Session Item

In oncology, follow-up imaging is designed to assess treatment response, detect secondary cancers, or record treatment-induced toxicities. Regular follow-up imaging offers a method to visualise a patient’s natural progression through time, where they may grow or undergo large anatomical changes. Such imaging provides information about the course of disease that you cannot obtain from images acquired before or during treatment.

The value of using follow-up images has been demonstrated in a range of research applications including early response assessment, automated cancer detection, and quantitative toxicity measurement. Often, this type of research requires prospective data collection of high-quality, standardised imaging at a specified frequency. It is possible to develop new imaging sequences for optimal follow-up, though this should not come at the cost of patient comfort.

Increasingly, researchers are interested in the use of ‘real-world’ retrospective data. With retrospective research, the original purpose of imaging is often not the same purpose the image is required for in research. This means images are often sub-optimal for the task at hand - a frustrating reality for researchers hoping to perform a retrospective analysis of follow-up images.

There are additional challenges in using ‘real-world’ follow-up imaging, as adherence rates can be low, and imaging is often performed at local centres where different protocols will be implemented. Inclusion criteria for typical research methods (e.g., single modality) are too strict in this context. Situations exist where researchers are required to ‘make the most’ of the data they already have. For example, in late-effects research 5+ years of follow-up data is required before investigation.

Advances in image processing can help us ‘deal with follow-up images’ in research. Super-resolution can increase the quality of low-resolution data eliminating the need for costly and time-consuming sequences in clinical practice. Image synthesis can create new datasets that replicate imaging protocols that are otherwise impractical to acquire. Alternatively, information can be combined from multiple images to improve the quality of existing datasets.