Being one of the largest organs in the human body, the liver is often the object of diagnostic and therapeutic procedures, which require accurate positioning of instruments with respect to the area of interest. This can be challenging due to respiratory motion. Training a motion model with X-ray images is not always possible and results in more radiation exposure for both the patient and the doctor. On the other hand, ultrasound can be used; however, this modality lacks spatial resolution.
In this work, we develop a motion model transformation workflow consisting of two steps. Firstly, the motion model is developed using ultrasound (US) modality. Secondly, it is transformed into a computed tomography (CT) modality space, which allows for better spatial resolution.
To achieve this, novel vessel segmentation algorithms for US and CT modalities were developed, along with two registration approaches: 2D US to 3D CT and 3D US to 3D CT. The initial US model is trained with data obtained from an electromagnetic (EM) sensor placed on the surface of the phantom. To evaluate model performance in the CT space, CBCT volumes were obtained in 10 different positions of the breathing cycle. Tumor coordinates in the CT space was extracted from these volumes and compared to those predicted by the motion model.
The possibility of applying a motion model trained in a different EM setup was also explored. The mean absolute error varied between 0.138 and 1.092 mm for different pipelines and setups. The obtained results indicate that the motion model transferred from the US to the CT modality can be used in interventions requiring high positional precision of instruments.