Spatio-Temporal Positron Emission Tomography Reconstruction with Attenuation and Motion Correction

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Nuclear Medicine

Background:

• Detecting small cancer lesions in Positron Emission Tomography (PET) is challenging due to limited scanner resolution.
• Current PET imaging struggles with motion and attenuation artifacts, further complicating lesion detection.

Purpose of the Study:

• To investigate the integration of image-registering convolutional neural networks with forward modeling for improved PET image reconstruction.
• To develop and evaluate novel PET reconstruction algorithms for enhanced detection of small pulmonary lesions, incorporating motion and attenuation correction.

Main Methods:

• Two novel PET reconstruction algorithms were developed, combining convolutional neural networks with static data acquisition modeling (forward model).
• Performance evaluation utilized synthetic data, assessing detectability of small pulmonary lesions using figures of merit, visual inspection, and an ideal observer model (Channelised Hotelling Observer).

Main Results:

• Traditional figures of merit (PSNR, RC, SDR) provided inconclusive results for evaluating the proposed algorithms.
• Visual inspection and the Channelised Hotelling Observer indicated that the proposed algorithms significantly outperform current clinical practices in detecting small pulmonary lesions.

Conclusions:

• The proposed PET reconstruction algorithms, enhanced by convolutional neural networks and forward modeling, demonstrate improved detectability of small pulmonary lesions.
• These advanced algorithms offer a promising solution to a long-standing challenge in PET imaging, potentially improving early cancer diagnosis.