Data-Driven Intra-Operative Estimation of Anatomical Attachments for Autonomous Tissue Dissection

Abstract

The execution of surgical tasks by an Autonomous Robotic System (ARS) requires an up-to-date model of the current surgical environment, which has to be deduced from measurements collected during task execution. In this work, we propose to automate tissue dissection tasks by introducing a convolutional neural network, called BA-Net, to predict the location of attachment points between adjacent tissues. BA-Net identifies the attachment areas from a single partial view of the deformed surface, without any a-priori knowledge about their location. The proposed method guarantees a very fast prediction time, which makes it ideal for intra-operative applications. Experimental validation is carried out on both simulated and real world phantom data of soft tissue manipulation performed with the da Vinci Research Kit (dVRK). The obtained results demonstrate that BA-Net provides robust predictions at varying geometric configurations, material properties, distributions of attachment points and grasping point locations. The estimation of attachment points provided by BA-Net improves the simulation of the anatomical environment where the system is acting, leading to a median simulation error below 5mm in all the tested conditions. BA-Net can thus further support an ARS by providing a more robust test bench for the robotic actions intra-operatively, in particular when replanning is needed. The method and collected dataset are available at https://gitlab.com/altairLab/banet.

Autonomous tissue retraction with a biomechanically informed logic based framework

“…2). The tissue is modelled as a St Venant-Kirchhoff material with Young's modulus 3 kPa and Poisson ratio 0.45, a common modelling choice for adipose tissues [32], [33], which are often responsible of hiding structures of interest during real surgery [28]. APs are defined as random tissue patches, following the strategy proposed in [33].…”

Accelerating Surgical Robotics Research: A Review of 10 Years With the da Vinci Research Kit

“…A lot of the remaining works focused on tissue interaction. This application category includes papers working on cutting and debridement [61], [87], [90], [92], [95], [97] as well as the retraction and dissection of tissues [101], [131], [132], [155], [257], [265] or blood suction [226]. Also included is tissue palpation for locating tumors or vessels and more general tissue manipulation, as in [13]- [15], [17], [82], [83], [86], [98], [99], [154], [164], [225], and [241], with experiments sometimes using just common fabric as a phantom for tissue [94], [100].…”

Intra-operative Update of Boundary Conditions for Patient-Specific Surgical Simulation

“…BANet The boundary condition update is performed by BANet, a DNN estimating at which points a given deformable tissue is attached to the surrounding environment [22]. BANet has been validated on phantom data with simple geometry, but has never been applied to a real PBM and within a realistic clinical situation.…”

“…The training dataset is composed of synthetic samples representing adipose tissues PBMs (with different random geometries and mechanical parameters) and annotated with randomly extracted BCs. In this work, we use the publicly available implementation of BANet with the provided pre-trained weights [22].…”

“…A synthetic dataset of 600 samples of adipose tissue manipulation is generated following the same training data generation strategy used in [22], but keeping the considered PBM fixed (Fig. 2b).…”