Monitoring of lung cancer patients during radiotherapy using combined texture and level set analysis of CBCT images
MetadataShow full item record
In the UK, radiotherapy (RT) contributes to a large amount of lung cancer treatment while its imaging information is limited to computed tomography (CT) and cone beam CT (CBCT) images. The oncologists deﬁnes the gross tumour volume (GTV) manually on the planning-CT images before any treatment starts. Manual contouring suffers from many disadvantages and the bad quality of CBCT images makes it very challenging for the clinicians to observe tumour behaviour in the time of treatment. CBCT is the only kind of image available throughout the whole course of RT which is used in the mechanical procedure of adjusting patient position before starting each session of treatment and is not generally used by clinicians for monitoring the tumour. The goal of this thesis is to develop a tumour detection model of non-small cell lung tumours on CBCT images in the course of treatment. By developing this process clinicians will be greatly aided in their role, helping them to detect lung tumours to allow better diagnosis and improving patient treatment outcome. Therefore a new segmentation approach is proposed as combined texture analysis and level set model. It has the potential capability to track the variation of the tumour shape over time of treatment solely using CBCT images, and evaluate the accountability of RT for different patients. The texture analysis, second-order statistics obtained from gray level co-occurrence matrices (GLCM), highlight the tumour boundary and help Chan-Vese and Li level set models convergence in the segmentation process. Further on a new parallel level sets model is proposed by combining Chan-Vese and Li models in the concept of vector-valued image level set. This new approach overcomes the difﬁculties in the parameter settings of current models by giving more freedom of choice in tuning parameters as well as selecting level set models. All proposed models were evaluated on the dataset of ﬁfty different patients suffered from non-small cell lung cancer. For the validation procedure, qualitative analysis was carried out by an oncologist as there is no ground truth in each CBCT image during RT. The decision of the oncologist based on patient history has proven the results of this work. For quantitative analysis, the Dice coefﬁcient is used to evaluate the tumour segmentation results on CBCT compared to GTV on CT images prior to treatment to evaluate the amount of changes especially after one third of RT on CBCT #10. Additionally, the proposed segmentation models had the accuracy of almost 90% to the GTV delineated by the oncologist for the only one patient in the dataset having GTV on CBCT images which proved the ability of these models for further analysis during the absence of GTV on CBCTs. For improving this research and helping the clinicians at most, the proposed segmentation model can be used as a notiﬁcation model to assist clinicians for a better understanding of the tumour during RT and subsequent use in ofﬂine adaptive radiotherapy (ART).