A Quantification Method for Comparing Low-Contrast Edge Definition on Clinical Head CT
Abstract
Purpose
To propose a clinically relevant, quantitative comparison of low-contrast edge definition on routine CT Head examinations.
Methods
Routine head CT exams were collected from an investigational photon-counting CT and an energy-integrating CT. The exams were processed with baseline reconstruction settings typically used in clinical practice. Three radiologists were tasked to create six ROIs over specific boundaries between anatomies in the Alberta Stroke Program Early CT Score over each scan volume. A Python-based tool was developed to facilitate ROI drawing by using a paintbrush-like ROI drawer. The measurement, termed “Blur Area”, was defined as the number of pixels within the ROI with a value within the 25th and 75th percentile range. This metric was calculated on the baseline reconstruction and additional reconstructions to compare the effect of different parameters on the low-contrast edges defined by the ROIs.
Results
ROIs were collected and the blur area was computed for the baseline image and three different reconstruction algorithms. We saw an overall decrease in the blur area across all reconstructions in comparison to the baseline. Blur area percentage change showed high variance at small ROI sizes, stabilizing at larger ROI sizes indicating that ROI maximization and sampling should be maximized for better precision of the measurement.
Conclusion
This work demonstrates a method for quantifying the impact of algorithms on the image quality of clinical images to serve their clinical task. In the framework presented, a radiologist manually asserted the areas analyzed in their clinical assessment to establish the most clinically relevant areas of the image. This method is more relevant than phantom measurements as it operates on clinical images which are inherently subject to the relevant conditions that may affect low-contrast edge measurements such as noise and contrast.