Current master thesis

M.Sc. candidate Alice Segato
Thesis Surgical Path Planning for Minimally Invasive Neurosurgery and Deep Brain Stimulation using DTI Tractography
Supervisors Elena De Momi, Alberto Favaro
Description The current trend in medical intervention favors a less invasive approach with a tendency to minimally-invasive, localized therapy. Common procedures employed in modern clinical practice involve percutaneous insertion of needles and catheters for biopsy and drug delivery and deep brain stimulation (DBS). A biomimetic flexible steerable probe is currently being developed at Imperial College London as one of the target of Eden2020 project which aim is the access of deep brain areas with minimum damage in order to accurately place minimally invasive instrumentation (catheters, electrodes for deep brain stimulation), to perform clinical analysis and diagnosis (biopsy, sampling), localized drug delivery and micro neurosurgery.

This master thesis is part of a PhD thesis whose aims is the development of a path planning algorithm for the pre-operative phase and defines surgical trajectories for brain tumor treatment in the context of Convention Enhanced Delivery.

In particular this thesis is focused on the use of Tractography to visually represent neural tracts using data collected by diffusion-weighted images in order to provide the planner with a certain pose in its terminal part and use the data of the tractography no longer as a target but as an obstacle for the path planning of deep brain stimulation having as target subthalamic nuclei.

M.Sc. candidate Federico Muretti, Federico Perrotta
Thesis Novel algorithms for steerable needle path planning with in vitro trajectory tracking feedback
Supervisors Elena De Momi, Alberto Favaro
Description Aim: To plan the optimal paths of an antitumoral-delivery steerable needle for advanced grade gliomas treatment

In the field of minimally invasive neurosurgery, various keyhole procedures (such as diagnostic biopsy, Deep Brain Stimulation, Stereoelectroencephalography and drug delivery) are usually performed by rigid linear tools, thus only straight-line trajectories are allowed. One of the efforts of the EU project EDEN2020 consists in planning curvilinear trajectories for a programmable bevel-tip needle. A recent, less invasive, approach has been introduced by using flexible steerable needles to overcome problems when the ideal straight path is not feasible or presents high risk level (blood vessels close). This task is achieved taking into account of mechanical constraints (radius of curvature) and of the overall safety (distance from the obstacles), crucial to avoid any damage to relevant anatomical regions.

The aim of this master thesis is twofold: at first, the work relies on the optimization of a path planning algorithm for the pre-operative (off-line) phase to obtain a set of 3D safety-corrected paths for a steerable controlled needle to be traveled [fig.1]. To this purpose, one part to be optimized is the interpolation of the curve which describes the catheter path to be planned. NURBS method was implemented to enhance the algorithm with the new property of 3D local support in the construction of the curves, which was a notable issue [fig. 2].

Secondly, a strictly related work focuses on the tracking of the needle during in vitro tests to evaluate its control system performances. The insertion of the steerable catheter is guided and controlled by a surgical robotic control system. The non-idealities in the kinematic model of the system actuators may cause significant errors in the case of a non-linear needle trajectory, which may bring to critical damages. It follows the importance of a test phase on jelly brains, joined with a tracking system based on cameras sending feedbacks to the control of the catheter, leading to an even improved path with minimized errors. The needle tracking which tests the control system capabilities will also provide a characterization of the needle itself.

M.Sc. candidate Beatrice van Amsterdam
Thesis Unsupervised surgical gesture segmentation
Supervisors Elena De Momi, Danail Stoyanov, Hirenkumar Nakawala
Collaboration Surgical Robot Vision research group, UCL
Description The adoption of robot-assisted minimally invasive surgery is generating datasets of kinematic and video recordings of surgical procedures. This data can facilitate robot learning from demonstrations, surgical training, assessment and automation.
Segmenting this demonstration data into meaningful sub-trajectories can benefit learning since individual segments are often less complex, have lower variance, and it is easier to remove outliers.
Different supervised methods have been developed either using segmented gestures or pre-characterized vocabulary of primitives. These methods rely on data that has been manually annotated by expert surgeons. However, providing such a faultless input is often unrealistic. Human annotations can be time-consuming and prone to error through missing segments or applying segmentation criteria inconsistently across a dataset. Hence, the need to develop unsupervised segmentation techniques, where the criteria is learned directly from data.
M.Sc. candidate Anna Mafrica
Thesis Design of an electromagnetic cortical stimulator for brain mapping during open skull neurosurgery
Supervisors Elena De MomiGiancarlo Ferrigno
Description Awake brain surgery is a neurosurgical technique mainly involved in brain tumors resection or in epilepsy surgery. In this procedure, the use of intraoperative brain mapping becomes essential firstly to reliably identify cortical areas and subcortical pathways involved in motor, sensory, language, and cognitive function in order to preserve the functional cortex and secondly to maximize the resection of the lesion. The actual golden standard to perform brain mapping is Direct Cortical Stimulation (DCS): a current is directly injected in the cortical tissue and the peripherical response is analysed in order to map the most important cortical areas near the lesion that has to be removed. The main drawbacks of DCS are firstly that the direct injection of currents in the brain cortex can induce seizures and secondly that this technique has a low resolution and very low penetration below the cortex surface.

The aim of this work is to propose an alternative tool for the intraoperative cortical mapping, allowing for a navigated, contactless stimulation of the cortex using coils inducing fast changing magnetic field on the brain tissues through a flexible magnetic circuit. The concept is to exploit an effect similar to Transcranial Magnetic Stimulation technique (already used to perform preoperative brain mapping) during the intraoperative phase and with a substantial higher resolution in space.

M.Sc. candidate Maria-Paola Forte 
Thesis Mathematical methods to recognize and analyse surgical tasks
Supervisors Elena De Momi, Nima Enayati
Description In recent years, the advancements in technology deeply influenced the surgical field leading to the development and wide adoption of Robotic Minimally Invasive Surgery (RMIS) systems, such as Intuitive Surgical’s daVinci. This technology has several advantages compared to the traditional surgery, among which greater precision, smaller incisions and shorter recovery time.

da Vinci’s ability to acquire the video from its stereo endoscope opens the opportunity of creating mathematical methods to recognize and analyse surgical tasks. These models can then be used to provide assistance to either trainees or expert surgeons. By now the robotic guidance is mainly rigidly-designed for specific tasks (especially by means of Virtual Reality). As a further step, the idea is to try to offer a context-aware assistance.

The aim of this project is to create a general platform able to provide the afore-mentioned guidance in real-time. Features from the acquired video are extracted with Computer Vision techniques and then Machine Learning is used to build the model of the under-analysis surgical task. Once the next phase of the task is predicted, the relative cognitive assistance will be provided.

M.Sc. candidate Laura Erica Pescatori
Thesis Action recognition during minimally invasive robot assisted surgery
Supervisors Elena De Momi, Hirenkumar Nakawala
Description Robot-assisted minimally invasive surgery has been developing and increasing over the last years. One of the main robots used with this aim is the da Vinci robot. It allows the surgeon to see inside the patient’s body thanks to a camera and to control the instruments inserted in the patient’s cavity through the use of a console. Many kind of surgery can be carried on with da Vinci and the one considered in this work is partial nephrectomy which consists of the removal of the tumor situated in the kidney.
The aim of this work is to use neural networks in order to allow the robot understand which action the surgeon has been carried on and, based on this, also what would be the following action. This network is created thanks to the use of annotated videos of such surgery procedures. Using images extracted from videos and the optical flow calculated between two subsequent images the network can be trained and thus will be able to classify the correct actions once a new video is fed to it.
M.Sc. candidate Andrea Mariani, Edoardo Pellegrini
Thesis Motor-Learning In Surgical Tele-Operation: Training Assessment And Improvement
Supervisors Elena De Momi, Peter Kazanzides (JHU), Nima Enayati
Description The main hypothesis of the work is that an adaptive training to surgical robotic procedures can measurably improve the outcomes of the skill acquisition. The training adaptation is analysed in terms of user’s performance-driven physical guidance application and exercise selection. All the experiments are performed in a virtual reality environment with the dVRK by Intuitive Surgical, which is the Research Kit for the da Vinci Surgical System, a minimally-invasive surgery robot.
More in detail, the project comprises two principal aims:
1) Design and evaluation of physical assistive methods in training
The first aim of the thesis is to investigate the effects of providing physical guidance in primary phases of tele-operated surgical training. This robot-assisted training guides the trainee through motion, force, torque, or vibration cues applied by the master device and is adapted to users’ progress. Such an assistive method is hypothesized to facilitate the initial stages of training by preventing cognitive overload. These methods are studied through multiple experiments using a tele-operation virtual reality setup with statistically significant populations of subjects. The results of the study should reveal potential benefits or harms of motion guidance in skills training in terms of both immediate improvements and long-term retention.
2) Design and evaluation of adaptive training programs
The second objective of the study is the evaluation of a method that aims at automatically scheduling the training on the base of an objective assessment of the trainee’s performances. More precisely, the underlying hypothesis turns into: can a “smart” training allow to obtain better final performances with respect to a self-managed training, where the word “smart” refers to the fact that the training schedule is arranged in an automatic and performance–based way? Moving from the identification of the basic and fundamental surgical skills towards the design of teleoperative tasks that could train these skills, the study focuses on the formulation of ML methods to assess the surgeon mastery of these abilities and the autonomous adaptation of the training program accordingly. Finally, the potentialities of this training approach are evaluated through the experimental comparison of statistically significant populations of subjects, one of which attending a ‘smart’ adaptive schedule and the other a canonical approach (i.e. employed by the commercial trainers) based on the user’s choices.
M.Sc. candidate  Francesco Grigoli
Thesis Segmentation of surgical gestures for tasks recognition and skills evaluation
Supervisors Nima Enayati, Elena De Momi, Aleks Attanasio
Description New technologies, as the robotic surgical systems, record motion and video data guaranteeing access on hidden features useful evaluate the surgeon’s skills. Data as the completion time, force/torque interaction and kinematic values can be directly used for this purpose.
Other approaches use the acquired features to segment surgical gestures and build models to classify surgical expertise using also the uncover concealed patterns. Considering the statistical distance from the user’s model from that of an expert a more complete measure of the user’s skills is reached.
To model surgical tasks, variations of the Hidden Markov models (HMMs), as in speech recognition, are widely used, however, due to the high complexity of the surgical procedures the definition of the observation model is challenging and the problem is still open. Using the kinematic labelled data from the JIGSAWS dataset, we are traying to overcome this issue paring a time series Neural Network (NN) algorithm, rude but able to catch the general aspects surgical gestures common to many tasks with a task specific HMM.
The HMM we are using for one task is composed by two levels. The deepest one is built considering an HMM for each gesture with multivariate Gaussian distributions as emitting states; these models are trained independently with the Baum-Welch algorithm, over data reduced with Linear Discriminant Analysis (LDA). The second level is an HMM that links all the models of the gestures considering task specific constrains. The final Algorithm will give a good classification of the gestures in a surgical procedure allowing us in building the user’s model to be compared with the expert one.
M.Sc. candidate Mohatashem Reyaz Makhdoomi
Thesis Adaptive Control of Teleoperated KUKA LWR 4+ in Minimally Invasive Surgery
Supervisors Elena De Momi
Description The well-known KUKA Light Weight Robots (LWR) have been popularly integrated into surgical systems for MIS with the advantage of achieving good surgical precision, enhanced dexterity and range of motion. These robots are light-weight, capable of high accuracy, have large DOFs and compliance. They can be manipulated manually and can follow defined tool-trajectories relative to anatomical features on the patient through different control modes. For the sake of achieving compliance during minimally invasive surgeries, cartesian impedance control is used to secure the accuracy of the surgical task. A secondary task, such as the guaranteeing a flexible workspace to the medical staff, by enabling a compliant swivel motion at the robot’s elbow can be obtained by controlling its null space. However, there are limitations in the mentioned strategies. Despite guaranteeing compliant behavior, Cartesian Impedance control does not guarantee the accuracy as high as demanded by a surgical intervention. During teleoperation, the surgical tool tracks the specified desired cartesian position albeit with some undesirable error. Since the uncertain human-robot interaction in the null space, the forces experienced by the robot manipulator are uncertain and time-varying, leading to the error in the surgical task. Impedance control parameters are not competent enough to work out such uncertainties.
In the field of control engineering, direct adaptive fuzzy controllers are known to perform well in presence of uncertainties and disturbances. In addition to it, supervisory control has been used to restrict the error within certain feasible limits. It is capable to achieve higher accuracy and quality in the MIS.
This work aims at addressing the above-mentioned limitations. It is divided into two objectives: Firstly, to develop an adaptive fuzzy controller to reduce the task error despite the uncertainties. Secondly, to develop a supervisory control to constrain the error within feasible limits. The work has been performed in three stages:
– Validating the proposed controllers in Simulink.
– Development and Implementation of controllers on simulations using V-rep and ROS.
– Implement of the proposed algorithm on the real robot
M.Sc. candidate Andrea Passoni
Thesis Human – Robot Interaction in Teleoperation: stiffness modulation during a reaching task
Supervisors Elena De Momi, Jacopo Buzzi
Description In tele-operated robot assisted surgery, surgeons use a robotic input device (master) to control the remote tele-manipulator (slave). With respect to traditional laparoscopy, tele-operated robotic surgery has brought several advantages, like improved ergonomics, motion scaling and the possibility of filtering the surgeon’s hand tremor, but remains a complex sensorimotor task also due to the multiple redundancies (kinematic and kinetic) that characterize the human motor control.

Up to now, the dynamic properties of the human arm have not been fully considered in the design and optimization of the master interfaces, considering both the human operator and the robot as passive elements. Through the estimation of user’s arm endpoint stiffness, that is the predominant component of the arm impedance, it would be possible to implement a master controller able to regulate the human – robot interaction.

The aims of this master thesis are, in particular:

  • to extract the arm endpoint stiffness while the user is performing a teleoperation task (reaching task) in order to better investigate the active arm dynamic properties control adopted by the user’s CNS. This is achieved by integrating user’s kinematic and kinetic data with a proper musculoskeletal model implemented in OpenSim;
  • to actively change the impedance characteristics (the viscosity, in particular) of the master controller to match the stiffness modulation strategies adopted by the human operator;
  • to test the impedance controller
M.Sc. candidate Anna Morelli
Thesis Intra-operative deformable registration for Augmented Reality in nephrectomy
Supervisors Elena De Momi, Sara Moccia
Description Renal cell cancer is a kind of kidney tumor that affects 14,068,000 patients all over the world. The treatment for this disease is nephrectomy, a surgical procedure in which the entire kidney, or only a part of it, is removed. Nowadays the procedure can be performed in minimally invasive surgery, allowing reduction of bleeding, of pain and of recovery time; the drawbacks of this technique is, for the surgeon, the impairment of haptic feedback for the discrimination of different structures. To overcome this limitation, Augmented Reality (AR) systems have been proposed. In an AR operating environment the pre-operative kidney model, usually extracted from CT or MRI, is superimposed onto the surgical field of view. Potential clinical advantages are offered in two stages: 1) in the identificiation of important structures, such as vessels, tumor, healthy tissues, in the initial phase; 2) during tumor resection, fixing negative surgical margins for the instruments.

Focusing on the first advantage, the superimposition during the first phase can be obtained starting from a landmark-based initial alignment of the model to the intra-operative scene. The intra-operative scene is recorded with a stereo-camera to obtain the intra-operative 3D reconstruction, i.e. the intra-operative point cloud. Then, with a deformable registration algorithm, it is possible to find the transformation to map the pre-operative model into the intra-operative point cloud.

The aims of this project are: i) implementing an algorithm for the initial registration based on manual identified landmarks; ii) implementing a deformable registration algorithm to robustly deal with the intra-operative deformation of the kidney.


M.Sc. Candidate Stefano De Nigris
Advisor Elena De Momi, Digna M González-Otero (EHU-UPV), Jesús Ruiz(EHU-UPV), Sofía Ruiz de Gauna(EHU-UPV)
Title Application of accelerometer-based chest compression feedback devices in novel scenarios
Collaborations Laboratorio GSC (Grupo Señales y Comunicaciones), Universidad del País Vasco (EHU-UPV), Bilbao, Spain
Description StefanoNigrisThesis

Sudden cardiac arrest is the largest cause of natural death worldwide. Early cardiopulmonary resuscitation (CPR) is key for patient survival. Resuscitation guidelines emphasize the importance of providing high quality chest compressions, that is, with a rate of 100 compressions per minute and a depth of 5cm in adults and one third of the diameter of the chest in children. However, meeting these requirements is difficult, even for well-trained rescuers.

The use of feedback systems can help rescuers increase chest compression quality. Most of these devices measure the acceleration of the chest during compressions, and guide the rescuer towards the target rate and depth. However, these systems present two main limitations. First, they are designed for adult patients, and thus present a target depth of 5cm, which is inadequate for children. Second, they could be inaccurate when used in moving vehicles, such as a train or a plane, as they would register the acceleration of the vehicle along with that of the patient’s chest.

The work tackles those two limitations. It is divided in two main objectives: First, to develop and test an algorithm to measure the diameter of the chest using an accelerometer. This algorithm could be used to adapt feedback systems to be used in pediatric patients. Second, to propose and test an additive model to evaluate the accuracy of accelerometer-based CPR feedback devices in moving vehicles, and to apply it to the case of a plane and train.

The results of this thesis could extend the scenarios of application of feedback systems, allowing their use in pediatric cardiac arrests or in public transportation means.


This site uses cookies to enhance your experience. By continuing to the site you accept their use. More info in our cookies policy.     ACCEPT