Conditional-Based Transformer Network With Learnable Queries for 4D Deformation Forecasting and Tracking

Current Research Status of Respiratory Motion for Thorax and Abdominal Treatment: A Systematic Review

“…In 2023, L. V. R. et al [ 113 ] proposed an attention-based time prediction network that treats features extracted from input images as markers of prediction tasks. The network architecture consists of three modules: feature coding and decoding, conditional transformer network, and parallel prior-based latent modeling.…”

Prediction of the position of external markers using a recurrent neural network trained with unbiased online recurrent optimization for safe lung cancer radiotherapy

“…Attention mechanisms can be applied to time-series prediction and help networks focus on time intervals that significantly impact forecasting accuracy by increasing corresponding weights. Despite initial works showing the superiority of attention-based architectures on RNNs in respiratory motion forecasting [16,39,44] and the general high performance of transformers on many tasks due to parallel processing and the absence of a vanishing gradient, transformers "are impractical for training or inference in resourceconstrained environments due to their computational and memory requirements" [46]. Indeed, their complexity quadratically grows with the input window length, which hinders their ability to learn long-range dependencies [23,5].…”

“…Indeed, their complexity quadratically grows with the input window length, which hinders their ability to learn long-range dependencies [23,5]. For instance, it was observed in [39] that transformers predicting breathing signal representations from chest cine-MR imaging led to an inference time approximately three times higher than convolutional GRUs. These shortfalls motivate further research on RNNs for respiratory motion forecasting.…”

“…Using Eq. 4, we can write: ∂Fst ∂x = ∂Φ ∂z ∂zn ∂x (39) The left and right factors can be directly calculated using the definition of Φ in Eq. 2:…”