Limited-angle SPECT image reconstruction using deep image prior.

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Reconstruction

Background:

• Limited-angle conditions in single-photon emission computed tomography (SPECT) cause frequency component loss, distorting tomographic images.
• Conventional iterative reconstruction methods struggle to achieve clinically acceptable image quality under these limitations.

Purpose of the Study:

• To develop an advanced limited-angle SPECT image reconstruction method using an end-to-end deep image prior (DIP) framework.
• To improve the quality of reconstructed SPECT images by mitigating distortion caused by missing projection data.

Main Methods:

• An end-to-end deep image prior (DIP) framework was implemented for limited-angle SPECT reconstruction.
• A forward projection model and a binary mask indicating collected data were incorporated into the neural network's loss function.
• The method optimizes the neural network to restore non-collected projection data and reconstruct images.

Main Results:

• The proposed DIP method demonstrated superior performance over existing back-projection methods in numerical simulations, assessed by peak signal-to-noise ratio and structural similarity index measure.
• Analysis using object-specific modulation transfer functions revealed significant improvements in spatial frequency response, even for data from the non-collected angle range.
• Reconstructed tomographic images showed reduced distortion in both simulated and clinical patient data.

Conclusions:

• The end-to-end DIP-based reconstruction method effectively restores lost frequency components in limited-angle SPECT imaging.
• Incorporating a binary mask within the loss function mitigates image distortion, enhancing diagnostic utility.
• This approach offers a promising solution for improving SPECT image quality in challenging limited-angle acquisition scenarios.