Correlation between Surface and Tumour Motion in Lung Cancer - including Deep Learning Perspectives

Sammanfattning: Purpose: The purpose of this master thesis was to retrospectively investigate correlation between surface and tumour motion in lung cancer patients, alongside deep learning applications of the results. Additional correlations such as age, tumour volume and anatomical placement of the tumour were also investigated. Materials and Methods: 48 lung cancer patients treated with Stereotactic Body Radiation Therapy (SBRT) were included in this study. Delineation of the tumours was made on 4-Dimensional Computed Tomography (4DCT) images where each tumour was delineated in each of the eight respiration phases. Tumour volume and centre of mass coordinates in Left-Right (LR), Anterior-Posterior (AP) and Superior-Inferior (SI) directions were retrieved in all respiration phases. The total translational shift from maximum exhale phase was computed. Surface motion data was acquired from a surface imaging system which was recorded during 4DCT simulation of the patient. The Spearman Correlation Coefficient (SCC) between tumour and surface motion was calculated. An additional Spearman correlation was calculated between the SCC and patient features (age, tumour volume, smallest distance to thoracic wall, distance to thoracic spine and to chest surface, lung volume, and use of abdominal compression belt). Wilcoxon's rank sum test was performed to determine statistical significance between the groups with and without abdominal compression belt. An artificial neural network (ANN) model was created to investigate the possibility to predict tumour motion given the surface motion as an input to the model. The model was built like a feed-forward ANN with two hidden layers using Rectified Linear Unit (ReLU) activation functions. Training, validation, and test data set split were 36/45, 6/45 and 3/45 respectively. Results: Strong correlation (0.70