A neural network approach for flexible needle tracking in ultrasound images using Kalman filter

Abstract

Percutaneous surgical procedures depend on the precise positioning of the needle tip for effectiveness. Although, robustly locating the tip of the needle still represents a challenge, specially when flexible needles are used. An imaging technique widely used for this task is 2D ultrasound, however the low signal/noise ratio makes it difficult to apply conventional image processing techniques. In this work, we propose an alternative method for detecting the needle in an ultrasound image and tracking it during a complete insertion. The proposed method combines a Multilayer Perceptron network with a Kalman filter estimator for improving robustness. In preliminary experimental characterization, we acquired ultrasound images for creating the data set and evaluated the performance of the tracker with a complete insertion video. However the tracking performance is still far from optimal, the obtained results suggests the neural network approach to be a feasible alternative to this problem.

Alignment Motion Control for an Automated Human Ear Surgery via Vision‐Servoing

“…Meanwhile, the actuator velocity is obtained with the backward approximation (v x , v y ). The observable measurement equation is, thus, given as Z(k) = HX (k) + n(k) (12) where H = I ∈ ℜ 4×4 and n(k) ∼ N(0, R) is the white Gaussian measurement noise with zero mean and covariance R. It is noted that the optimality of the Kalman filter depends on the quality of prior assumptions about the processing noise covariance Q and the measurement noise matrix R. Inadequate prior assumptions to represent the real noise level could lead to unreliable results and sometimes to filter divergence [25].…”

“…To estimate the variable matrix regarding to the system of two measurement inputs, we apply the adaptive Kalman filtering which makes the elements of the residual-based covariance matrix consistent with their theoretical values proposed by Wang [26]. According to (12), the residual v(k) which is the difference between the real observations and its predicted value can be computed as follows:…”

“…After the determination of the noise covariance Q * andR(k), according to the kinematic model in (8) and the observer equation in (12), the discrete Kalman filtering algorithm is illustrated in Fig. 8.…”

Dispositivo robótico para guiagem de agulhas flexíveis em procedimentos minimamente invasivos

“…Em alguns trabalhos, utilizam-se imagens de ultrassom 2D alinhadas com o plano de inserção da agulha [49]. Esta estratégia tem a vantagem de permitir visualizar uma área maior da agulha, no entanto apresenta problemas caso a ponta da agulha saia do plano da imagem.…”

“…Título: A neural network approach for flexible needle tracking in ultrasound images using Kalman filter [49] Autores: André A. Geraldes, Thiago S. Rocha…”