University of Toronto
Dr. Caldwell has a strong interest in applying methods of digital image analysis and artificial intelligence to clinical problems. For example, Dr. Caldwell is conducting research focused on defining the location and extent of tumours accurately and reproducibly for radiation therapy planning purposes. Currently, the tumour contours are defined manually by physician experts from computed tomographic (CT) images. While it is often possible for the physician to determine if a tumour exists in a general area, it is much more difficult for the expert to determine the boundaries of the tumour. Dr. Caldwell and colleagues have studied this problem for both lung and head and neck cancers, finding that, in many cases, there may be relatively little overlap among the tumour volumes defined by different experts. This is a crucial problem for radiation therapy planning, as the tumour target must be accurately defined for a curative radiation dose to be delivered without giving too high a radiation dose to surrounding normal tissues.
Dr. Caldwell has found that the addition of functional information from Positron Emission Tomography (PET) imaging of the distribution of radio-labeled sugar molecules can lead to improvement in tumour volume definition. In addition, recent work has pointed the way toward using texture information, extracted from both CT and PET images, to automatically define the tumour target. Some of the types of information that can be derived by processing PET/CT images are shown in the figure below. Projects for interested students may be arranged in several areas, including (1) developing methods of combining different sources of information (i.e., anatomic images, functional images and non-imaging information such as measures of lung function) to devise radiation therapy targets, (2) developing methods of quantative assessment of tumour response to treatment from functional images and (3) investigation of the effect of imaging protocols on the quantitative nature of the information from PET images.