Effect of Noise Adaptive Wavelet Filter on Diagnostic Performance in Stroke Perfusion
| Autor: | Gökçe Kaan Ataç, Tolga Inal, Ziya Telatar |
|---|---|
| Rok vydání: | 2021 |
| Předmět: |
Computer science
business.industry ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION Health Informatics Pattern recognition medicine.disease Adaptive wavelet 030218 nuclear medicine & medical imaging 03 medical and health sciences Noise 0302 clinical medicine Filter (video) medicine Radiology Nuclear Medicine and imaging Artificial intelligence business Perfusion Stroke 030217 neurology & neurosurgery |
| Zdroj: | Journal of Medical Imaging and Health Informatics. 11:720-729 |
| ISSN: | 2156-7018 |
| Popis: | Background: Computed tomography perfusion (CTP) images include more noise than routine clinic computed tomography (CT) images. Singular value decomposition based deconvolution algorithms are widely used for obtaining several functional perfusion maps. Recently block circulant singular value decomposition algorithms become popular for its superior property of immunity to contrast bolus lag. It is well known from literature that these algorithms are very sensitive to noise. There are a lot of examples of noise reduction filters in the literature as well as commercial ones. Functional maps which help physicians in the diagnostic process can be obtained with better image quality by de-noising CTP images with adaptive noise reduction filters. Objective: In this study, the effect of a noise adaptive wavelet filtering method on diagnostic performance on CTP stroke patient images is investigated. Method: Images of acute stroke patients were de-noised by this method and their diagnostic value were evaluated by visual means, peak signal-to-noise ratio and time intensity profile metrics. An observer evaluation study was carried out in order to validate quantitative image quality metrics. The results are compared with Gaussian and a bilateral filter based filtering method called TIPS (Time Intensity Profile Similarity) on same images sets to benchmark proposed method. Results: The diagnostic value of the images obtained from noise adaptive wavelet filtering method were better than Gaussian filter method and were compatible with a wellknown time intensity profile similarity bilateral filter method. Diagnostic performance of the both observers were improved compared to both Gaussian and TIPS methods. Conclusion: The noise adaptive wavelet filter method succeeded to reduce noise while preserving details contained in the contrast bolus. Its final effect on the timeintensity profiles and generated perfusion maps are compatible with the literature and showed improvements on diagnostic performance on specificity and overall accuracy when compared to other methods. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|