Alongside mainstream scanning modalities such as MRI, X-rays or Computerized Tomography (“CAT” or “CT”), Positron Emission Tomography (PET) imaging is widely used in routine medical practice, in particular in the management of cancer. PET scans offer unique insight into the metabolic status and activity of a tumor mass, whereas the other above mentioned modalities rather provide information of an anatomical or non-metabolic nature. As such, the analysis of PET images is key in understanding aspects of the disease that are not captured by these other modalities. In particular, PET imaging can be useful for early assessment of therapeutic response, by capturing early changes in metabolism resulting from therapy. This enhances opportunities for timely and individualized treatment planning. The question of personalized treatment assessment and decision-making arises for a range of cancers (carcinomas in particular), and potentially also for the design of novel protocols for clinical trials.

The authors focus here on the comparative analysis of PET imaging data obtained pre- and post-chemotherapy in order to assess treatment response. They investigate how to appropriately evaluate the statistical uncertainty in the measured response for an individual patient. The information content within a PET scan is highly statistical, in that it typically comprises of a large amount of information that is acquired subject to specific types of variability. Sources of such variability are mainly inherent to the scanner characteristics, the patient’s own physiological features, and the dose of PET radioactive tracer that is injected into the patient for imaging. Moreover, the comparison of two scans for a given patient is also made complicated by the random sources of variability such as the way the patient lies on the table, differences in metabolic activity between scans, the potential development or shrinkage of the tumor mass, etc.

In current clinical practice, PET scans acquired at two successive treatment time-points are compared in terms of a set of summary statictics that are obtained independently from each PET image. Therapeutic effectiveness is then evaluated and rated on the basis of significant changes in these summaries (for example, an increase greater than 20% in maximum overall activity would be indicative of disease progression). In their work the authors suggest that the comparative analysis of these sets of PET scans should be performed instead by pairing up the pre- and post-therapy scans (by co-registration of the two images). They have developed a mathematical framework (based on a Gamma distribution) that incorporates typical characteristics of PET measurements which allow for this paired analysis to be carried out despite the difficulties introduced by variations occurring between imaging time points. The authors provide mathematical formulations for the estimation of the treatment effect and its variability. They also describe simulation studies to explore the performance of this comparative assessment in the context of testing for a treatment effect. The impact of mis-registration (or mis-alignment) errors and how test power is affected by estimation of variability using simplified sampling assumptions is evaluated and clarified. The results illustrate a potentially marked benefit in using a properly constructed paired approach. Remarkably the power of the paired analysis is maintained even if the pre- and post- image data are even rather poorly aligned. The authors provide a theoretical explanation for this. The methodology is further illustrated in the context of a series of FDG-PET sarcoma patient studies. These data demonstrate the additional prognostic value of the proposed treatment effect test statistic.

Statistical assessment of treatment response in a cancer patient based on pre‐therapy and post‐therapy FDG‐PET scans

E. Wolsztynski F. O’Sullivan J. O’Sullivan J. F. Eary.

Statistics in Medicine, Volume 36, Issue 7, 30 March 2018, pages 1172-1200, https://doi.org/10.1002/sim.7198