Assessment of mammography screening using clinical and virtual data
Sammanfattning: Aim: Virtual clinical trials (VCT) in medical imaging can be used to predict the outcome of clinical trials, by simulating anatomy, imaging methods and image interpretation. VCTs could reduce the cost and duration of clinical trials and their dependence on available patients. This thesis is motivated by the limitations in the field of breast cancer screening. The aim of the project is to expand a specific VCT-software, OpenVCT, to support future VCTs and complement the results from the Malmö Breast Tomosynthesis Screening Trial (MBTST). The virtual patients and tumours currently simulated in OpenVCT need to be further improved and assessed, especially when it comes to their progression over time. The foremost objective of this study was to initiate a simulation of breast tumour growth and to implement growing lesions into virtual breast phantoms and thus allow for the simulation of multiple examinations over time. More specifically, a tumour growth model that is based on the characteristics of the Malmö screening population. A secondary objective was to evaluate the tumour growth model in a virtual clinical environment by estimating tumour volume doubling times (TVDT) from virtual mammograms and comparing with the theoretical values of the model. Material and Methods: The tumour growth model was based on previous studies of TVDT in breast cancer patients in Malmö, Sweden. A gamma probability distribution was fitted to the existing data and a program was developed that randomly samples a TVDT for a virtual breast cancer patient. Based on this, 30 virtual breasts were simulated using simplified tumour characteristics such as spherical lesions and exponential growth functions. The patient age and TVDT was specific for the Malmö population. Two mammograms, at different time points, were simulated per patient in order to display the tumour growth. TVDTs were then estimated from the mammograms by having a radiologist measure the lesion size. The estimated TVDTs were compared with their corresponding nominal values. Results: The initial tumour growth model was successfully implemented, and virtual mammograms were simulated for the 30 patients, depicting tumour growth. The model was estimated to have a mean TVDT of 297 ± 169 days, whereas the sampled virtual patient cohort had 322 ± 217 days. The estimated TVDT from the simulated mammograms had a mean of 306 ± 209 days. The data sets were found to originate from the same distribution as no significant difference was found between them (p>0.54). However, it was observed that the median difference between the sampled and estimated TVDTs was 12 days (IQR = 20.75) and significantly larger than zero (p<0.01). The mean difference between the sampled and estimated TVDTs was 16 ± 57 days. Median differences between the other data sets showed no significant distinction from zero (p>0.64). Conclusion: The initial tumour growth model displayed high accuracy and reliability when used in a possible virtual clinical trial and showed potential for further development.
HÄR KAN DU HÄMTA UPPSATSEN I FULLTEXT. (följ länken till nästa sida)