Lesion Synthesis Toolbox: Development and Validation of Dedicated Software for Synthesis of Lesions in Raw PET and CT Patient Images

Abstract

Positron Emission Tomography (PET) is one of the most sensitive clinical tools for early detections of small tumours, but its performance is dependent on multiple factors including image quality and human perception. Current methods relying on physical and numerical image quality phantoms are inadequate for evaluating clinical task-based performance, such as limits of lesion detection, due to lack of physiologic realism of the images.In this work, we describe development and validation of the Lesion Synthesis Toolbox, easy-to-use software to synthetize well-characterized, user-defined lesions in real patient PET data prior to image reconstruction and in corresponding Computed Tomography (CT) images on GE Discovery line of PET/CT scanners.We employ synthetic lesions in a preliminary perception study to estimate the limits-of-detection (LOD) in representative clinical setting. Finally, we describe a custom developed, web-based tool for conducting perception studies using synthetic lesions of varying parameters to characterize the LOD. We conclude with a discussion of how these tools can be employed in future studies to compare LOD between imaging methods, image generation algorithms, observers, display types, and computer-based lesion detection solutions.ii

Evaluation of a method based on synthetic data inserted into raw data prior to reconstruction for the assessment of PET scanners

“…The most realistic is particle-tracking-based simulation, which is generically referred to as Monte Carlo method [8,9]. In medical imaging, the SimSET package uses Monte Carlo techniques to model the physical processes and instrumentation, in particular in PET imaging [10]. A major limitation in particle-tracking simulation is the significant computation time which limits the generation of large datasets.…”

“…Analytical simulation is another method that models the average probability of photon interactions instead of individual photon tracking. Therefore, it significantly improves computational cost and can generate large datasets promptly [10][11][12][13][14][15].…”

“…In order to design a hybrid method, Juma [10] used and compared two simulation techniques, SimSET and a forward projector, that generates lesion sinogram estimating events from an estimated activity map. He considered the analytical simulation as the most appropriate method, taking into account simulation times, technical limitations related to file formats and the opportunity to easily simulate time-of-flight (TOF).…”

Uncovering the Limits of Detection of Artificial Intelligence using Synthetic Lesions in Positron Emission Tomography

“…LST is based of the Master's work of Hanif Gabrani-Juma [53], [54] and Quinn de Bourbon, under the supervision of Ran Klein at The Ottawa Hospital. LST can be run directly in MATLAB © by: 1) Copying the LST program files to a local directory.…”