Purpose: O-(2-[ 18 F]fluoroethyl)-L-tyrosine (FET), a PET radiotracer of amino acid uptake, has shown potential for diagnosis and treatment planning in patients with glioblastoma (GBM). To improve quantitative assessment of FET PET imaging, we evaluated the repeatability of uptake of this tracer in patients with GBM. Methods: Test-retest FET PET imaging was performed on 8 patients with histologically confirmed GBM, who previously underwent surgical resection of the tumour. Data were acquired according to the protocol of a prospective clinical trial validating FET PET as a clinical tool in GBM. SUV mean , SUV max and SUV 98% metrics were extracted for both test and retest images and used to calculate 95% Bland-Altman limits of agreement (LoA) on lesion-level, as well as on volumes of varying sizes. Impact of healthy brain normalization on repeatability of lesion SUV metrics was evaluated. Results: Tumour LoA were [0.72, 1.46] for SUV mean and SUV total , [0.79,1.23] for SUV max , and [0.80,1.18] for SUV 98% . Healthy brain LoA were [0.80,1.25] for SUV mean , [0.80,1.25] for SUV max , and [0.81,1.23] for SUV 98% . Voxel-level SUV LoA were [0.76, 1.32] for tumour volumes and [0.80, 1.25] for healthy brain. When sampled over maximum volume, SUV LoA were [0.90,1.12] for tumour and [0.92,1.08] for healthy brain. Normalization of uptake using healthy brain volumes was found to improve repeatability, but not after normalization volume size of about 15 cm 3 . Conclusions Advances in Knowledge and Implications for Patient Care: Repeatability of FET PET is comparable to existing tracers such as FDG and FLT. Healthy brain uptake is slightly more repeatable than uptake of tumour volumes. Repeatability was found to increase with sampled volume. SUV normalization between scans using healthy brain uptake should be performed using volumes at least 15 cm 3 in size to ensure best imaging repeatability.

Intro. Current radiation therapy (RT) planning guidelines handle uncertainties in RT using geometric margins. This approach is simple to use but oversimplifies complex underlying processes and is cumbersome for non-homogeneous dose prescriptions. In this work, we characterize the performance of a novel probabilistic target definition and planning (PTP) approach, which uses voxel-level tumor likelihood information in treatment plan optimization. Methods. We expanded a treatment planning system with probabilistic therapy planning functionality that utilizes non-binary target maps (TM) as voxel-level input to dose plan optimization. Different dose plans were calculated and compared for twelve prostate cancer patients with multiparametric magnetic resonance imaging derived TMs. Dose plans were created using both classical and PTP approaches for uniform and integrated dose boost prescriptions. Dose performance between the different approaches was compared using dose benchmarks on target and organ-at-risk (OAR) volumes. Results. Over all dose metrics, PTP was shown to be comparable to classical planning. For plans of uniform dose prescription, the PTP approach created plans within 1 Gy of the classical planning approach across all dose metrics, with no significant differences (p > 0.2). For plans with the integrated dose boost, PTP plans exhibited higher dose heterogeneity, but still showed target doses comparable to the classical approach, without increasing doses to OAR. Conclusion. In this work we introduce direct incorporation of probabilistic target definition into treatment planning. This treatment planning approach can produce both uniform dose plans and plans with integrated dose boosts that are comparable to ones created using classical dose planning. PTP is a flexible way to optimize external beam radiotherapy, as it is not limited by the use of margins. PTP can produce dose plans equivalent to classical planning, while also allows for greater versatility in dose prescription and direct incorporation of patient target definition uncertainty into treatment planning.

Purpose. To investigate image intensity histograms as a potential source of useful imaging biomarkers in both a clinical example of detecting immune-related colitis (irColitis) in 18 F-FDG PET/CT images of immunotherapy patients and an idealized case of classifying digital reference objects (DRO). Methods. Retrospective analysis of bowel 18 F-FDG uptake in N=40 patients receiving immune checkpoint inhibitors was conducted. A CNN trained to segment the bowel was used to generate the histogram of bowel 18 F-FDG uptake, and percentiles of the histogram were considered as potential metrics for detecting inflammation associated with irColitis. A model of the colon was also considered using cylindrical DRO. Classification of DRO with different intensity distributions was undertaken under varying geometry and noise settings. Results. The most predictive biomarker of irColitis was the 95th percentile of the bowel SUV histogram (SUV 95% ). Patients later diagnosed with irColitis had a significantly higher increase in SUV 95% from baseline to first on-treatment PET than patients who did not experience irColitis (p=0.02). An increase in SUV 95% > + 40% separated pre-irColitis change from normal variability with a sensitivity of 75% and specificity of 88%. Furthermore, histogram percentiles were ideal metrics for classifying 'hot center' and 'cold center' DRO, and were robust to varying DRO geometry and noise, and to the presence of spoiler volumes unrelated to the detection task. Conclusions. The 95th percentile of the bowel SUV histogram was the optimal metric for detecting irColitis on 18 F-FDG PET/CT. Image intensity histograms are a promising source of imaging biomarkers for clinical tasks.

Previous studies on personalized radiotherapy (RT) have mostly focused on baseline patient stratification, adapting the treatment plan according to mid-treatment anatomical changes, or dose boosting to selected tumor subregions using mid-treatment radiological findings. However, the question of how to find the optimal adapted plan has not been properly tackled. Moreover, the effect of information uncertainty on the resulting adaptation has not been explored. In this paper, we present a framework to optimally adapt radiation therapy treatments to early radiation treatment response estimates derived from pre- and mid-treatment imaging data while considering the information uncertainty. The framework is based on the optimal stopping in radiation therapy (OSRT) framework. Biological response is quantified using tumor control probability (TCP) and normal tissue complication probability (NTCP) models, and these are directly optimized for in the adaptation step. Two adaptation strategies are discussed: (1) uniform dose adaptation and (2) continuous dose adaptation. In the first strategy, the original fluence-map is simply scaled upwards or downwards, depending on whether dose escalation or de-escalation is deemed appropriate based on the mid-treatment response observed from the radiological images. In the second strategy, a full NTCP-TCP-based fluence map re-optimization is performed to achieve the optimal adapted plans. We retrospectively tested the performance of these strategies on 14 canine head and neck cases treated with tomotherapy, using as response biomarker the change in the 3’-deoxy-3’[(18)F]-fluorothymidine (FLT)-PET signals between the pre- and mid-treatment images, and accounting for information uncertainty. Using a 10% uncertainty level, the two adaptation strategies both yield a noteworthy average improvement in guaranteed (worst-case) TCP.

No abstract