Denoising of CT and MRI Images Using Decomposition-Based Curvelet Thresholding and Classical Filtering Techniques

New publication by Mahmoud Nasr, Krzysztof Brzostowski, Rafał Obuchowicz and Adam Piórkowski

Noise is the invisible enemy of medical imaging. In a new paper, Mahmoud Nasr from Sano, together with Krzysztof Brzostowski, Rafał Obuchowicz and Adam Piórkowski, proposes a powerful framework to clean CT and MRI scans while preserving diagnostically critical details.  

Their method combines multiscale decomposition techniques with curvelet-based denoising and classical spatial filters, allowing it to separate noise from real anatomical structures in a highly targeted way. The framework was tested on controlled phantom data, clinical CT images reconstructed with different kernels, and MRI scans acquired with accelerated protocols, reflecting real-world imaging challenges.

The hybrid variants using MEMD–Curvelet and VMD–Curvelet consistently achieved high structural similarity and signal quality, outperforming standard filters even for sharp-kernel CT images. In MRI, MEMD–Curvelet and BEMD–Curvelet reduced perceptual distortion, improving image naturalness compared to popular Gaussian and median filters.

Importantly, the authors also benchmarked their approach against deep learning baselines, showing that it can match high-fidelity methods while remaining computationally efficient—an important factor in clinical workflows. This versatility across modalities and acquisition settings makes the framework a promising candidate for integration into therapeutic and diagnostic pipelines where high-quality denoising is essential under constrained imaging conditions.

See how smarter denoising can sharpen medical diagnosis  — read the full paper: https://www.mdpi.com/2076-3417/16/3/1335?utm_source=researchgate.net&utm_medium=article