Optimisation Of Static [F-18]Faza Pet Threshold Analysis For Hypoxia Detection In Non-Small Cell Lung Cancer Patients

1749 Introduction: Hypoxia develops in solid tumours when oxygen demand exceeds supply and is associated with poor treatment response and prognosis [1,2]. PET imaging with radiolabelled nitroimidazoles such as [18F]FAZA can be used to measure tumour hypoxia to select patients for hypoxia-targeted therapies. A threshold based on a normoxic reference region is often applied to static PET images to demarcate hypoxic tissue [3,4]. However, a consensus has yet to be reached on the most appropriate definition for this threshold [5]. Aims: To compare fixed and confidence interval (CI) based thresholds using either muscle or image derived blood (aorta) as the reference; to investigate the influence of image noise by varying scan duration, image smoothing and region delineation; and to quantify measurement uncertainty in order to maximise hypoxia detectability. Methods: Twelve patients with non-small cell lung cancer were imaged using a Siemens TrueV PET-CT scanner after bolus injection of [18F]FAZA (373±5 MBq) employing a dual-dynamic scan protocol (0-60 min and 90-150 min). Six of these patients were imaged on return visits within 1-8 days for repeatability data. The tumour (median 8.5 cm3, range 3.5-308 cm3), descending aorta and contralateral back muscles were manually contoured on the CT scan at the tissue boundary. Increasing Gaussian smoothing was applied to the PET image and edges were sequentially removed from the muscle contour via erosion. The hypoxic volume was derived using the 95% CI of the muscle histogram and compared against two commonly used fixed thresholds, 1.2 and 1.4. The hypoxic fraction was computed by dividing the hypoxic volume by the total lesion volume. The voxel uncertainty was estimated by subtracting the median filtered image from the muscle region assuming local homogeneity. Results: The muscle group exhibited a slow equilibration time (71 ± 24 min) compared to the tumour (7.3±4.1 min) confirming that it had reached equilibrium with blood after 90 min. When using a fixed threshold, the hypoxic fraction of the reference muscle exceeded 20% in some patients and decreased as a function of contour erosion (Figure 1a). In contrast, CI based thresholds adjust to define the maximum measurable hypoxic fraction in the reference for a given image quality, in this case, 2.5% for a 95% CI. Figure 1b shows an example of muscle and lesion histograms overlaid with the corresponding thresholds, illustrating that the 95% CI defines both high (hypoxia) and low (necrosis or edges) uptake regions within the lesion. Gaussian smoothing improves uncertainty by reducing image noise, but by adding interregional blurring, excessive smoothing reduces hypoxia detectability (Figure 1c). Lengthening scan duration also reduces noise but without blurring the image and leads to improve hypoxia detectability (Figure 1d). The extent of the uncertainty on the hypoxic fraction is also influenced by lesion size and spatial distribution. Conclusions: Fixed thresholds have the potential to severely over or underestimate detectable hypoxia due to their insensitivity to image noise. CI-based thresholds adjust to the noise to maximise hypoxia detectability and as such benefit from lengthening scan duration and optimised smoothing. References: 1. McKeown SR, Br. J. Radiol. 2014;87:20130676. 2. Vaupel P, Cancer Metastasis Rev. 2007;26:225-39. 3. Koh W-J, Int. J. Radiat. Oncol. Biol. Phys. 1992;22:199-212. 4. Taylor E, Phys. Med. Biol. 2016;61:7957. 5. Chirla R, Phys. Med. 2016;32:23-35.