Fuzzy Clustering of CT scans to assess image noise

Computed Tomography (CT) scans are the largest source of radiation exposure in diagnostic imaging. Studies show that CT scan radiation doses are relatively high in the United States compared to Japan, Israel, and Switzerland even af- ter adjusting for population effects. The increased use of CT scans in the United States has caused great concern in the public health community. Numerous stud- ies have proposed methods for noise estimation in CT scans to help reduce dose. However, there are few efforts that have leveraged fuzzy logic, which is mod- elled after human reasoning and perception to assess the relationship between CT image quality and dose. We show a new application of fuzzy c-means (FCM) clustering to group images. Image characteristics such as noise, contrast, reso- lution, etc. are identified by the algorithm and used to cluster the images. Our dataset included CT scans from the abdominal and pelvic region of adults, re- sulting in a total of 4,951 images, which vary by radiation dose level, image slice thickness, reconstruction filter and machine specifications. One of the disadvan- tages of FCM is its sensitivity to image noise and artifacts. However, we exploit this disadvantage for our purpose to obtain clusters that would correlate well with image noise as calculated by noise algorithms. Our results showed that FCM was successful at clustering images based on image noise if the difference in radiation dose levels was significant. This implies that intelligent image classi- fication algorithms can detect differences in image noise. Results also indicated that the noise from high dose scans were similar to that from low dose scans, thereby indicating that there is potential to lower doses. It was also evident that the choice of the convolution kernel and slice thickness had a greater effect on image noise than radiation dose. We would like to draw attention of the fuzzy logic community to this area of work that will particularly benefit from percep- tion rules in fuzzy logic. While fuzzy logic has been used in image analyses and medical decision-making, it has not been applied to this objective of lowering radiation dose. We aim to show some of the work performed to achieve this goal, aligning with national initiatives to reduce radiation dose that Americans receive, and identified areas that need further exploration.