| dc.description.abstract | Computed tomography (CT) is an extremely important tool to realize medical imaging exams
and obtain an accurate medical diagnosis of internal structures of a patient. Compared to other
traditional imaging exams, CT is more efficient, where a digital geometry processing is used
to generate a 3D image of an internal structure of an object, or patient, from a series of 2D
images obtained during various rotations of the CT scan around the scanned object. Also taking
in consideration traditional imaging exams such as X-ray, MRI or ultrasound, for example,
the CT technique uses higher radiation doses than such exams, providing high quality images.
However, exposing patients to constant high doses of radiation is a negative factor of this type
of examination and, because of that, the medical community has focused on decreasing the radiation
dose of CT scans following the ALARA principle (As Low As Reasonably Achievable),
which aims to minimize the radiation dose to a certain level while maintaining an acceptable obtained
image and medical diagnosis. Nevertheless, the images acquired in low-dose CT (LDCT)
scans are degraded by undesirable artifacts, known as noise, which affects negatively the image
quality. That is, the pixels values of the obtained image are corrupted. Given the presented
scenario, in this study is proposed a method based on mathematical morphology operators and
Block-Matching 3D (BM3D) filtering to reduce noise and preserve details in low-dose computed
tomography medical images. The method is divided into two main stages, named as:
image segmentation and noise filtering. The first stage is responsible for separating the input
image in two main regions named as foreground and background which will pass through a
noise filtering process by morphological operators and BM3D technique in the second stage.
The proposed method was tested on 991 dental images and 460 lung images from low-dose
CT scans. The experimental results obtained by the proposed method were compared with the
results of several state-of-the-art filters and validated using the PSNR (Peak Signal-to-Noise
Ratio), SSIM (Structural Similarity), MSE (Mean-Squared Error) and EPI (Edge Preservation
Index) quantitative metrics. The average results obtained by the proposed method for dental
images demonstrate superior performance compared to the evaluated filters, with average
quantitative PSNR values of 28.78, 0.82 in SSIM metric, MSE of 177.06 and 91% in edge
preservation metric. For the images of the lung, the average PSNR value obtained was 19.77,
0.74 for the SSIM metric, 181.98 on MSE and average EPI of 0.82. The results, both visual and
quantitative, were satisfactory, proving that the proposed method can reduce noise and preserve
details efficiently. | |
