Evaluation of Deep Learning–Based Approaches to Segment Bowel Air Pockets and Generate Pelvic Attenuation Maps from CAIPIRINHA-Accelerated Dixon MR Images

Abstract

Attenuation correction (AC) remains a challenge in pelvis PET/MR imaging. In addition to the segmentation/model-based approaches, deep learning methods have shown promise in synthesizing accurate pelvis attenuation maps (μ-maps). However, these methods often misclassify air pockets in the digestive tract, which can introduce bias in the reconstructed PET images. The aims of this work were to develop deep learning-based methods to automatically segment air pockets and generate pseudo-CT images from CAIPIRINHA-accelerated MR Dixon images. Methods: A convolutional neural network (CNN) was trained to segment air pockets using 3D CAIPIRINHA-accelerated MR Dixon datasets from 35 subjects and was evaluated against semi-automated segmentations. A separate CNN was trained to synthesize pseudo-CT μ-maps from the Dixon images. Its accuracy was evaluated by comparing the deep learning-, model-and CT-based μ-maps using data from 30 of the subjects. Finally, the impact of different μ-maps and air pocket segmentation methods on the PET quantification was investigated. Results: Air pockets segmented using the CNN agreed well with semi-automated segmentations, with a mean Dice similarity coefficient of 0.75. Volumetric similarity score between two segmentations was 0.85  0.14. The mean absolute relative change (RCs) with respect to the CT-based μ-maps were 2.6% and 5.1% in the whole pelvis for the deep learning and model-based μ-maps, respectively. The average RC between PET images reconstructed with deep learning and CT-based μ-maps was 2.6%. Conclusion: We presented a deep learningbased method to automatically segment air pockets from CAIPIRINHA-accelerated Dixon images with comparable accuracy to semi-automatic segmentations. We also showed that the μ-maps synthesized using a deep learning-based method from CAIPIRINHA-accelerated Dixon images are more accurate than those generated with the model-based approach available on integrated PET/MRI scanner.

Quantitative evaluation of a deep learning-based framework to generate whole-body attenuation maps using LSO background radiation in long axial FOV PET scanners

“…The use of deep learning-based methods in PET attenuation correction has been increasingly popular, particularly in PET/MRI imaging where lack of CT-based attenuation maps introduced significant challenges to accurate PET quantification [45]. In previous work, CNNs were trained using coregistered MR and CT images to generate pseudo-CT based µ-maps for head [46,47] and pelvis [48][49][50], which were shown to be more accurate compared to vendor-provided atlas based µ-maps. Besides, the use of supervised deep learning techniques such as CNN has limited performance in generating whole-body µ-maps as these techniques require perfectly aligned MR and CT whole-body images which is not straightforward.…”

A review of PET attenuation correction methods for PET-MR

“…The challenge in this case is to accurately identify the bone which is where most MR-based techniques are prone to error. A number of methods which used paired 2D Dixon and CT [ 241 , 242 ], 3D ZTE and CT [ 40 ], 3D T 1 and CT [ 243 ] images or an additional deep learning-based segmentation step to segment the air from the bowl areas [ 244 ], resulted in comparable biases of approximately 5% in the pelvic bone regions. Moreover, it was recently shown that if the uncertainty in the prediction is also taken into account, implants could be more easily identified [ 245 , 246 ].…”

Recent advances in combined Positron Emission Tomography and Magnetic Resonance Imaging

“…The main challenge of these approaches was that neglecting air pockets in the pelvis AC maps could lead to inaccuracies in reconstructed PET images. However, these innovative methods based on CNNs even showed the capability of predicting air pockets and include them in the AC maps [103].…”