WP6 Machine Learning
The main objective of WP6 is to make the robotic platform autonomous and able to learn optimal strategies for therapy planning and monitoring directly from experience and through efficient interaction with physicians. WP6 will play a key role in the overall achievement of project goals because it will provide the basic infrastructure to endow the FUTURA platform with cognitive abilities. In more detail, in WP6 CAMELOT will design and implement novel algorithms and develop software modules based on both supervised and semi-supervised machine learning in order to improve and automate the algorithmic workflow developed in WP4 (therapy planning) and WP5 (therapy monitoring). Ultimately, all the modules developed in WP6 will be responsible for data collection from the different input sources (both pre- and intra-operative) available to the FUTURA platform and will integrate with clinical constraints in order to deliver data for robust and reliable robot control.
The supervised learning algorithms will improve the crucial phase of data fusion and knowledge extraction from radiological images, thus making the system less dependent from human intervention during the design and image-based guidance of therapeutic procedures. The semi-supervised algorithms will leverage key inputs from physicians and support them during the identification of the optimal ultrasonic window to perform the therapy. A common research and development theme underlying all the activities in WP6 is so called optimization theory, which studies optimality conditions for learning, designs efficient training algorithms, and provides guarantees for convergence towards the optimal solution. Optimization is at the very core of almost all machine learning problems and is also the mathematical tool required for planning and controlling the therapeutic treatment. CAMELOT will identify the best State of the Art methods to meet the goals of FUTURA and study how to integrate them within the system architecture to exploit all the resources offered by the innovative robotic system. The following list of Tasks will allow CAMELOT to make significant progresses towards the advanced robotic functionalities foreseen in FUTURA, and to meet all the key requirements of autonomy, adaptability, robustness and safety that will be necessary to validate and use the platform in a real environment.
DESCRIPTION OF WORK
T6.1 Image fusion - CAMELOT only will be involved, and will be solely responsible for the accomplishment of all the activities. CAMELOT will standardize the protocol for pre-operative MR and intra-operative US image acquisition on each robotic arm and design robust algorithms and methods for multi-modal image registration and fusion. The algorithms will be designed to exploit parallel many-core architectures in order to obtain high performance with limited time resource. CAMELOT will also develop a stand-alone application with a simple easy-to-use graphical user interface enabling the physicians to monitor the fusion process and interact with the system during the definition of the therapy plan.
T6.2 Automation of therapy planning - Both SSSA and CAMELOT will be involved. CAMELOT will lead the R&D activities specifically related to machine learning and SSSA will provide support for the integration of the results within other WPs. CAMELOT will be responsible for the accomplishment of all the activities. Supervised machine learning methods will be used to extract information from data and to generate optimal strategies for the therapy. The system will understand which are the main constraints posed by the patient anatomy and suggest viable configurations of the robot arm and medical equipment to execute the therapy on the lesions with minimum impact on other healthy organs or tissues. This part is extremely critical for the overall success of the therapy and CAMELOT will take specific attention to design suitable control mechanisms to exploit the autonomy provided by the cognitive modules without compromising the safety of the patients.
T6.3 Automation of therapy monitoring - Both SSSA and CAMELOT will be involved. CAMELOT will lead the R&D activities specifically related to machine learning and SSSA will provide support for the integration of the results within other WPs. CAMELOT will be responsible for the accomplishment of all the activities. CAMELOT will develop methods for the automatic detection of the relevant regions of interest in the fused images and track them during the different phases of the intervention, thus providing quantitative information on the therapy progression. The tracking of all the regions of interest will make it possible to monitor the successful delivery of the therapy and to control the robot’s arm in order to be compliant with the dynamic environment. The developed software will be integrated in the FUTURA platform in straight combination with the hardware system for guaranteeing a smooth, accurate and responsive management of the hardware modules by the user.