Preclinical application and evaluation of an energy-based spectral unmixing method for multi-radionuclide SPECT.

Abstract

Objective.Crosstalk poses significant challenges in multi-radionuclide SPECT, where γ-rays from one radionuclide are incorrectly attributed to a different radionuclide, compromising image fidelity and interpretations. Although the triple energy window (TEW) method is widely accepted for correcting crosstalk in the acquired energy spectrum, its effectiveness depends on the gamma camera's energy resolution and statistical counts on a pixel-by-pixel basis. Furthermore, subtractive corrections applied before image reconstruction, such as TEW, alter the Poisson distribution of detected events, negatively impacting statistical reconstruction accuracy, reconstructed image noise, and image quality. To address these challenges, we introduce an energy-based spectral unmixing crosstalk correction approach in a preclinical setting using the open-source Synergistic Image Reconstruction Framework (SIRF) and Core Imaging Library (CIL). This method employs a mixing matrix to decompose hyperspectral data into distinct radionuclide distributions during synergistic reconstruction. Approach.Data were acquired with the Cubresa Spark preclinical SPECT scanner using^99mTc/¹²³I in measurement and simulation, and^99mTc/¹¹¹In in simulation. A mouse-sized phantom study employed five 4 MBq line sources with linearly varying mixtures of radioactivity. The images were reconstructed using primary energy windows, TEW, spectral unmixing, and gold standard simulation data representing activity distributions free of crosstalk. Performance evaluation used root-mean-square error (RMSE) for activity quantitation and image contrast, while noise was assessed separately in each image.Main results.In the^99mTc/¹²³I measurement, primary energy windows yielded a mean quantitation RMSE of 0.49, with improvements of 32% observed with TEW and 74% with spectral unmixing. Simulation results were consistent, giving an RMSE of 0.44 for primary energy windows, with improvements of 51%, 83%, and 95% observed with TEW, spectral unmixing, and gold standard methods. For the^99mTc/¹¹¹In simulation, the mean quantitation RMSE was 0.15 with primary energy windows, while TEW, spectral unmixing, and gold standard methods improved accuracy by 57%, 57%, and 84%, respectively. Across all cases, spectral unmixing also demonstrated superior contrast and reduced noise compared to TEW.Significance.This energy-based spectral unmixing approach provides enhanced crosstalk correction, quantitative accuracy, and image quality compared to conventional methods, facilitating flexible application in diverse SPECT systems and radionuclide combinations.