Search the portal

Please enter a term

Full record

« Back to home page
TitleQuantitative Analysis of Radiation-Associated Parenchymal Lung Change
AuthorChandy, Edward T. J.
AbstractRadiation-induced lung damage (RILD) is a common consequence of thoracic radiotherapy (RT). We present here a novel classification of the parenchymal features of RILD. We developed a deep learning algorithm (DLA) to automate the delineation of 5 classes of parenchymal texture of increasing density. 200 scans were used to train and validate the network and the remaining 30 scans were used as a hold-out test set. The DLA automatically labelled the data with Dice Scores of 0.98, 0.43, 0.26, 0.47 and 0.92 for the 5 respective classes. Qualitative evaluation showed that the automated labels were acceptable in over 80% of cases for all tissue classes, and achieved similar ratings to the manual labels. Lung registration was performed and the effect of radiation dose on each tissue class and correlation with respiratory outcomes was assessed. The change in volume of each tissue class over time generated by manual and automated segmentation was calculated. The 5 parenchymal classes showed distinct temporal patterns We quantified the volumetric change in textures after radiotherapy and correlate these with radiotherapy dose and respiratory outcomes. The effect of local dose on tissue class revealed a strong dose-dependent relationship We have developed a novel classification of parenchymal changes associated with RILD that show a convincing dose relationship. The tissue classes are related to both global and local dose metrics, and have a distinct evolution over time. Although less strong, there is a relationship between the radiological texture changes we can measure and respiratory outcomes, particularly the MRC score which directly represents a patient’s functional status. We have demonstrated the potential of using our approach to analyse and understand the morphological and functional evolution of RILD in greater detail than previously possible.
Date2022-05-28
TypeThesis; Doctoral
PublisherUCL (University College London)
Languageeng
Source Doctoral thesis, UCL (University College London).
Formattext
Identifier
Identifier
Rightsopen