Truncation compensation and metallic dental implant artefact reduction in PET/MRI attenuation correction using deep learning-based object completion

Abstract

The susceptibility of MRI to metallic objects leads to void MR signal and missing information around metallic implants. In addition, body truncation occurs in MR imaging for large patients who exceed the transaxial field-of-view of the scanner. Body truncation and metal artefacts translate to incomplete MRI-derived attenuation correction (AC) maps, consequently resulting in large quantification errors in PET imaging. In this work, we propose a deep learning-based approach to predict the missing information/regions in MR images affected by metallic artefacts and/or body truncation aiming at reducing quantification errors in PET/MRI. Twenty-five whole-body (WB) co-registered PET, CT, and MR images were used for training and evaluation of the object completion approach. CT-based attenuation corrected PET images were considered as reference for the quantitative evaluation of the proposed approach. Its performance was compared to the 3-class segmentation-based AC approach (containing background air, soft-tissue and lung) obtained from MR images. The metal-induced artefacts affected 8.1 ± 1.8% of the volume of the head region when using the 3-class AC maps. This error reduced to 0.9 ± 0.5% after application of object completion on MR images. Consequently, quantification errors in PET images reduced from −57.5 ± 11% to −18.5 ± 5% in the head region after metal artefact correction. The percentage of the torso volume affected by body truncation in the 3-class AC maps reduced from 9.8 ± 1.9% to 0.6 ± 0.3% after truncation compensation. PET quantification errors in the affected regions were also reduced from −45.5 ± 10% to −9.5 ± 3% after truncation compensation. The quantitative results demonstrated promising performance of the proposed approach towards the completion of MR images corrupted by metal artefacts and/or body truncation in the context of WB PET/MR imaging.

Deep learning–based metal artefact reduction in PET/CT imaging

“…Any issues or image artefacts affecting CT images would be reflected in the attenuationcorrected PET images, potentially leading to misdiagnosis and/or quantitative errors [4,5]. Moreover, CT images commonly serve as standard of reference for the evaluation of MRI-guided attenuation correction [6][7][8] and MRI-only treatment planning in radiation oncology [9].…”

Performance of PROPELLER FSE T2WI in Reducing Metal Artifacts for Patients with Various Material Porcelain Fused To Metal Crown: A Preliminary Study

“…The previous studies showed that the spin-echo (SE) sequence signi cantly reduces the susceptibility artifact compared with the gradient-echo (GRE) sequence, yet, this still did not meet the expected standards. Furthermore, advances in MR sequence (e.g., VAT, SEMAC, MAVRIC, UTE) and serious deep learn-based methods now allow signi cantly improved image quality in the presence of ferromagnetic materials [10][11][12][13][14][15] . The PROPELLER sequence has the advantages of mature technology, imaging easily and high signal-to-noise ratio, and has been widely applied in clinical MR imaging to reduce motion artifacts [17][18][19][20] .…”

“…Furthermore, various deep learning-based approaches were developed to reduce metal artifacts, improve image quality, and even predict the missing information/regions in MR images affected by metal artifacts [13][14][15] . However, the clinical uses of these methods were restricted due to safety and quality control issues, as well as complex principles and higher demand for MRI equipment in hardware and software 16 .…”

The promise of artificial intelligence and deep learning in PET and SPECT imaging

“…In Atlasbased algorithms [80,81], pairs of co-registered MR and CT images (considered as template or atlas) are aligned to the target MR image to generate a continuous attenuation map. The main disadvantage of atlasbased algorithms is the high dependence on the atlas dataset and suboptimal performance for subjects presenting with anatomical abnormalities [82,83].…”