Click here to

Session: Biologically- and Functionally-Guided Radiation Therapy [Return to Session]

Ventilation Derived From Clinical 4D-CBCT Using a Deep Learning-Based Model: First Comparison with Technegas SPECT Ventilation

Z Liu*, Y Tian, J Dai, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical CollegeBeijingCN


TU-C-TRACK 4-6 (Tuesday, 7/27/2021) 1:00 PM - 2:00 PM [Eastern Time (GMT-4)]

Purpose: Ventilation derived from 4D cone-beam computed tomography (CBCT) may predict specific radiation sensitivity, but the current algorithms for measuring ventilation highly depend on deformable image registration (DIR). With single-photon emission-computed tomography ventilation (SPECT-V) acting as a gold standard, this study proposes new deep learning (DL) model-independent of DIR for deriving ventilation from 4D-CBCT and investigates whether the accuracy could be improved in comparison with DIR dependent methods.

Methods: The study consists of 4D-CBCT and 99mTc-Technegas SPECT/CT scans of 28 esophagus or lung cancer patients. For each patient, the scans were acquired before the first radiotherapy treatment and rigidly registered. Using these data, ventilation images from 4D-CBCT (CBCTVI) were derived by using a deep-learning-based model with two forms of the dataset [(a) ten phases and (b) two phases of peak-exhalation and peak-inhalation] as input. A Sevenfold cross-validation procedure was used to evaluate the performance. For comparative evaluation, density-change- and Jacobian-based (HU and JAC) methodologies are also used to calculate CBCTVIs for each patient. The SPECT-V and derived CBCTVIs were segmented into high, medium, and low functional lung (HFL, MFL, and LFL) regions. The spatial overlap of corresponding HFL, MFL, and LFL for SPECT-V against CBCTVIs was evaluated by using the dice similarity coefficient (DSC) and analyzed with a one-factor ANONA model among different methods.

Results: The averaged DSC values were 0.34, 0.34, and 0.59/0.58 for the CTVI(HU), CTVI(JAC), and CTVI(DL(a)/DL(b)), respectively. These results showed the DL method yielded the highest similarity with SPECT-V with the prominently significant difference.

Conclusion: This study proposed a DL model for deriving CBCTVI and performed a validation against SPECT-V. The results demonstrated that DL method increased the accuracy in comparison to HU and JAC methods, and can be a stepping stone to extract dynamic changes in the respiration patterns of patients during radiotherapy.

Funding Support, Disclosures, and Conflict of Interest: The authors sincerely thank Dr. Li Mei and her team in nuclear medicine department of Peking TongRen Hospital for their great help in acquiring SPECT image for patients. This work is supported by the National Natural Science Foundation of China [11905295, 81502649, 11875320]; the Beijing Hope Run Special Fund of Cancer Foundation of China (LC2018B07); the Natural Science Foundation of Beijing Municipal (7204295).



    Not Applicable / None Entered.


    Not Applicable / None Entered.

    Contact Email