Our paper on developing a no-gold-standard (NGS) technique to evaluate quantitative imaging methods with patient data in the absence of ground truth was accepted for publication in a special section of the Journal of Medical Imaging (link).
This work was in collaboration with radiologists at the Johns Hopkins Division of Nuclear Medicine. We were interested in being able to evaluate segmentation methods for PET images based on how reliably they could assist in computing quantitative imaging biomarkers such as metabolic tumor volume (MTV). The issue was that a gold standard for the imaging biomarker value was not available, as is typically the case. To address this issue, in this paper, we developed a NGS technique, along with developing methods to overcome some practical difficulties in applying this technique to patient data. The developed technique was then applied to evaluate four segmentation methods for FDG-PET imaging on the task of measuring metabolic tumor volume.
The developed NGS framework for evaluating PET segmentation methods on the task of measuring metabolic tumor volume (MTV).
The idea of evaluating imaging methods in the absence of ground truth is undoubtedly useful but has continued to confound! In this paper, we provide an intuitive explanation of the technique. Let us know what you think!
The special section of the Journal of Medical Imaging is devoted to development, challenges, and opportunities of PET imaging (link) and was jointly edited by Drs. Norbert J. Pelc, Paul E. Kinahan, and Roderic I. Pettigrew.