A simulation framework for virtual clinical trials in chest radiography
Chest radiography (CXR) is the core imaging examination for the chest and one of the most frequently performed diagnostic procedures in clinical practice. CXR is subject to many technical challenges because of the many organs and structures that need to be captured by the image. Additionally, a wide range of clinical questions are addressed with this imaging technique.
Achieving the image quality required for the physician to answer the clinical questions, requires setting up optimal acquisition protocols. In principle, acquisition protocols are optimized to account for both the dose delivered to the patient and image quality. However, imaging protocols currently used were established for screen/film technology, while now most of x-ray departments use the more advanced digital systems. Thus, the revision and optimization of these protocols for the newer CXR technology is fundamental.
This PhD thesis describes the creation of a simulation platform to be used in optimization of image quality and dose in chest radiography. The framework allows to generate computer-simulated radiographic images of realistic anatomical models of patients and pathologies. To generate the images, different simulation techniques were used, such as Monte Carlo simulations and ray tracing algorithms. Additionally, the noise and sharpness levels measured in real x-ray detectors were added to the simulated images to further increase the realism of the images
The work also includes the development of a library of anthropomorphic computational phantoms. The phantoms represent a range of body types, like male, female and several Body Mass Indexes. Additionally, a range of pathologies and devices commonly found in CXR were modelled and added within the phantoms to simulate different clinical tasks. The validation of the framework was performed by comparing the simulated images to real images acquired in an x-ray system using several image quality metrics. On the other hand, the realism of the phantom models was validated by experienced radiologists. The simulation platform and the anatomical models were used to generate a set of images representing radiographies acquired under different acquisition settings, i.e., dose level, tube voltage and antiscatter device. Additionally, the dose absorbed in the phantom organs were also estimated for all settings.
To study the influence of the different acquisition settings in diagnostic performance, four radiologists read the images under normal clinical conditions. The readers reported on the localization of pathologies detected, together with a confidence score about the presence of the pathologies.
Several practical conclusions were derived from the results of the reading, some of which are in contrast with previous findings:
- For images with the antiscatter device in place, a tube voltage of 120 kVp showed the lower organ doses with no significant decrease in diagnostic performance compared to lower tube voltages, justifying the choice made in current clinical practice regarding tube voltage.
- For images without the antiscatter device, no significant difference in diagnostic performance was found between 80 and 100 kVp, however organ doses were in average 10% lower at 100 kVp compared to 80 kVp.
- A reduction of dose by 50% can be achieved from the standard working dose level, without affecting the detectability of the clinical tasks modelled (nodules, catheters, rib fractures, pneumothorax and pleural effusion).
- In general, no significant difference was found in images with and without the antiscatter device. Further analysis of this effect is necessary. A comparison of different image processing techniques in terms of how they can restore the contrast for images without antiscatter device is also relevant.
Although developed for chest radiography applications, the phantoms created are not limited to this modality and can also be employed in other imaging techniques and/or for dosimetry studies. There are no other phantom libraries that depict this range of clinical tasks for chest radiography. The proposed simulation framework should allow the exploration of a wide range of system configurations and rule out suboptimal scenarios. This should also ease the optimization of the newest x-ray detectors being brought on the market these days.
Hilde Bosmans (KU Leuven)
SCK CEN mentors:
Lara Struelens (SCK CEN)
Filip Vanhavere (SCK CEN)
Click here for a list of obtained PhD degrees.