed. While great progress has been made in the development of these high-fidelity models, they are only valid within certain boundary conditions limiting their match with reality. This gap in realism is aggravated by variability in human tissues, needles, and the modes of interaction with the tissue. In an effort to develop more realistic models, the current paper was developed to create and test an event (i.e. changes of variability) detection method based on the probability distribution of residues-difference between force models and measurements. To obtain force measurements, we repeated robotic-driven needle insertion into a simulated mannequin. Needle types and tissue thickness were varied in the measurements in order to add realistic variability. To obtain the force model, the measurement data was used as an input to a Grey-Box model. From the measurements and models, we estimated the probability distribution of residues. For validation, a Gaussian-Mixture Model (GMM) was used to confirm the stochastic model successfully distinguishes the residual distributions under different variability. We found that by examining the residual distributions it is possible to detect unexpected variability in needle-tissue interactions. The findings from this paper have implications for developing real-time event detection methods and simulating patient-variability in haptic applications.