Evaluation of CT-Derived Lung Perfusion with Deep Learning in Radiotherapy Patients
G Ren1*, B Li2,1, Y Lu2, R Mao2, H Ge2, F Kong3, W Ho4, J Cai1 (1) Department of Health Technology and Informatics,The Hong Kong Polytechnic University, Hong Kong, CN (2) Radiation Oncology Department Of Henan Cancer Hospital,CN (3) Case Western Reserve University- School of Medicine, Cleveland, OH (4) Department of Nuclear Medicine, Queen Mary Hospital,Hong Kong, CN
Presentations
TU-C-TRACK 4-1 (Tuesday, 7/27/2021) 1:00 PM - 2:00 PM [Eastern Time (GMT-4)]
Purpose: Deep learning-based CT perfusion mapping (DL-CTPM) has been proposed to provide regional function information for functional lung avoidance radiotherapy (FLART). This study aims to evaluate the DL-CTPM synthesized perfusion images in radiotherapy patients for FLART.
Methods: SPECT/CT perfusion scans of 19 lung cancer patients were retrospectively collected from Henan Anti-cancer Hospital. All patients underwent radiation therapy afterward. The left/right lungs were separated to augment the dataset to 38 (20 for transfer training; 18 for testing). These images were processed by a deep learning-based framework to extract features from 3D-CT images and synthesize CT-derived perfusion (CTDP). Transfer learning strategies of fine-tuning and fixed-encoder were used to reuse information learned from a previously built dataset of different types of lung diseases. The Spearman’s correlation coefficient (R) and structural similarity index measure (SSIM) between the CTDP and the corresponding SPECT were computed to assess the statistical and perceptual image similarities, respectively. To assess the function-wise concordance, the Dice similarity coefficient (DSC) was computed to determine the similarity of the low/high functional lung volumes. The dosimetry evaluation is undergoing.
Results: The fine-tune transfer learning model has the best performance in terms of the voxel-wise agreement. The evaluation shows a high voxel value correlation (0.8111±0.1090) and high structural similarity (0.8094±0.0517) between the CTDP images and SPECT images. The evaluation of the function-wise concordance obtains an average DSC value of 0.7869±0.0634 for high-functional lungs, ranging from 0.6460 to 0.8984, and 0.7806±0.0727 for low-functional lungs, ranging from 0.6743 to 0.8902. Of the testing cases, 94% have high functional lung larger than DSC > 0.7, and 83% for low functional lung.
Conclusion: The CT-derived lung perfusion using the DL-CTPM method showed a high agreement with the SPECT perfusion for lung cancer patients. This method holds great promise to provide lung function images for image-guided FLART.
Funding Support, Disclosures, and Conflict of Interest: This work is supported by the Health and Medical Research Fund (HMRF 07183266); the General Research Fund (GRF 151022/19M).
Handouts
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email