Development and in silico imaging trial evaluation of a deep-learning-based transmission-less attenuation...

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Artificial Intelligence

Background:

• Dopamine transporter (DaT) uptake quantification using SPECT is crucial for Parkinsonism diagnosis and progression monitoring.
• Current methods rely on CT-based attenuation correction (CTAC), which presents challenges like radiation exposure, cost, and system availability.
• CTAC can also introduce errors due to SPECT-CT misalignment.

Purpose of the Study:

• To develop and evaluate a deep learning-based, transmission-less attenuation compensation (AC) method for DaT-SPECT (DaT-CTLESS).
• To assess the performance of DaT-CTLESS in quantifying regional DaT uptake compared to CTAC and uncorrected (UAC) methods.
• To validate the generalizability, repeatability, and robustness of DaT-CTLESS.

Main Methods:

• An in silico imaging trial, ISIT-DaT, was designed to evaluate DaT-CTLESS.
• The study compared DaT-CTLESS with CTAC and UAC for regional DaT uptake quantification.
• Performance metrics included correlation, agreement, ability to distinguish patient groups, generalizability, repeatability, and robustness to training data size.

Main Results:

• DaT-CTLESS demonstrated a significantly higher correlation with CTAC than UAC for regional DaT uptake quantification.
• Excellent agreement was observed between DaT-CTLESS and CTAC.
• DaT-CTLESS effectively distinguished patients with normal versus reduced putamen SBR, showing good generalizability, repeatability, and robustness.

Conclusions:

• DaT-CTLESS provides a viable alternative to CTAC for DaT-SPECT quantification, overcoming CT-related limitations.
• The method shows strong potential for improving Parkinsonism diagnosis and disease management.
• Further clinical evaluation of DaT-CTLESS is warranted based on its promising performance in silico.