@article{oai:repo.qst.go.jp:00079593,
 author = {Takahashi, Wataru and Oshikawa, Shota and Mori, Shinichiro and Mori, Shinichiro},
 issue = {1109},
 journal = {British Journal of Radiology},
 month = {Mar},
 note = {Purpose
Large amounts of subject data are required for training data sets in deep learning (DL) for medical imaging. Collection of these data is crucial, but some collection strategies are hampered by the heterogeneity of subject data. For real-time markerless tumour tracking, we propose a different approach which uses patient-specific DL using a personalized data generation strategy.
Methods
We developed lung tumour tracking for radiotherapy using a convolutional neural network trained for each patient’s lesion, avoiding the need for a large data set by using multiple digitally reconstructed radiographs (DRRs) generated from each patient’s treatment planning 4D-CT. We trained tumour-bone differentiation using huge training DRRs generated with different projection geometries to simulate tumour motion. We solved the problem of using DRRs for training and X-ray images for tracking by using the training DRRs with random contrast transformation and random noise addition. These solutions also double as data augmentation.
Results
We defined adequate target coverage (TC) as the % frames satisfying < 1 mm tracking accuracy of the isocentre. In a simulation study, we achieved 100% TC in 3 cm spherical and 1.5 × 2.25 × 3 cm ovoid masses. In a phantom study, we achieved 100% and 94.7% TC in 3- and 2-cm spherical masses, respectively. This required 32.5 ms/frame (30.8 fps) real-time processing.
Conclusions
We proposed and validated a real-time markerless tumour tracking framework based on patient-specific DL with a personalized data generation strategy.
Advances in Knowledge
Using DL with personalized data generation is an efficient method for real-time tumour tracking.},
 title = {Real-time markerless tumour tracking with patient-specific deep learning using a personalized data generation strategy: Proof of concept by phantom study},
 volume = {93},
 year = {2020}
}