Large scale medical image search technology - a proof of concept

点击翻译按钮获取中文摘要

Radiotherapy is a key method for treating cancer in which high energy radiation is directed at a tumour to disrupt DNA replication and thereby destroy the tumour. However in the process, it is inevitable that surrounding healthy tissue is also irradiated. It is therefore necessary to plan any therapy to maximise the tumour dose while minimising the dose to the healthy tissue.In this process, known as RT planning, teams of clinical experts spend a great deal of time developing a treatment plan prior to its administration. This process is very time consuming and can typically take some tens of hours, leading to high costs and limited throughput for the clinic.A major step in the planning process is the delineation of the tumour and surrounding healthy organs in a medical image scan of the patient, a task known as contouring. The resultant delineations, called RTstructures, are subsequently used to estimate the delivered dose and optimise the treatment plan. Typically, tumour and organ delineation is a laborous manual process.An image processing technique known as atlas-based contouring has been shown to be effective in speeding up this step. Here, expert delineations of healthy organs on an example scan, known as an atlas, are warped automatically onto the scan of the patient. However, it has been found that this method is effective only in cases where the anatomy of the patient is similar in appearance to the atlas.The aim of this proof-of-concept project is to make atlas-based contouring work accurately for most if not all patients. We will do this by developing innovative technology that can build and rapidly search large-scale databases of atlases containing thousands of example delineations representing the wide variability in human anatomy. In a process analogous to Web search of keywords, when applied to a new patient, the system will first retrieve the best matching case or cases from the database and use only those for the atlas-based contouring process.