Accelerating radiotherapy and neurosurgery treatment for patients with brain tumours

Nearly 40% of all patients cured of cancer receive radiotherapy. Yet the methods used to target the radiations to the tumours are extremely time-consuming. New computer vision technologies could help save 4 million man hours globally.

Radiotherapy is an important anti-cancer therapy. Modern radiotherapy uses information from CT, MRI and PET scans to build virtual representations of the tumour and patient anatomy in a computer workstation. In this way, radiation oncologists can deliver precisely targeted treatment to the tumour, whilst minimising the risk of injury to adjacent healthy tissues. This is particularly important in the treatment of brain tumours, where injury to surrounding tissues can lead to serious disability.

At present, the radiation oncologist must painstakingly outline the edge of the tumour, and all healthy adjacent tissues for every patient undergoing radiotherapy. For complex brain tumour treatment, this can take up to 4 hours for a single patient. Automatic image segmentation tools have been developed to try and reduce the amount of time taken for this process, using information from an anatomical atlas to infer the extent of a particular organ. There is a significant learning curve for the use of such systems, particularly when the oncologist is required to correct the output of an automated segmentation system.
The Geodesic Image Segmentation system Geos3D developed by Antonio Criminisi and his colleagues at Microsoft Research presents a highly attractive alternative to existing technologies, using an interactive segmentation tool to achieve fast segmentation of anatomical structures in a matter of seconds. The technique shows great potential for workflow acceleration in radiation oncology. At the global level, some 10,000 linear accelerators are used for the treatment of cancer patients, and if a 10-fold reduction in segmentation time could be achieved for the most complex radiotherapy treatment, this would equate to a 4 million man hour time saving.

Interactive segmentation of 3D medical images for various organs and structures of the body

Dr Criminisi and Dr Jena are working on Geos3D, adapting it to work with radiotherapy image objects, and assessing the ability of the system to segment normal intact anatomical structures. The speed of operation, simple user interface, and ease of interaction make Geos3D an ideal tool for image segmentation in radiation oncology. By performing the bulk of the necessary image segmentation in a matter of seconds, the oncologist has more time to evaluate the patient anatomy and imaging findings, and map out the location and extent of the tumour in more detail. They propose to evaluate Geos3D in the clinic, comparing it with a panel of expert oncologists, and then move on to address the difficult task of automatic segmentation of brain tumours from their surrounding tissues.

Having generated a high-precision radiotherapy treatment, the radiation oncologist must make the final judgement as to whether or not the proposed treatment will be safe and effective for their patient. This requires the oncologist to assess multiple parameters in 3-dimensional space and is a labour intensive process. New volumetric rendering techniques offer unique potential for the visualisation of multiple-parameter datasets over anatomical image data. Toby Sharp and his colleagues at Microsoft Research have developed a volumetric rendering tool using acceleration of graphics cards in computers to achieve real-time interactive displays of 3D image datasets. Dr Criminisi and Dr Jena will work to adapt the volumetric rendering tool to provide an intuitive visualisation tool for radiation oncologists and medical physicists, reducing the time taken to review radiotherapy plans, and facilitating rapid treatment for patients.

For brain tumour patients, treatment typically involves a combination of surgery and radiotherapy techniques, which are used in conjunction to maximise the likelihood of treatment success. Neurosurgery, like radiation oncology, has become increasingly dependent on the use of high resolution brain imaging to plan surgical treatment. Often the neurosurgeon may need to refer to patient imaging data during the course of the operation. Interactive image display systems for neurosurgery are hampered by the need for the surgeon to maintain a sterile field with their hands. Using the natural movement tracking technology underlying the Microsoft Kinect hardware, research will be performed on a natural user interface for image navigation, which can be operated by the surgeon without having to remove gloves or compromise the sterile surgical field. Reducing the time taken for image review during surgery brings benefit in terms of reduced time under anaesthesia, and also a lower risk of introducing infection during the operation.

Dr. Raj Jena, Clinician Scientist & Consultant Radiation Oncologist, Department of Oncology, Cambridge University Hospitals NHS Foundation Trust

We are pleased to host on our website contributions from external experts and stakeholders and we are grateful for their time and thoughts. The content developed by our guest bloggers is purely the reflections of the author and does not necessarily reflect Microsoft positions.

Related content