Sujets de stages

Aller à : navigation, rechercher


Sujets en Vision par Ordinateur / Deep Learning pour le Médical (Projet CAMMA: Computational Analysis and Modeling of Medical Activities)

We are looking for motivated and talented students with knowledge in computer vision, machine learning and/or augmented reality who can contribute to the development of our computer vision system for the operating room.

Please feel free to contact Nicolas Padoy if you are interested to do your master's thesis or an internship with us (funding of ~500Euros/month will be provided during 3 to 6 months). The successful candidates will be part of a dynamic research group hosted within the IRCAD institute at the University Hospital of Strasbourg. They will thereby have direct contact with clinicians, industrial partners and also have access to an exceptional research environment. The CAMMA project is supported by the laboratory of excellence CAMI, the IdEx Unistra and the MixSurg Institute.

Topics:

  • Deep Learning for the Analysis of Large Surgical Video Databases
  • Multi-view Human Body Tracking for the Operating Room
  • 3D Simulation and Visualization of X-ray Radiations for Radiation Safety Analysis

More information about CAMMA

Links:


Computer vision and medical images processing

  • Title: Robust tracking of moving tissues in colonoscopy videos
  • Training period: 5-6 monthes, between January and August 2018
  • Supervisor: F. Nageotte (nageotte_at_unistra.fr)
  • Keywords: robust tracking, features detection, surgical endoscopy
  • Context:

This work takes place in the development of tools for assisting gastroenterologists and surgeons in manipulating flexible endoscopes for medical tasks in the digestive tract. One of the goals is to automatically and actively track areas of tissues with robotic endoscopes during physiological motion. For this purpose, areas of tissues have to be robustly tracked in the endoscopic images.

Peristalsis.jpg

Many solutions have been proposed in the past few years to track tissues in endoscopic images [M. Yip 2012, V.Penza 2017]. However, most of these techniques have been tailored for laparoscopic surgery. Images provided by colonoscopes (see figure) differ notably from laparoscopic images for the following reasons:

    • The environment is uniform, with only the mucosa of the digestive tract visible,
    • The orientation of tissues with respect to the camera is more varied, with tissues on the side (like in a pipe),
    • Cameras embedded in conventional flexible endoscopes provide images with lower resolution and with lower quality than in laparoscopes,
    • Physiological motion such as peristaltic waves create very strong tissue deformations.
  • Work of the intern:

The objective of the internship will be to identify the tracking techniques which are adapted to these particular conditions. Real images acquired during surgery in the colon of pigs will be used as support. The work will therefore consist in: - Implementing and testing existing algorithms on colonoscopy sequences - Identifying limitations of these algorithms - Adapting and improving them to fill the requirements of robust tracking. The implementation will be made in Matlab and / or C++/ open CV depending on the availability of existing codes. The work will be carried out on the robotic platform of the AVR team of the ICube laboratory, located at the hospital in the center of Strasbourg.

  • Profile of candidates : Master student (or student from engineering school) with major in computer vision or real-time image processing. Interest for / knowledge in medical robotics application can be a plus but is not a requirement. Proficiency in Matlab and C/C++ are required. Knowledge of OpenCV will be appreciated.
  • To apply: send a CV, cover letter, available grades at master level and programs of master courses to Florent Nageotte : Nageotte@unistra.fr


Image-based tracking of continuum robots

  • Timeframe: 5-6 months, between January and August 2018
  • Supervisors: B. Rosa and F. Nageotte
  • Keywords: medical robotics, continuum robots, endoscopy, image-based tracking
  • Context:

Continuum robots are increasingly used in minimally invasive surgery. The mechanical models alone, however, do not allow to reliably estimate the shape of the robot (due to friction and other nonlinearities). This work focuses on image-based tracking of a continuum robotic arm in the context of surgical endoscopy.

Existing solutions use either markers are a substantial set of pre-labeled images for supervised machine learning. We have developed an alternative method using the robot mechanical model to train a classifier to recognize the instrument, without needing to pre-label any image before the surgery. THe method is thus an unsupervised, online machine learning method to track a continuum instrument on endoscopic images.

  • Work of the intern:

The objective of this internship is to incorporate finer mechanical information into the algorithm to make it more robust. In particular, the algorithm now uses only static information (i.e. inferred robot pose from mechanics), and we are interested in incorporating differential kinematics (i.e. robot jacobian). The work will therefore consist in :

  • Analyzing the current algorithm to understand its main features and shortcomings
  • Developing a theorretical framework to link the differential kinematics of the robot to image features
  • Implementing the algorithm, and benchmarking it on benchtop / in vivo data from our STRAS platform

The work will be carried out on the robotic platform of the AVR team of the ICube laboratory, located at the hospital in the center of Strasbourg.

More information about the subject is available here : File:Master_Internship_Subject_2018_Tracking.pdf

  • Profile of candidates : Master student (or student from engineering school) with major in robotics and/or computer vision, with strong analytical skills. Proficiency in python and/or C++ is required. Knowledge of computer vision libraries such as OpenCV, VTK, and an experience with machine learning algorithms, would be appreciated, but are not a requirement.
  • To apply: send a CV, cover letter, available grades at master level and programs of master courses to Benoit Rosa : b.rosa@unistra.fr


Unsupervised machine learning for continuum robot visual tracking using deep learning

  • Timeframe: 5-6 months, between January and August 2018
  • Supervisors: B. Rosa and N. Padoy
  • Keywords: medical robotics, continuum robots, instrument tracking, deep learning
  • Context:

Continuum robots are increasingly used in minimally invasive surgery. The mechanical models alone, however, do not allow to reliably estimate the shape of the robot (due to friction and other nonlinearities). This work focuses on image-based tracking of a continuum robotic arm in the context of surgical endoscopy.

Existing solutions use either markers are a substantial set of pre-labeled images for supervised machine learning. We have developed an alternative method using the robot mechanical model to train a classifier to recognize the instrument, without needing to pre-label any image before the surgery. The method is thus an unsupervised, online machine learning method to track a continuum instrument on endoscopic images.

  • Work of the intern:

THe developed method uses handcrafted features to track the instrument. The objective of this work is to investigate the use of deep learning approaches to enhance the unsupervised learning algorithm. Depending on the progress, the following directions will be explored to improve the current algorithm :

  • Use better image features learned by a deep neural network as input for the current method based on random forests
  • Replace the hybrid approach relying on deep features and random forests by a method based on a single deep network, considering that this network needs to be optimized online when new images become available
  • Jointly optimize both the pose and the tool detection model as opposed to using a two steps optimization strategy
  • Improve the overall method to make it applicable in real-time

The work will be carried out on the robotic platform of the AVR team of the ICube laboratory, located at the hospital in the center of Strasbourg.

More information about the subject is available here : File:Master_Internship_Subject_2018_DeepLearning.pdf

  • Profile of candidates : Master student (or student from engineering school) with major in computer vision and/or robotics, with a taste for applied mathematics and strong analytical skills. Proficiency in python and/or C++ is required, as well as knowledge in computer vision and machine learning. Knowledge in deep learning and/or computer vision libraries such as OpenCV, is a plus.
  • To apply: send a CV, cover letter, available grades at master level and programs of master courses to Benoit Rosa : b.rosa@unistra.fr



Medical robotics and electromagnetic tracking

  • Title: Electromagnetic tracking of continuum robots
  • Training period: 5-6 monthes, between January and August 2018
  • Supervisor: F. Nageotte (nageotte_at_unistra.fr) and P. Cantillon-Murphy
  • Keywords: Continuum robots, endoscopy, EM tracking

The complete proposal in pdf is available here.

  • Context:

In this work, one aims at investigating the use of electromagnetic (EM) sensing for measuring and controlling flexible endoscopic systems. EM tracking is an attractive technique for providing external sensing on continuum systems because no line of sight is needed and sensors have very small sizes. For this purpose, the ICube laboratory is collaborating with the team of Padraig Cantillon-Murphy at University College Cork, Ireland. This team has developed an open EM tracking system called Anser [2].

The objective of the work will consist in developing systems, techniques and software for an efficient tracking of endoscopes and endoscopic instruments.

STRAS.jpg

  • Work of the intern:

The work will consist in three main stages:

    • First the working of the ANSER EM will have to be understood and the tracking of flexible instruments will be analyzed to serve as a baseline for subsequent developments. This will be done in static as well as dynamic conditions (movements).
    • The origin of accuracy limitations, for instance the presence of metallic parts and motorization, or delays in acquisition will be identified
    • Improvement on the setup, on different aspects: mechatronics (attachment of sensors), electronics (adaptation of EM frequencies), software (filtering, etc.).

Testing will be performed with the ANSER EM tracking system both on manual flexible instruments and on the STRAS robotic system of the ICube laboratory [3] (see figure).

  • Profile of candidates : Student at the master level with major in electrical engineering or mechatronics with strong experimental skills (electronics, mechatronics, software development).

Experience in electromagnetism would be a plus.

  • To apply: send a CV, cover letter, available grades at master level and programs of master courses to Florent Nageotte : Nageotte@unistra.fr


Mechatronic design of a medical robot

Position is now closed.

  • Title: Mechatronic Design of a Robotic Device for Olfactory Cells Inspection
  • Training period: 5-6 months, starting February / March 2018.
  • Supervisor: P. Renaud
  • Keywords: Robot design, CAD design, mechatronics, integration, concentric tube robots

The complete proposal in pdf is available in french here: Proposal

  • Context:

In this work, one aims at building a prototype of the robotic system developed in the NEMRO project, a national project aiming at providing a new approach for olfactory cell inspection.

The objective of the work is to build a functional robotic device based on a concentric tube robot in order to show the feasibility of the robotised approach as developed during the first 3 years of the NEMRO project.

  • Work of the intern:

The work will consist in three main stages: i) to understand concentric tube robot principle and existing results issues from the work of C. Girerd, ii) CAD design of the robotic system, iii) integration to build the experimental platform.

  • Profile of candidates : Student at the master level with major in mechanical engineering / mechatronics with strong skills in CAD design, integration and experimental evaluation.
  • To apply: send a CV, cover letter, available grades at master level and programs of master courses to Pierre Renaud : pierre.renaud@insa-strasbourg.fr