(Thursday, 7/18/2019) 7:30 AM - 9:30 AM
Room: Stars at Night Ballroom 2-3
Purpose: The purpose of this study is to develop a deep learning method for producing pulmonary ventilation images from four-dimensional computed tomography (4DCT) and perform the first comparison of produced computed tomography ventilation images (CTVI) with clinical single-photon emission computed tomography ventilation (SPECT-V).
Methods: Fifty patients with esophagus or lung cancer who received 4DCT scans paired with 99mTc-Technegas SPECT/CT and underwent thoracic radiotherapy were enrolled in this study. Using these data, a deep learning model was trained to produce CTVI. Ten time phases’ 4DCT and SPECT/CT were first rigidly registered using MIMvista, and converted to three-dimensional matrix with a program based on a developed image toolkit with MATLAB, and then transferred to the model for correlating 4DCT features and SPECT-V. Ten-fold cross validation was used to estimate the performance of the model. In this study, density change-based and Jacobian-based methods are also used to compute a 4DCT-based ventilation map for the purpose of comparison. The spatial overlap of corresponding percentile ventilation distributions (high, median and low) are evaluated for each 4DCT ventilation map with reference to the clinical standard SPECT-V by using the dice similarity coefficient (DSC). The averaged DSC of different percentiles was calculated.
Results: The averaged DSC of different percentiles for one-fold results with density change-based, Jacobian-based and deep learning-based methods are 0.38, 0.33 and 0.76, respectively.
Conclusion: The study developed a new deep learning method for producing 4DCT ventilation image and performed a validation of CTVI with SPECT-V. The preliminary results demonstrate that deep learning based method can generate ventilation image from 4DCT with improved accuracy in comparison with density change-based and Jacobian-based methods. The produced ventilation maps can be useful for lung functional avoidance radiotherapy.
Funding Support, Disclosures, and Conflict of Interest: This work is supported by the Beijing Hope Run Special Fund of Cancer Foundation of China (LC2018B07); the National Natural Science Foundation of China [81502649,11875320].
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
