Jan 29, 2013
PET features help predict therapy response
Virtually all cancers exhibit intra-tumour heterogeneity, a characteristic that plays an important role in tumour growth and may also contribute to treatment failure. Quantifying this heterogeneity could enable prediction and monitoring of a tumour's response to treatment. With this aim, a UK research team has studied a range of textural variables derived from PET images, and explored their associations with clinical response to chemotherapy in breast-cancer patients (Phys. Med. Biol. 58 187).
The researchers, from Imperial College London and King's College London, analysed PET images using the tracer 18F-fluorothymidine (FLT), the uptake of which is correlated with cell proliferation in breast cancer. They employed texture analysis to assess the heterogeneity of cell proliferation in individual tumours and explored potential associations between textural variables and clinical response to chemotherapy.
"PET is a powerful molecular-imaging modality that can be used to study many processes that are specific to tumour biology," explained Julien Willaime, from Imperial's Comprehensive Cancer Imaging Centre. "Chemotherapy agents primarily target proliferating cells. Our aim was to assess whether heterogeneous FLT-PET uptake in tumour volumes – an indirect measure of proliferation – can be reliably quantified using textural metrics. And if so, whether this information could potentially be used to discriminate between patient groups."
Willaime and colleagues examined FLT-PET images from 15 breast cancer patients. All patients underwent one scan prior to the start of chemotherapy and one week into treatment, and some had an additional pre-treatment validation scan. PET images were reconstructed using the OSEM iterative method and compared with filtered back-projection (FBP) reconstructed images.
The researchers calculated 28 textural variables for each FLT-PET image. Five were derived from the intensity distribution of radiotracer uptake, to measure intra-tumour uptake variability. The remaining variables described the spatial organization of radiotracer uptake. Standardized uptake values (SUV) were also determined 60 minutes after radiotracer injection.
The team first studied differences in textural variable values across tissue types. Mean radiotracer uptake (SUVmean) was lowest in normal breast, medium in tumours and nodes, and highest in the liver and vertebra. The values of two intensity-based features – the coefficient of variation (CV) and EntropyFOS, a measure of information content – suggested that the uptake of FLT in healthy breast tissues was characterized by the predominance of a group of voxels with similar intensity, but different from the mean intensity. Only slight variations in spatial textural features were observed across tissue types.
For five of the patients who had two FLT-PET scans prior to therapy, the researchers used an intra-class correlation coefficient (ICC) to assess the repeatability of textural feature measurements. An ICC of 0.7 or above indicates that intra-patient variability is reasonably small compared to inter-patient variability, and suggests that the feature could potentially be used to study associations with treatment outcome.
To explore such associations, they selected a subset of eight textural features that had ICC of 0.7 or above and the same intra-patient variability range (below ±30%) as SUV measurements. These features were: CV EntropyFOS, AUC-CSH, ContrastGLCM, Dissimilarity, Homogeneity, ContrastNGTDM and Complexity.
They then tested associations between these features (as measured on pre-treatment PET scans) and the clinical response to chemotherapy at 60 days. Potential correlations were evaluated in 11 patients: six responders and five non-responders.
Intensity histograms of FLT uptake (at baseline) were positively skewed, with a greater average shift in the intensity distribution seen for non-responding patients than responders. Better separation of the groups was achieved when using OSEM rather than FBP to reconstruct the images.
Values of CV and (for certain normalization parameters) Complexity were statistically greater in lesions of non-responders. Conversely, AUC-CSH and EntropyFOS values were greater in lesions of responders. SUVmean at baseline was also greater, although not statistically significant, in lesions of responders versus non-responders.
The researchers surmised that this technique could help inform the likelihood of response to neo-adjuvant chemotherapy before treatment starts. They note that observed differences in radiotracer distribution between the two groups at baseline may be due to the presence of fewer highly proliferative tumour cells in lesions of non-responders than responders.
Willaime and colleagues also examined changes between baseline and one week after the start of therapy. They saw that a number of textural features were altered following chemotherapy, and suggest that these should be evaluated for clinical relevance in larger clinical trials.
For example, true increases in CV and Homogeneity were seen for four of six responder patients after one week of therapy. No changes in these features were noted for the other two responders. A true decrease in CV was seen for one non-responder patient, and no changes for the others. No true changes in Homogeneity were observed for any of the non-responders.
Willaime noted that it may potentially be possible to combine results from more than one textural variable to increase the technique's predictive power. "Some of the features might provide complementary information and enhance the discriminatory power. We are now combining intensity and spatial features and are planning to extend the technique to bigger sample sizes," he told medicalphysicsweb.
• Related articles in PMB
Quantification of intra-tumour cell proliferation heterogeneity using imaging descriptors of 18F fluorothymidine-positron emission tomography
J M Y Willaime et al Phys. Med. Biol. 58 187
Kinetic filtering of [18F]Fluorothymidine in positron emission tomography studies
Katherine R Gray et al Phys. Med. Biol. 55 695
About the author
Tami Freeman is editor of medicalphysicsweb.