MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Computed Tomography Based Lung Perfusion Mapping Using Attention Residual Neural Network for Functional Avoidance Radiation Therapy

G Ren1*, W Ho2, H Xiao1, A Cheung1, J Qin1, J Cai1, (1) The Hong Kong Polytechnic University, Hong Kong, ,CN, (2) Queen Mary Hospital, Hong Kong, ,CN

Presentations

(Monday, 7/13/2020) 1:00 PM - 3:00 PM [Eastern Time (GMT-4)]

Room: Track 1

Purpose:
Lung functional avoidance radiation therapy is an emerging technique that holds the promises to reduce radiation induced lung injury. The purpose of this study is to develop a lung perfusion mapping method that estimates CT based lung perfusion image using an attention residual neural network (ARNN) for functional avoidance radiation therapy.

Methods:
Tc-99m MAA SPECT/CT scans of 41 patients with lung diseases were retrospectively analyzed. To clean the data and detect the useful features, a preprocessing pipeline was built, including pseudo-functional region exclusion, CT equalization, histogram-based SPECT regularization, and target discretization. Next, a 3D ARNN model was built to learn and exploit the underlying functional information in the CT image and generate 3D functional perfusion image. The attention module was used to focus on the synthetic of the low function regions. The model was trained using randomly selected datasets of 30 patients for 1000 epochs and tested using the remaining 11 scans. Binary cross entropy with auxiliary correlation was used as the loss function. The accuracy of the predicted perfusion was qualitatively and quantitively evaluated, using intensity plot, voxel-wise correlation, and average perfusion in each lobe.

Results:
Preliminary results show that the preprocessing pipeline can improve the prediction correlation on the testing dataset from 0.53 to 0.73. The ARNN model can save computational time (2.5h) as compared with U-net (11.3h). The average voxel-wise correlation with SPECT perfusion is 0.73 (S.D.= 0.05). The correlation of the average perfusion in each lobe is 0.85. Intensity plot and scatter plot comparison demonstrate good agreement between the SPECT perfusion and predicted perfusion.

Conclusion:
We developed a lung functional mapping method for deriving lung perfusion images from 3D CT images. This method holds great promise to provide lung function images for image guided functional lung avoidance radiation therapy.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email