Respiratory lung motion using an artificial neural network

One of the possibilities to enhance the accuracy of lung radiotherapy is to improve the understanding of the individual lung motion of each patient. Indeed, using this knowledge, it becomes possible to follow the evolution of the clinical target volume defined by a set of points according to the lung breathing phase. This paper presents an innovative method to simulate the positions of points in a person’s lungs for each breathing phase. Our method, based on an artificial neural network (ANN), allowed us to learn the lung motion of five different patients and then to simulate it accurately for three other patients using only beginning and end points. The training set for our ANN consisted of more than 1,100 points spread over ten breathing phases from the five patients on a specific area of the lungs. The points were defined by a medical expert. The first results are very promising: we obtain an average accuracy of 1.5 mm while the spatial resolution is 1  ×  1  ×  2.5 mm3. The accuracy of the method will be improved even more with additional data and providing complete lung coverage.