Diagnostic imaging is the use of non-invasive techniques to produce information, in the form of images, about structures of the body, and has wide applications for the diagnosis and study of disease. The rapid development of computer technology has resulted in increased availability and major advances in diagnostic imaging techniques in both diagnostic and research settings. This has increased the use and applications for diagnostic imaging and allows for new approaches to investigations of the presence, development, and even underlying causes of disease. These technological advances also present challenges, notably in the interpretation of the often large and complex data sets produced by some studies. Image evaluation is often done by a specialist radiologist, however, artificial intelligence (AI), with its capacity to handle and process large amounts of data, automation of repetitive tasks, and potential to remove human error, now offers major possibilities for diagnostic imaging research and diagnosis.
My research focuses on the use of computed tomography (CT), a method that calculates the amount of x-rays that are absorbed and deflected by specific tissues (attenuation) and produces three-dimensional cross-sectional images of the body. The x-ray attenuation data in the CT images gives information about the amount and distribution of specific types of tissues in the body and the three-dimensional data is ideal for evaluating complex structures such as joints. X-ray attenuation values in CT images are known to vary due to how the CT examination is performed and external factors. How much variation and what factors cause the variation are areas of focus in my research.
When a diagnostic imaging technique is used in a new way to investigate a disease, the specific changes in the image that indicate the disease process need to be identified. Changes can be identified subjectively (a person evaluating the appearance of the images) or objectively (measurement of the image data). Suspected changes that are identified need to be compared to a reference method to confirm the true cause of changes, a process that is called validation.
Furthermore, it is highly desirable to identify validated, and accurate “key lesions”, a change in an image that is strongly predictive of a specific disease. Identification of “key lesions” not only has the potential to improve diagnostic accuracy but can also provide simplified image interpretation methods for veterinarians and researchers who are not diagnostic imaging specialists. In this lecture, I will describe the approach to the identification and validation of changes in images in my research and present examples of whole-body CT examinations of cats to investigate joint disease and associations to changes in body composition and bone density.
At the end of the lecture, I will outline my research visions, including investigations of methods for detecting “key lesions” in radiographs based on validated CT changes, non-invasive longitudinal studies of joint disease, and studies to develop AI methods based on x-ray attenuation values in CT images. The availability of modern advanced diagnostic imaging methods offers great possibilities for veterinary research but brings the challenge of extracting specific information from large data sets and ensuring that the interpretation of that data is correct. This research will involve collaboration between radiologists, computer scientists, and pathologists.