Dynamic order selection analysis in adaptive polynomial Kalman filtering: implementation and integration of sensor data and hybrid image processing for bio-inspired needle systems

Abstract

This study investigates dynamic-order selection in Adaptive Polynomial Kalman Filtering (APKF) for tracking the bioinspired dual-sheath needle systems used in biopsy procedures. Emphasizing integration of sensor data and hybrid image processing, the goal is to achieve precise motion estimation, which is critical to medical robotics. A hybrid image tracking system combined with APKF was implemented for real-time needle tip tracking and validated using a linear rail setup. Initial simulations showed that the standard APKF significantly outperformed traditional Kalman Filtering (KF), achieving an average reduction of 46.9% in Root Mean Square Error (RMSE), 57.8% in Mean Absolute Error (MAE), and 64.5% in Median Absolute Deviation (MAD). To further improve the performance, model-order selection criteria–Mean Squared Error (MSE), Akaike Information Criterion (AIC), Corrected AIC (AICc), and Bayesian Information Criterion (BIC)–were applied within the APKF framework. This led to even greater reductions in RMSE (55.4%), MAE (61.2%), and MAD (65.9%) compared with KF. The results highlight the effectiveness of combining model-order selection with adaptive filtering to enhance real-time estimation. The proposed tracking system demonstrates improved accuracy and control, reinforcing the potential of bioinspired needle systems in robot-assisted biopsy procedures.