Anatomy tracking by fluoroscopic image processing combined with 3D finite element motion model

In 2020, nearly 3 million deaths were caused by lung or liver cancer. These cancers are particularly difficult to treat with X-ray treatment, also called radiotherapy. First because the tumor is positioned near vital organs, and secondly because the position of the tumor and its shape may vary during treatment due to chest movements, i.e. breathing. This type of tumor is called a mobile tumor. To cope with this, safety margins, that is a volume irradiated greater than the tumor volume, are considered. This ensure that the tumor will always be reached. However, traditional radiotherapy does not make it possible to target a precise 3D volume, the X-rays will irradiate everything in their path, before and after the tumor. Through proton therapy, volume can be targeted more precisely. Nevertheless, this type of beam involves more energy and, in the case of a mobile tumor, can lead to significant injuries if the radiating beam is not perfectly synchronized with the movement of the tumor. This results in futile irradiation of the surrounding tissue, in this cases lungs or liver, which can be quite dangerous. Thus, it appears to be crucial to have precise information concerning the exact position of the tumor in real-time during the treatment. In this work, we tested a new method that could potentially meet this need. This method combines two main parts: the determination of a metric from fluoroscopic images which gives information on the patient’s respiratory state, and the construction of a patient-specific finite element model representing a tumor in a liver for several positions. The idea is as follows: based on the metric calculated on the images available in real time (fluoroscopy), the closest corresponding 3D model is chosen to represent the current respiratory state of the patient. This latter gives precise information on the position and shape of the tumor. The obtained results are encouraging since they show the possibility of reducing the tumor margins, which are currently necessary to be sure of irradiating the mobile tumor whatever the respiratory state, and thus save more healthy tissues. To verify the accuracy of the method we used DICE coefficients indicating a level of region similarity between the ground truth tumor volume, and the simulated tumor volume chosen based on metric. The average DICE values calculated for the tumor region were: 0.8176 ± 0.0547, 0.8755 ± 0.0333 and 0.7906 ± 0.0446 respectively for patients 1, 2 and 3, over about 200 iterations for each of them.