Multiple training interventions significantly improve reproducibility of PET/CT-based lung cancer radiotherapy target volume delineation using an IAEA study protocol

BACKGROUND AND PURPOSE: To assess the impact of a standardized delineation protocol and training interventions on PET/CT-based target volume delineation (TVD) in NSCLC in a multicenter setting.

MATERIAL AND METHODS: Over a one-year period, 11 pairs, comprised each of a radiation oncologist and nuclear medicine physician with limited experience in PET/CT-based TVD for NSCLC from nine different countries took part in a training program through an International Atomic Energy Agency (IAEA) study (NCT02247713). Teams delineated gross tumor volume of the primary tumor, during and after training interventions, according to a provided delineation protocol. In-house developed software recorded the performed delineations, to allow visual inspection of strategies and to assess delineation accuracy.

RESULTS: Following the first training, overall concordance indices for 3 repetitive cases increased from 0.57±0.07 to 0.66±0.07. The overall mean surface distance between observer and expert contours decreased from -0.40±0.03 cm to -0.01±0.33 cm. After further training overall concordance indices for another 3 repetitive cases further increased from 0.64±0.06 to 0.80±0.05 (p=0.01). Mean surface distances decreased from -0.34±0.16 cm to -0.05±0.20 cm (p=0.01).

CONCLUSION: Multiple training interventions improve PET/CT-based TVD delineation accuracy in NSCLC and reduces interobserver variation.

A parallel pattern for iterative stencil + reduce

We advocate the Loop-of-stencil-reduce pattern as a means of simplifying the implementation of data-parallel programs on heterogeneous multi-core platforms. Loop-of-stencil-reduce is general enough to subsume map, reduce, map-reduce, stencil, stencil-reduce, and, crucially, their usage in a loop in both data-parallel and streaming applications, or a combination of both. The pattern makes it possible to deploy a single stencil computation kernel on different GPUs. We discuss the implementation of Loop-of-stencil-reduce in FastFlow, a framework for the implementation of applications based on the parallel patterns. Experiments are presented to illustrate the use of Loop-of-stencil-reduce in developing data-parallel kernels running on heterogeneous systems.