Percutaneous procedures based on needle insertion are common medical interventions that require high targeting accuracy. This work introduces a method to automated robotic needle steering within deformable tissues, guaranteeing both high targeting accuracy and low computation times. Automated needle steering along a pre-defined trajectory with high targeting accuracy requires control strategies that take into account the complex interaction between the needle and the tissue, which includes needle shaft bending phenomenon and tissue deformations arising from the insertion. In order to control the needle movements inside the tissue, it is necessary to solve the inverse problem that provides the next robot end-effector position that allows to place the needle tip in the desired position. Even if the inverse problem for motion control is a widely used approach in the robotic field, its resolution is complex and computationally demanding. We propose to solve the inverse problem through the use of an Extreme Learning Machine (ELM) artificial neural network. A realistic Finite Element simulation of robotic needle steering into deformable tissue is used to generate the training database for the network. Ground truth control inputs for the robot end-effector are computed based on the inverse simulation approach leveraging on a needle-tissue biomechanical model. The proposed method is tested in a simulation environment where the ELM predictions are used to control the end-effector movements. The main contribution of this work is that our method reaches a sub-millimetre targeting accuracy while reducing the computational time by two-thirds with respect to the state of the art. The method has proved able to generalize to trajectories with new morphologies and tissue with mechanical properties never seen at training time, obtaining a targeting accuracy compatible with medical percutaneous procedures in all the considered scenarios.
Le procedure percutanee basate sull'inserimento dell'ago sono interventi medici comuni che richiedeno un'elevata precisione di targeting. Questo lavoro introduce un metodo per il controllo automatizzato dei movimenti degli aghi all'interno di tessuti deformabili, garantendo sia un'elevata precisione di targeting che tempi di calcolo ridotti. Il controllo dei movimenti dell'ago lungo una traiettoria predefinita con un'elevata precisione di targeting richiede strategie di controllo che tengano conto della complessa interazione tra l'ago e il tessuto, che include il fenomeno della curvatura dell'ago e le deformazioni dei tessuti derivanti dall'inserimento. Per controllare i movimenti dell'ago all'interno del tessuto, è necessario risolvere il problema inverso. Esso calcola la successiva posizione che l’end-effector del robot occuperà perché la punta dell'ago sia nella posizione desiderata. Anche se il problema inverso per il controllo del movimento è un approccio ampiamente utilizzato in campo robotico, la sua risoluzione è complessa ed impegnativa dal punto di vista computazionale. Ci proponiamo di risolvere il problema inverso attraverso l'uso di una rete neurale artificiale Extreme Learning Machine (ELM). Una simulazione realistica agli elementi finiti dell’inserzione dell'ago in un tessuto deformabile viene utilizzata per generare il database di addestramento per la rete. Gli input di controllo Ground truth per l'end-effector del robot vengono calcolati in base all'approccio della simulazione inversa sfruttando un modello biomeccanico ago-tessuto. Il metodo proposto viene testato in un ambiente di simulazione in cui le previsioni ELM vengono usate per controllare i movimenti dell'end-effector. Il contributo principale di questo lavoro è che il nostro metodo raggiunge una precisione di targeting inferiore al millimetro riducendo il tempo di calcolo di due terzi rispetto allo stato dell'arte. Il metodo si è dimostrato in grado di generalizzare a traiettorie con nuove morfologie e tessuti con proprietà meccaniche mai viste al momento dell'allenamento, ottenendo in tutti gli scenari considerati una precisione di targeting compatibile con le procedure mediche percutanee.
Robotic needle steering into deformable tissues with extreme learning machines
CAZZANIGA, ANNA
2020/2021
Abstract
Percutaneous procedures based on needle insertion are common medical interventions that require high targeting accuracy. This work introduces a method to automated robotic needle steering within deformable tissues, guaranteeing both high targeting accuracy and low computation times. Automated needle steering along a pre-defined trajectory with high targeting accuracy requires control strategies that take into account the complex interaction between the needle and the tissue, which includes needle shaft bending phenomenon and tissue deformations arising from the insertion. In order to control the needle movements inside the tissue, it is necessary to solve the inverse problem that provides the next robot end-effector position that allows to place the needle tip in the desired position. Even if the inverse problem for motion control is a widely used approach in the robotic field, its resolution is complex and computationally demanding. We propose to solve the inverse problem through the use of an Extreme Learning Machine (ELM) artificial neural network. A realistic Finite Element simulation of robotic needle steering into deformable tissue is used to generate the training database for the network. Ground truth control inputs for the robot end-effector are computed based on the inverse simulation approach leveraging on a needle-tissue biomechanical model. The proposed method is tested in a simulation environment where the ELM predictions are used to control the end-effector movements. The main contribution of this work is that our method reaches a sub-millimetre targeting accuracy while reducing the computational time by two-thirds with respect to the state of the art. The method has proved able to generalize to trajectories with new morphologies and tissue with mechanical properties never seen at training time, obtaining a targeting accuracy compatible with medical percutaneous procedures in all the considered scenarios.File | Dimensione | Formato | |
---|---|---|---|
2021_04_Cazzaniga.pdf
Open Access dal 01/04/2022
Dimensione
3.1 MB
Formato
Adobe PDF
|
3.1 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/175115