Brady, professor of information engineering at the University of Oxford, UK, detailed a key challenge when using imaging for cancer detection: namely, the sheer diversity of disease indicators that can be imaged. Mammography, for example, is the most appropriate modality for imaging radiologically dense cancer tissues. Biomechanically dense cancer tissue, meanwhile, is best observed using ultrasound, while angiogenesis can be seen using MRI and rapid cell division is picked up by PET scanning.
"There's no single imaging modality that will work for all tumours in all cases," said Brady. What's needed, he told the IPEM delegates, is the combination of images from more than one technique, such as PET/CT, or the emerging yet highly promising MRI/PET systems. "This idea of fusing images to determine the optimal way to present information is something we'll become increasingly familiar with in years to come."
Of course, there are plenty of other obstacles to face when it comes to analysing medical images, one of which is the "apparently impossible task" of detecting subtle biological changes in complex images with inherently poor signal-to-noise ratios. The solution, according to Brady, lies in scientific modelling. "In order to make progress, we have to devise and mobilize models: physical, mathematical, biological and chemical," he said.
Clinical imaging
Brady and his research group are currently focusing much of their efforts on image analysis in the field of colorectal and liver cancer. Their recent work includes developing a model that uses medical images of a tumour's morphology to determine its heterogeneity, enabled by applying a basic biological model of tumour growth.
The simplest tumour-growth model envisages the tumour as an expanding sphere that sprouts additional spheres as it progresses. A tumour's heterogeneity can impact significantly upon its development and therapeutic response: if the secondary growths have different compositions to the parent lesion they may exhibit differing responses to radio- or chemotherapy. "Image analysis gives us morphological information, related to the fundamental biology of what's going on in a tumour," Brady noted.
By using medical imaging to determine the morphology of such a composite lesion, physicians can measure the degree of response in the different spheres following treatment. With a heterogeneous tumour, some areas may shrink while others continue to grow. This technique therefore provides a more accurate reflection of therapeutic response than that given by simple tumour-volume measurements.
Scientific modelling has also proven beneficial for improving the registration of pre- and post-chemotherapy images, used to assess the effects of neoadjuvent chemotherapy prior to surgical resection, for example. Brady described a recent study undertaken by his research group in which six well known non-rigid registration algorithms were employed to align a series of pre- and post-therapy colorectal MR images.
Typically, the algorithms failed to align the images in about 30% of cases. In terms of accuracy, the misalignment of lymph nodes ranged from 7 to 30 mm for the six schemes under investigation, often too high to be of clinical value. The problem, Brady explained, lies not in the algorithms themselves but in the fact that they are too general and don't embody any specific anatomical knowledge.
To address this issue, the researchers built a simple model that identifies the bone structures and other well defined features seen in a colorectal MR image. When this anatomical model was incorporated into the six algorithms and reapplied to the same series of MR data sets, the misalignment was reduced to clinically useful levels (less than 6 mm in half of the cases). "The maths was OK," Brady explained. "It was the lack of application-specific knowledge that was the real problem."
Exciting science
Brady concluded his presentation with an insight into the use of PET imaging for determining a tumour's hypoxia, a crucial factor in determining its response to chemotherapy or radiotherapy. The application of molecular imaging for non-invasive in vivo measurement of oxygenation status is garnering enormous interest at present.
One issue under debate in the literature is whether fluorodeoxyglucose (FDG) uptake can act as a marker for hypoxia in PET imaging. Some publications conclude that it correlates perfectly, while others report negligible correlation. By developing mathematical and biological models for tracer pharmacokinetics, the researchers determined that FDG uptake is significantly influenced by the activity of oncogenic pathways such as Akt (a molecule involved in cellular survival).
"Using PET images together with biological models [helps us] understand biochemical processes as a result of image analysis. We're going to see a vast amount of this over the next few years," said Brady. "I honestly think molecular imaging is one of the most exciting pieces of science I've ever come across."