Search

As more complex predictive models are used for gamma-ray spectral analysis, methods are needed to probe and understand their predictions and behavior. Recent work has begun to bring the latest techniques from the field of Explainable Artificial Intelligence (XAI) into the applications of gamma-ray spectroscopy, including the introduction of gradient-based methods like saliency mapping and Gradient-weighted Class Activation Mapping (Grad-CAM), and black box methods like Local Interpretable Model-agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP). In addition, new sources of synthetic radiological data are becoming available, and these new data sets present opportunities to train models using more data than ever before. In this work, we use a neural network model trained on synthetic NaI(Tl) urban search data to compare some of these explanation methods and identify modifications that need to be applied to adapt the methods to gamma-ray spectral data. We find that the black box methods LIME and SHAP are especially accurate in their results, and recommend SHAP since it requires little hyperparameter tuning. We also propose and demonstrate a technique for generating counterfactual explanations using orthogonal projections of LIME and SHAP explanations.

By combining radiation detection technologies with robotics sensing, the ability to continuously conduct gamma-ray imaging using freely-moving systems was demonstrated in 2015.1 This new method, which was named free-moving 3D Scene Data Fusion (SDF), was then applied to mapping radioactive contamination and to contextualizing the extent of contamination and the efficacy of radiological clean-up efforts.2,3 Since then, further studies into the types of radiation detection systems to which SDF could be applied resulted in the discovery and demonstration that neutron activity could be mapped using neutron-sensitive CLLBC scintillators, arrays of pix-elated CZT detectors could be used to create multi-modal imagers, and more rudimentary detector systems such as arrays of four CsI modules could still achieve good-quality mapping by inferring source positioning through the encoded modulation of source-to-detector distance. This paper provides an overview of the SDF technology, highlights recent measurements leveraging SDF-equipped systems, discusses the continued development of quantitative algorithms4,5 and their ramifications for developing autonomous SDF-capabilities, and summarizes future directions of research and application development for free moving radiation detection systems.

Integration of contextual sensors into vehicle-borne mobile radiation detector systems delivers a rich description of the environment that could be used to estimate the complex and variable environmental gamma-ray backgrounds in urban areas. The predictions could potentially increase the sensitivity to illicit radiological and nuclear materials and could provide realistic inputs to urban radiological search simulations and algorithms. Recent work in this field has focused mainly on the predictive power of segmenting and classifying imagery from cameras and elected in its approach to aggregate the locations of gamma-ray interactions within the fielded detector array to a single point. This work builds upon the previous effort by leveraging LiDARs to create a 3D representation of the detector system and the surrounding scenery and demonstrates further improvement in the capability of attributing observed gamma-ray backgrounds to classes of surrounding materials.