Premium
SU‐E‐J‐54: A Framework of Physical‐Law‐Based Respiratory‐Perturbation Modeling for Motion Prediction
Author(s) -
Li G,
Gaebler C,
Huang H,
Yuan A,
Wei J
Publication year - 2015
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4924141
Subject(s) - torso , breathing , trajectory , motion (physics) , perturbation (astronomy) , thorax (insect anatomy) , computer vision , computer science , physics , anatomy , medicine , quantum mechanics , astronomy
Purpose A physics‐law‐based respiratory motion model should be more resilient to breathing irregularities. We assess the accuracy of a physical respiratory motion‐perturbation (RMP) model, which is derived with physical relationships, patient‐specific anatomy, and measureable breathing parameters. Methods The analytical RMP model was developed under the assumption that chest raise (Δthorax) contributes primarily to the anterior‐posterior motion of the lung while belly motion (Δabdomen) caused by diaphragmatic motion contributes to superior‐inferior motion. All model parameters were determined from patient‐specific anatomy. This RMP model aims to use the base motion trajectory from simulation 4DCT and motion perturbations due to breathing irregularities at treatment from optical surface imaging (OSI) to calculate the tidal volume (TV=Δtorso) and breathing pattern (BPv=Δthorax/Δtorso). To test the prediction accuracy, two sets of 4DCT from eleven patients were used to predict lung motion from one set to the other. An in‐house ITK‐based C/C ++ program was developed and used to automatically identify 20 bifurcation points per patient, calculate their motion trajectories, and sets their correspondence between the two 4DCT with visual verification. The base motion from one 4DCT and perturbations (ΔTV and ΔBP) from the other 4DCT (mimicking OSI) were used for prediction and the accuracy was assessed by comparing the predicted with the actual motion trajectory: A comparison was made using an established 5D model that was trained with one 4DCT to predict the other. The motion difference between the two 4DCT was used as a control. Results The RMP model prediction provides the prediction accuracy of −0.1±1.9mm, improved from the control of 0±3mm. The RMP produces similar results to the 5D model, but does not require model training. The motion variations range is 0–15mm. This work is in part supported by NIH (U54CA137788 and U54CA132378).